CN115641330A - Flexible circuit board defect detection method and system based on image processing - Google Patents

Flexible circuit board defect detection method and system based on image processing Download PDF

Info

Publication number
CN115641330A
CN115641330A CN202211440566.7A CN202211440566A CN115641330A CN 115641330 A CN115641330 A CN 115641330A CN 202211440566 A CN202211440566 A CN 202211440566A CN 115641330 A CN115641330 A CN 115641330A
Authority
CN
China
Prior art keywords
description
image
array
analysis space
defect
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202211440566.7A
Other languages
Chinese (zh)
Other versions
CN115641330B (en
Inventor
汪永新
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Dongguan Zonja Printing Co ltd
Original Assignee
Dongguan Zonja Printing Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Dongguan Zonja Printing Co ltd filed Critical Dongguan Zonja Printing Co ltd
Priority to CN202211440566.7A priority Critical patent/CN115641330B/en
Publication of CN115641330A publication Critical patent/CN115641330A/en
Application granted granted Critical
Publication of CN115641330B publication Critical patent/CN115641330B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Image Processing (AREA)
  • Image Analysis (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)

Abstract

According to the method and the system for detecting the defects of the flexible printed circuit board based on the image processing, the image sequence to be identified is separated to obtain a plurality of separation images, the first analysis space description array extraction and the second analysis space description array extraction are respectively carried out on the plurality of separation images, the description array fusion is carried out, the target fusion description array is obtained, the image definition description arrays corresponding to the plurality of separation images are obtained, based on the target fusion description array, the extracted image definition description arrays simultaneously contain information of two analysis spaces, the extracted image definition description arrays can restore basic properties of the images to the utmost extent, then the reliability coefficients of the defects of the flexible printed circuit board corresponding to the plurality of separation images are obtained, the accuracy of image defect detection is improved, then a plurality of defect image chains are determined in the image sequence, the corresponding defect definition description arrays are determined, the unified defect image chain combination is finally obtained, and the accuracy of reasoning and the accuracy of the unified defect image chain combination are improved.

Description

Flexible circuit board defect detection method and system based on image processing
Technical Field
The application relates to the field of image processing and artificial intelligence, in particular to a method and a system for detecting defects of a flexible printed circuit board based on image processing.
Background
The flexible circuit board has many advantages such as saving space, reducing weight and high flexibility, and thus has very wide application in production and life. Common defects of the flexible circuit board include solder joint short circuit or loss, component burn-out circuit short circuit, solder bridge, open circuit, solder joint and circuit board routing crack, and in conventional defect detection, a manual mode is adopted, obviously, in the face of large-batch defect circuit board detection, the efficiency and accuracy of manual detection cannot be guaranteed, and auxiliary defect detection through an artificial intelligent mode is a trend. In the face of a large-batch circuit board defect detection scene, such as factory return batch detection, the defects of the flexible circuit board need to be reasonably collected, so that unified processing and product yield analysis are facilitated, and therefore, the technical problem to be solved is how to quickly collect the defective products on the premise of ensuring efficient identification.
It should be noted that the above section of defect clustering is only for the background of the technology of the present application and is not intended to be a prerequisite or reference for evaluating the creativity of the present application.
Disclosure of Invention
The present invention provides a method and a system for detecting defects of a flexible printed circuit based on image processing, so as to solve the above technical problems.
In order to achieve the above purpose, the implementation process of the embodiment of the present application is as follows:
a first aspect of the embodiments of the present application provides a method for detecting defects of a flexible printed circuit board based on image processing, which is applied to a defect detection system, and the method includes:
acquiring an image sequence to be identified, and separating the image sequence to be identified to obtain a plurality of separated images; the image sequence to be identified comprises a plurality of images of the flexible circuit board to be detected;
respectively extracting a first analysis space description array from the multiple separated images to obtain a first analysis space description array corresponding to the multiple separated images, wherein the first analysis space description array comprises a first analysis space temporary description array and a first analysis space final description array;
respectively extracting second analysis space description arrays from the multiple separated images to obtain second analysis space description arrays corresponding to the multiple separated images, wherein the second analysis space description arrays comprise a second analysis space temporary description array and a second analysis space final description array;
performing description array fusion through a first analysis space temporary description array and a second analysis space temporary description array corresponding to the multiple separation images to obtain target fusion description arrays corresponding to the multiple separation images;
extracting image definition description arrays through a first analysis space final description array, a second analysis space final description array and a target fusion description array corresponding to the multiple separated images to obtain image definition description arrays corresponding to the multiple separated images, and performing image defect detection through the image definition description arrays to obtain defect credibility coefficients of the flexible circuit board corresponding to the multiple separated images;
determining a plurality of defect image chains from the image sequence to be identified through the credibility coefficients of the defects of the flexible printed circuit board, and determining defect paraphrase description arrays corresponding to the plurality of defect image chains through the image paraphrase description arrays;
and identifying the defects of the defect image chains through the defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain a unified defect image chain combination.
Optionally, the identifying the defect of the defect image chain through the defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain a unified defect image chain combination includes:
performing specified array extraction through the defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain a selected description array;
analyzing and restoring through the selected description array and the defect credibility coefficients of the flexible circuit board corresponding to the multiple separated images to obtain target defect paraphrase description arrays corresponding to the multiple defect image chains;
and identifying the defects of the plurality of defect image chains through the target defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain the unified defect image chain combination.
Optionally, the performing specified array extraction through the defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain a selected description array includes:
extracting original image description arrays corresponding to the multiple separated images respectively, and determining defect image chain original description arrays corresponding to the multiple defect image chains from the original image description arrays corresponding to the multiple separated images respectively;
integrating original description arrays of the defect image chains corresponding to the defect image chains with corresponding paraphrase description arrays of the defects respectively to obtain target integrated description arrays corresponding to the defect image chains;
and putting the target integration description arrays corresponding to the plurality of defect image chains into a description array extraction network of a description array processing model for processing to obtain a target selection description array.
Optionally, the performing defect identification on the plurality of defect image chains through the target defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain the unified defect image chain combination includes:
determining the matching degree among the plurality of defect image chains through target defect paraphrase description arrays corresponding to the plurality of defect image chains;
and performing image bucket division according to the matching degree among the plurality of defect image chains to obtain the unified defect image chain combination.
Optionally, the extracting the first analysis space description array from the multiple separated images respectively to obtain a first analysis space description array corresponding to the multiple separated images, where the first analysis space description array includes a first analysis space temporary description array and a first analysis space final description array, and includes:
respectively carrying out first analysis space convolution operation on the multiple separated images to obtain multiple temporary convolution description arrays and ending convolution description arrays corresponding to the multiple separated images;
performing dimension transformation operation on the plurality of temporary convolution description arrays to obtain a plurality of first analysis space temporary description arrays corresponding to the plurality of separation images;
carrying out dimension transformation operation on the ending convolution description array to obtain a first analysis space final description array corresponding to the multiple separation images;
the extracting of the second analysis space description array is respectively carried out on the multiple separated images to obtain the second analysis space description array corresponding to the multiple separated images, and the second analysis space description array comprises a second analysis space temporary description array and a second analysis space final description array, and the extracting of the second analysis space description array comprises the following steps:
extracting original image description arrays corresponding to the multiple separated images respectively;
and performing second analysis space convolution operation on the original image description arrays corresponding to the multiple separated images to obtain multiple second analysis space temporary description arrays and second analysis space final description arrays corresponding to the multiple separated images.
Optionally, the number of the first analysis space temporary description arrays is multiple, and the number of the second analysis space temporary description arrays is multiple;
the obtaining of the target fusion description array corresponding to the multiple separation images by performing description array fusion through the first analysis space temporary description array and the second analysis space temporary description array corresponding to the multiple separation images includes:
integrating a first analysis space temporary description array tensor1-1 in the plurality of first analysis space temporary description arrays with a corresponding second analysis space temporary description array tensor2-1 in the plurality of second analysis space temporary description arrays to obtain a first integrated description array, and performing convolution processing through the first integrated description array to obtain a first fusion description array;
integrating the first fusion description array, a first analysis space temporary description array tensor1-2 in the plurality of first analysis space temporary description arrays and a corresponding second analysis space temporary description array tensor2-2 in the plurality of second analysis space temporary description arrays to obtain a second integration description array, and performing convolution processing through the second integration description array to obtain a second fusion description array;
and when the plurality of first analysis space temporary description arrays and the plurality of second analysis space temporary description arrays are finished, obtaining a target fusion description array.
Optionally, the extracting an image definition description array through the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images to obtain the image definition description array corresponding to the multiple separated images, and performing image defect detection through the image definition description array to obtain the defect reliability coefficients of the flexible printed circuit corresponding to the multiple separated images includes:
integrating a first analysis space final description array, a second analysis space final description array and a target fusion description array corresponding to the multiple separation images to obtain a target integration description array corresponding to the multiple separation images;
performing convolution processing through the target integration description arrays corresponding to the multiple separated images to obtain convolution description arrays corresponding to the multiple separated images;
obtaining a description array maximum value and a description array average value corresponding to each order in the convolution description array through the convolution description arrays corresponding to the multiple separated images;
summing the maximum value of the description array and the average value of the description array to obtain paraphrase extraction description array values corresponding to all orders in the convolution description array, and obtaining the paraphrase extraction description array values corresponding to the multiple separated images through the paraphrase extraction description array values corresponding to all orders in the convolution description array;
performing linear transformation on the paraphrase extraction description arrays corresponding to the multiple separated images to obtain image paraphrase description arrays corresponding to the multiple separated images;
and detecting and judging the defects and non-defects of the flexible circuit board through the image definition description arrays corresponding to the multiple separated images to obtain the credibility coefficients of the defects of the flexible circuit board corresponding to the multiple separated images.
Optionally, the method further comprises:
putting the image sequence to be identified into an image defect detection model, and separating the image sequence to be identified through the image defect detection model to obtain a plurality of separated images;
respectively performing first analysis space description array extraction on the multiple separated images through the image defect detection model to obtain first analysis space description arrays corresponding to the multiple separated images, wherein the first analysis space description arrays comprise a first analysis space temporary description array and a first analysis space final description array;
respectively extracting second analysis space description arrays from the multiple separated images to obtain second analysis space description arrays corresponding to the multiple separated images, wherein the second analysis space description arrays comprise a second analysis space temporary description array and a second analysis space final description array;
performing description array fusion on a first analysis space temporary description array and a second analysis space temporary description array corresponding to the multiple separated images through the image defect detection model to obtain target fusion description arrays corresponding to the multiple separated images;
and extracting image definition description arrays from the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images through the image defect detection model to obtain the image definition description arrays corresponding to the multiple separated images, and detecting image defects through the image definition description arrays to obtain the defect reliability coefficients of the flexible printed circuit board corresponding to the multiple separated images.
Optionally, the image defect detection model includes a first analysis space description array extraction network, a second analysis space description array extraction network, a description array fusion network, an image definition description array extraction network, and a defect discrimination network;
the method further comprises the following steps:
putting the image sequence to be identified into an image defect detection model, and separating the image sequence to be identified through the image defect detection model to obtain a plurality of separated images;
putting the plurality of separated images into the first analysis space description array extraction network to extract a second analysis space description array to obtain a first analysis space temporary description array and a first analysis space final description array;
putting the plurality of separated images into a second analysis space description array extraction network to extract a second analysis space description array so as to obtain a second analysis space temporary description array and a second analysis space final description array;
putting a first analysis space temporary description array and a second analysis space temporary description array corresponding to a plurality of separation images into the description array fusion network for description array fusion to obtain target fusion description arrays corresponding to the plurality of separation images;
and putting the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images into the image paraphrase description array extraction network for image paraphrase description array extraction to obtain image paraphrase description arrays corresponding to the multiple separated images, and putting the image paraphrase description arrays into the defect judgment network for image defect detection to obtain the defect credibility coefficients of the flexible printed circuit corresponding to the multiple separated images.
A second aspect of the embodiments of the present application provides a defect detection system, which includes a processor and a memory, where the memory stores a computer program, and when the processor runs the program, the method described above is implemented.
According to the method and the system for detecting the defects of the flexible printed circuit board based on image processing, a plurality of separation images are obtained by separating an image sequence to be identified, then first analysis space description array extraction is carried out on the plurality of separation images respectively to obtain a first analysis space temporary description array and a first analysis space final description array, second analysis space description array extraction is carried out on the plurality of separation images respectively to obtain a second analysis space temporary description array and a second analysis space final description array, description array fusion is carried out on the first analysis space temporary description array and the second analysis space temporary description array corresponding to the plurality of separation images to obtain target fusion description arrays corresponding to the plurality of separation images, and the target fusion description arrays obtained through the description array fusion order contain nesting information of different analysis spaces. And then, extracting the image paraphrase description array through the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images to obtain the image paraphrase description array corresponding to the multiple separated images, so that the extracted image paraphrase description array simultaneously contains information of two analysis spaces, and the extracted image paraphrase description array can greatly restore the basic properties of the images. Then, image defect detection is carried out through the image paraphrase description array to obtain the defect credibility coefficients of the flexible circuit board corresponding to the multiple separated images, so that the accuracy of image defect detection can be improved, then, multiple defect image chains are determined from the image sequence to be identified through the defect credibility coefficients of the flexible circuit board, and the defect paraphrase description arrays corresponding to the multiple defect image chains are determined through the image paraphrase description array; and identifying the defects of the defect image chains through the defect paraphrase description arrays corresponding to the defect image chains to obtain a unified defect image chain combination, so that the accuracy of identifying the defects of the defect image chains is improved, and the accuracy of the obtained unified defect image chain combination is further ensured.
In the description that follows, additional features will be set forth, in part, in the description. These features will be in part apparent to those skilled in the art upon examination of the following and the accompanying drawings, or may be learned by production or use. The features of the present application may be realized and attained by practice or use of various aspects of the methodologies, instrumentalities and combinations particularly pointed out in the detailed examples that follow.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings used in the description of the embodiments of the present application will be briefly described below.
FIG. 1 is a flowchart of a method for detecting defects of a flexible printed circuit board based on image processing according to an embodiment of the present application.
Fig. 2 is a schematic view of a scene of a method for detecting defects of a flexible printed circuit based on image processing according to an embodiment of the present disclosure.
Fig. 3 is a schematic diagram of a functional module architecture of a defect detection apparatus according to an embodiment of the present disclosure.
Fig. 4 is a schematic composition diagram of a defect detection system according to an embodiment of the present application.
Detailed Description
The embodiments of the present application will be described below with reference to the drawings. The terminology used in the description of the embodiments of the examples herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.
In the embodiment of the present application, an implementation subject of the method for detecting defects of a flexible printed circuit board based on image processing is a defect detection system, including but not limited to a defect detection system with high performance data processing capability such as a server, a personal computer, and a notebook computer. For example, the defect detection system is a server, and specifically includes, but is not limited to, a single network server, a server group consisting of a plurality of network servers, or a cloud consisting of a large number of computers or network servers in cloud computing, where cloud computing is one of distributed computing and is a super virtual computer consisting of a group of loosely coupled computers. The computer equipment can run independently to realize the application, and can also be accessed to the network to realize the application through the interactive operation with other computer equipment in the network. The network in which the computer device is located includes, but is not limited to, the internet, a wide area network, a metropolitan area network, a local area network, a VPN network, etc. The defect detection system can communicate with the terminal equipment and receive the images shot by the terminal equipment, such as the images of the flexible circuit board to be detected.
The embodiment of the application provides a method for detecting defects of a flexible printed circuit board based on image processing, which is applied to a defect detection system, and as shown in fig. 1, the method comprises the following steps:
and 110, acquiring an image sequence to be identified, and separating the image sequence to be identified to obtain a plurality of separated images.
In the embodiment of the present application, the image sequence to be recognized includes a plurality of images of the flexible printed circuit board to be detected, which are captured images of a plurality of flexible printed circuit boards to be detected, and it should be specifically noted that, because of requirements on the size of the flexible printed circuit board and the definition of the images, the number of captured images is a plurality of images, for example, 5 images, in other words, the 5 images are captured images of the same flexible printed circuit board, each image corresponds to a portion of the flexible printed circuit board, and the images are continuous but may include images of repeated regions. In practical application, the images corresponding to the same flexible printed circuit board can be packed to serve as an image packet, and the image sequence to be recognized includes a plurality of image packets corresponding to the flexible printed circuit board, in other words, the image sequence to be recognized includes a plurality of images of the flexible printed circuit board to be detected. The embodiment of the application aims to identify images with defects in a plurality of flexible circuit boards to be detected and classify and collect the images according to the types of the defects. For example, please refer to fig. 2, which is a scene diagram of an image sequence to be recognized according to an embodiment of the present disclosure, and the scene diagram includes a plurality of image packets, such as A, B, C, D … …, where an image packet a includes images a1, a2, a3, a4, and a5, an image packet B includes images B1, B2, B3, B4, and B5, an image packet C includes images C1, C2, C3, C4, and C5, and an image packet D includes images D1, D2, D3, D4, and D5 … …, and a corresponding flexible printed circuit board of the image packet a has defects, specifically, among the plurality of images included in the image packet a, the defects indicated by the images a2, a3, and a4, that is, a defect types of a ', a' are type I defects, such as a tin ring defect; correspondingly, the flexible printed circuit board corresponding to the image packet B has defects, specifically, the defects indicated by the image B1, the image B2 and the image B3 in the plurality of images included in the flexible printed circuit board, that is, the defect type B 'and B' are type II defects, such as board surface defects; the flexible printed circuit board corresponding to the image packet C has defects, specifically, the defects indicated by the image C3, the image C4 and the image C5 in the plurality of images included in the flexible printed circuit board, that is, the defect type of C 'and C' is a type I defect; the flexible printed circuit board corresponding to the image packet D has defects, specifically, among the plurality of images included in the flexible printed circuit board, the defects indicated by the image D2, the image D3, the image D4 and the image D5, that is, the defect type of D ' is a type II defect … …, because the defect types of a ' and C ' are consistent, the two are collected into the same set, and the defect types of B ' and D ' are consistent, and the two are collected into the same set. In the following description, how to perform the above steps is described, it can be understood that, in step 110, the image sequence to be recognized is separated to obtain a plurality of separated images, and the separated images are a1, a2, a3, a4, a5, b1, b2, b3, b4, b5, c1, c2, c3, c4, c5, d1, d2, d3, d4, d5 … …, in other words, the minimum unit obtained by separation is the minimum component unit of the image packet corresponding to the flexible printed circuit board to be detected.
And 120, extracting the first analysis space description array from the plurality of separated images respectively to obtain a first analysis space description array corresponding to the plurality of separated images, wherein the first analysis space description array comprises a first analysis space temporary description array and a first analysis space final description array.
The first analysis space description array is an image description array extracted from the first analysis space, in this embodiment, the first analysis space characterization is analyzed in an image space domain or an image frequency domain, in this embodiment, the first analysis space further includes a second analysis space, and the second analysis space is analyzed in the image space domain or the image frequency domain, it is understood that when the first analysis space is analyzed in the image space domain, the second analysis space is analyzed in the image frequency domain, and conversely, when the first analysis space is analyzed in the image frequency domain, the second analysis space is analyzed in the image space domain, and the first and second analysis spaces are only used for distinguishing the image description array. The first analysis space temporary description array is a paraphrase array obtained when the first analysis space final description array is extracted, the first analysis space final description array is a first analysis space description array corresponding to the separated image obtained by the last extraction, and the paraphrase array is a characteristic numerical value expression for representing the semantics of the image. The first analysis space description array may be regarded as a feature of an image obtained in the first analysis space.
For example, the convolution processing may be repeated on the separated image, a first analysis space temporary description array is obtained each time the processing is performed, the current first analysis space temporary description array is used as the loading information of the next processing, and the array obtained by the final processing is used as the first analysis space final description array after the loop is finished. It is easy to understand that the above-mentioned processing is performed once for all the separated images, and a first analysis space temporary description array and a first analysis space final description array corresponding to the plurality of separated images are obtained.
And step 130, extracting second analysis space description arrays from the multiple separated images respectively to obtain second analysis space description arrays corresponding to the multiple separated images, wherein the second analysis space description arrays comprise a second analysis space temporary description array and a second analysis space final description array.
And 140, performing description array fusion through the first analysis space temporary description array and the second analysis space temporary description array corresponding to the multiple separated images to obtain a target fusion description array corresponding to the multiple separated images.
The process of describing array fusion is to carry out the negotiation of image content on the first analysis space temporary description array and the corresponding second analysis space temporary description array, so that the stability of defect identification is improved, and a deeper paraphrase array can be obtained. The target fusion description array is a paraphrase array obtained after the first analysis space description array and the second analysis space description array are fused. For example, the first analysis space temporary description array and the second analysis space temporary description array corresponding to the separated images are combined to obtain a target fusion description array corresponding to the separated images, and the first analysis space temporary description array and the second analysis space temporary description array corresponding to each separated image are combined to obtain a target fusion description array corresponding to each separated image.
And 150, extracting the image paraphrase description array through the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images to obtain the image paraphrase description array corresponding to the multiple separated images, and detecting image defects through the image paraphrase description array to obtain the defect reliability coefficients of the flexible printed circuit corresponding to the multiple separated images.
The image definition description array is obtained by integrating the first analysis space description array, the second analysis space description array and the fusion description array, and each separated image comprises a corresponding image definition description array. The image defect detection is a process of judging images, specifically judging whether the images have defects of the flexible circuit board, and the detection result comprises the defects of the flexible circuit board and the defects of the flexible circuit board. The confidence coefficient of the defects of the flexible circuit board is used for representing the confidence level that the corresponding separated images are the defects of the flexible circuit board, and the higher the confidence coefficient of the defects of the flexible circuit board is, the higher the confidence level that the corresponding separated images are the defects of the flexible circuit board is. For example, the image paraphrase description array is aggregated and obtained through the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to each separated image, and the aggregated description array is obtained, so that the image paraphrase description array corresponding to each separated image is obtained. And then, carrying out defect identification through the image definition description array, determining whether the separated image is a flexible circuit board defect or a non-flexible circuit board defect, and obtaining the reliability coefficient of the flexible circuit board defect corresponding to each separated image.
And step 160, determining a plurality of defect image chains from the image sequence to be identified through the defect confidence coefficients of the flexible printed circuit board, and determining defect paraphrase description arrays corresponding to the plurality of defect image chains through the image paraphrase description arrays.
The defect image chain is an image combination obtained by combining a plurality of continuous separated images with defects of the flexible printed circuit board, specifically, refer to a ', B', C ', and D' in fig. 2, where the separated images with defects of the flexible printed circuit board are separated images with a reliability coefficient of defects of the flexible printed circuit board that is greater than a predetermined reliability coefficient of defects of the flexible printed circuit board, and the predetermined reliability coefficient of defects of the flexible printed circuit board is a predetermined numerical value, and the size of the predetermined reliability coefficient of defects of the flexible printed circuit board can be selected according to actual conditions. The defect paraphrase description array can indicate paraphrase of a defect image chain, and is obtained by fusing image paraphrase description arrays corresponding to defects of all flexible printed circuits, for example, a reliable coefficient of defects of the flexible printed circuits corresponding to all separated images is compared with a reliable coefficient of defects of preset flexible printed circuits, and if the reliable coefficient of defects of the flexible printed circuits is larger than the reliable coefficient of defects of the preset flexible printed circuits, the separated images corresponding to the reliable coefficients of defects of the flexible printed circuits have defects of the flexible printed circuits. And then combining the separated images corresponding to the defects of the flexible circuit board belonging to one image package in the image sequence to be identified into a defect image chain through the spatial ordering of the images to obtain a plurality of defect image chains, combining the image paraphrase description arrays corresponding to the defects of the flexible circuit board in the defect image chain to obtain a defect paraphrase description array corresponding to the defect image chain, and processing each defect image chain to obtain a defect paraphrase description array corresponding to each defect image chain.
And 170, identifying the defects of the defect image chains through the defect paraphrase description arrays corresponding to the defect image chains to obtain a uniform defect image chain combination.
The defect image chain defect identification is used to determine whether the defect image chains are the same defect image chain, the unified defect image chain combination includes each of the same defect image chains (such as the type I defect and the type II defect in fig. 2), and the same defect image chain is a defect image chain whose matching degree is greater than a preset matching degree.
The plurality of defect image chains can be grouped by the defect paraphrase description arrays corresponding to the plurality of defect image chains, for example, a preset grouping operator such as a mean value algorithm is adopted, and a grouping result is obtained by distance clustering of the description arrays, so that at least one unified defect image chain combination is obtained.
The method for detecting the defects of the flexible printed circuit board based on image processing includes the steps of separating an image sequence to be identified to obtain a plurality of separation images, respectively extracting a first analysis space description array from the plurality of separation images to obtain a first analysis space temporary description array and a first analysis space final description array, respectively extracting a second analysis space description array from the plurality of separation images to obtain a second analysis space temporary description array and a second analysis space final description array, performing description array fusion on the first analysis space temporary description array and the second analysis space temporary description array corresponding to the plurality of separation images to obtain target fusion description arrays corresponding to the plurality of separation images, and enabling the target fusion description arrays obtained through the description array fusion to contain nesting information of different analysis spaces. And then, extracting the image paraphrase description array through the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images to obtain the image paraphrase description array corresponding to the multiple separated images, so that the extracted image paraphrase description array simultaneously contains the information of the two analysis spaces, and simultaneously the extracted image paraphrase description array can restore the basic properties of the images to the utmost extent. Then, image defect detection is carried out through the image paraphrase description array to obtain the defect credibility coefficients of the flexible circuit board corresponding to the multiple separated images, so that the accuracy of image defect detection can be improved, then, multiple defect image chains are determined from the image sequence to be identified through the defect credibility coefficients of the flexible circuit board, and the defect paraphrase description arrays corresponding to the multiple defect image chains are determined through the image paraphrase description array; and identifying the defects of the defect image chains through the defect paraphrase description arrays corresponding to the defect image chains to obtain a unified defect image chain combination, so that the accuracy of identifying the defects of the defect image chains is improved, and the accuracy of the obtained unified defect image chain combination is further ensured.
In an implementation of the embodiment of the present application, for step 170, identifying the defect in the defect image chain by using the defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain a unified defect image chain combination includes:
step 171, performing specified array extraction through the defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain a selected description array.
Specifically, the embodiment of the present application provides a description array processing model, which may be built by using a transform architecture, where the specified array extraction is a process of describing that an Encoder in the array processing model adopts encoding, which is substantially a process of extracting a specific feature (description array), and selecting a description array is a result of encoding a defect paraphrase description array that completes the specified array extraction. For the training of describing the array processing model, a general training process can be adopted, namely training data (sample-sample label) is prepared, a preset model is input, an inference result is obtained, the cost of the sample label is compared, model parameters are updated, iteration is carried out until convergence, and in addition, the describing array processing model can be used by directly selecting an existing adaptive model. the transformer is a mature model, and specific structures can refer to the prior art, which is not described herein.
And 172, analyzing and restoring the defect credibility coefficients of the flexible printed circuit board corresponding to the selected description array and the multiple separated images to obtain target defect paraphrase description arrays corresponding to the multiple defect image chains.
And continuing to step 171, decoding the result output by the Encoder in the analytic reduction processing process, i.e. the Decoder in the description array processing model, for example, identifying the defect confidence coefficient of the flexible printed circuit board of the separated image corresponding to the current defect image chain from the defect confidence coefficients of the flexible printed circuit board corresponding to the multiple separated images, then inputting the selected description array corresponding to the defect image chain and the defect confidence coefficient of the flexible printed circuit board of the separated image corresponding to the current defect image chain into the Decoder of the description array processing model to obtain the target defect paraphrase description array corresponding to the defect image chain, and performing the above processes on all the defect image chains to obtain the target defect paraphrase description array corresponding to all the defect image chains.
Step 173, the defect recognition is performed on the plurality of defect image chains through the target defect paraphrase description arrays corresponding to the plurality of defect image chains, so as to obtain a unified defect image chain combination.
For example, the target defect paraphrase description arrays corresponding to the plurality of defect image chains are grouped by a mean algorithm to obtain a plurality of defect image chains, and the defect image chains of each category are determined to be the same defect image chain to obtain a unified defect image chain combination of a unified category, such as the type I defect and the type II defect in fig. 2.
In an implementation of the embodiment of the present application, for step 171, performing specified array extraction on the defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain a selected description array includes: extracting original image description arrays corresponding to the multiple separated images respectively, and determining defect image chain original description arrays corresponding to the multiple defect image chains from the original image description arrays corresponding to the multiple separated images respectively; integrating original description arrays of the defect image chains corresponding to the defect image chains with corresponding paraphrase description arrays of the defects respectively to obtain target integrated description arrays corresponding to the defect image chains; and putting the target integration description array corresponding to the plurality of defect image chains into a description array extraction network of a description array processing model for processing to obtain a target selection description array.
The original image description array is the most basic description information of the image, the defect image chain original description array is the original image description array corresponding to the defect image chain, and is obtained by integrating the original image description arrays of a plurality of separated images corresponding to the defect image chain, the target integration description array is the description array combined with the original characteristic information, and the target selection description array is the selected description array combined with the original characteristic information. For example, extracting the original image description arrays corresponding to the multiple separated images, then integrating the original image description arrays of the separated images corresponding to the defective image chains to obtain the original description array of the defective image chain corresponding to each defective image chain, for example, connecting (for example, stitching) the original description arrays of the separated images corresponding to each defective image chain, then connecting the original description array of the defective image chain corresponding to each defective image chain with the defect paraphrase description array corresponding to each defective image chain to obtain the target integration description arrays corresponding to the multiple defective image chains, and then sequentially putting the target integration description arrays corresponding to each defective image chain into the description array extraction network of the description array processing model to process to obtain the target selection description array. Through the process, the original description array of the defect image chain is integrated with the corresponding defect paraphrase description array and then processed, so that the accuracy of the obtained target selected description array is improved, and the accuracy of the obtained target defect paraphrase description array is improved.
In one implementation of the examples of the present application, for step 173: the defect recognition is carried out on the plurality of defect image chains through the target defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain a unified defect image chain combination, and the method comprises the following steps: determining the matching degree among the plurality of defect image chains through the target defect paraphrase description arrays corresponding to the plurality of defect image chains; and performing image bucket division through the matching degree among the plurality of defect image chains to obtain a unified defect image chain combination.
The matching degree between the defect image chains represents the closeness degree between the defect image chains, and the matching degree can be obtained by calculating the vector distance (such as Euclidean distance, manhattan distance and cosine distance) of the defect image chains, and the distance and the matching degree are in negative correlation. For example, through the target defect paraphrase description arrays corresponding to each defect image chain, the target defect paraphrase description array A and the target defect paraphrase description array B are determined in the target defect paraphrase description arrays corresponding to the defect image chains, the matching degree of the target defect paraphrase description array A and the target defect paraphrase description array B is determined, the matching degree among all the target defect paraphrase description arrays is determined, all the matching degrees are classified in a bucket, the defect image chains corresponding to the target defect paraphrase description arrays with the matching degree larger than the preset matching degree are integrated into a uniform defect image chain combination, and the image bucket is performed in a mode of determining the matching degree, so that the process of pre-specifying the number of centroids during bucket division is omitted, and the accuracy and the obtaining speed of the obtained uniform defect image chain combination are improved conveniently.
In one implementation of the embodiments of the present application, for step 120: carry out the extraction of first analysis space description array respectively to many separation images, obtain the first analysis space description array that many separation images correspond, first analysis space description array includes the interim description array of first analysis space and the final description array of first analysis space, includes: respectively carrying out first analysis space convolution operation on a plurality of separated images to obtain a plurality of temporary convolution description arrays and ending convolution description arrays corresponding to the plurality of separated images; performing dimension transformation operation on the plurality of temporary convolution description arrays to obtain a plurality of first analysis space temporary description arrays corresponding to the plurality of separation images; and carrying out dimension transformation operation on the ending convolution description array to obtain a first analysis space final description array corresponding to the multiple separated images.
The first analysis space convolution operation can obtain a description array of the image in the first analysis space, the ending convolution description array is a finally obtained convolution description array, the temporary convolution description array is a convolution description array obtained before the ending convolution description array is obtained, and the dimension transformation operation is an operation of transforming the first analysis space description array into the same dimension as the second analysis space description array. For example, specifically, a first analysis space convolution operation is performed on each separated image to obtain a plurality of temporary convolution description arrays and a final ending convolution description array corresponding to each separated image, then dimension transformation operation is performed on each temporary convolution description array to obtain a plurality of first analysis space temporary description arrays corresponding to the plurality of separated images, and dimension transformation operation is performed on the ending convolution description array to obtain a first analysis space final description array corresponding to the plurality of separated images.
In one implementation of the embodiment of the present application, for step 130: the extraction of a second analysis space description array is respectively adopted to a plurality of separation images, a second analysis space description array corresponding to the plurality of separation images is obtained, and the second analysis space description array comprises a second analysis space temporary description array and a second analysis space final description array, and the extraction comprises the following steps: extracting original image description arrays corresponding to the multiple separated images respectively; and performing second analysis space convolution operation on the original image description arrays corresponding to the multiple separated images to obtain multiple second analysis space temporary description arrays and multiple second analysis space final description arrays corresponding to the multiple separated images, wherein the second analysis space convolution operation can acquire the description arrays of the images in the second analysis space, extracting the description arrays to the original image description arrays corresponding to the multiple separated images, performing multiple times of second analysis space convolution operation (the times of the second analysis space convolution operation are equal to those of the first analysis space convolution operation) on the original image description arrays, finally obtaining the second analysis space final description arrays, obtaining the second analysis space temporary description arrays before obtaining the second analysis space final description arrays, and finally obtaining the multiple second analysis space temporary description arrays and the second analysis space final description arrays corresponding to the multiple separated images. Extracting original image description arrays corresponding to the multiple separated images respectively; and then, a second analysis space convolution operation is carried out through the original image description array to obtain a plurality of second analysis space temporary description arrays and a second analysis space final description array corresponding to the plurality of separation images, so that the accuracy of the obtained second analysis space description array is improved.
In an implementation of the embodiment of the present application, the first analysis space temporary description array includes a plurality of temporary description arrays, and the second analysis space temporary description array includes a plurality of temporary description arrays. With respect to step 140: the temporary description array of the first analysis space and the temporary description array of the second analysis space corresponding to the multiple separated images are used for describing array fusion, and the target fusion description array corresponding to the multiple separated images is obtained, and the method comprises the following steps: step 141, integrating the first analysis space temporary description array tensor1-1 in the plurality of first analysis space temporary description arrays and the corresponding second analysis space temporary description array tensor2-1 in the plurality of second analysis space temporary description arrays to obtain a first integrated description array, and performing convolution processing through the first integrated description array to obtain a first fusion description array. The integrated description array is obtained by adding the description arrays, and the fused description array is obtained by fusing the description arrays. For example, a first analysis space temporary description array tensor1-1 and a corresponding second analysis space temporary description array tensor2-1 are obtained, the first analysis space temporary description array tensor1-1 and the corresponding second analysis space temporary description array tensor2-1 are obtained according to a first convolution calculation, then the first analysis space temporary description array tensor1-1 and the corresponding second analysis space temporary description array tensor2-1 are subjected to a splicing process at a set level (channel, dimension) to obtain a first integrated description array, and then the first integrated description array is subjected to a convolution process to obtain a first fusion description array. Step 142, integrating the first fusion description array, the first analysis space temporary description array tensor1-2 in the plurality of first analysis space temporary description arrays and the corresponding second analysis space temporary description array tensor2-2 in the plurality of second analysis space temporary description arrays to obtain a second integration description array, and performing convolution processing through the second integration description array to obtain a second fusion description array. In the step, in the process of fusing the first analysis space temporary description array and the second analysis space temporary description array for the next time, the first fusion description array obtained in the previous time is integrated together to obtain a second integration description array, and then convolution processing is performed on the second integration description array to obtain a second fusion description array. And step 143, obtaining the target fusion description array when the plurality of first analysis space temporary description arrays and the plurality of second analysis space temporary description arrays are completed. For example, description array fusion is performed on each first analysis space temporary description array and the corresponding second analysis space temporary description array in sequence to obtain a last fusion description array, the last fusion description array is integrated with the current first analysis space temporary description array and the current second analysis space temporary description array, then convolution processing is performed on the integration description array to obtain a current fusion description array, when description array fusion is performed finally, the fusion description array, the final first analysis space temporary description array and the final second analysis space temporary description array are integrated to obtain a final integration description array, and then convolution processing is performed on the final integration description array to obtain a target integration description array. According to the method and the device, the description array fusion is carried out on the first analysis space temporary description array and the corresponding second analysis space temporary description array, the description arrays of different analysis spaces are mutually nested and complemented, and in addition, the top-level network is made to be clear to information contained in the network below, so that the acquired target integration description array is made to be more accurate.
In one implementation of the embodiments of the present application, step 150: extracting the image paraphrasing description array through the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images to obtain the image paraphrasing description array corresponding to the multiple separated images, and performing image defect detection through the image paraphrasing description array to obtain the defect reliability coefficients of the flexible printed circuit corresponding to the multiple separated images, which may include:
and 151, integrating the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images to obtain a target integration description array corresponding to the multiple separated images.
And 152, performing convolution processing on the target integration description arrays corresponding to the multiple separated images to obtain convolution description arrays corresponding to the multiple separated images.
The target integration description array is a description array obtained by integrating the first analysis space final description array, the second analysis space final description array and the target fusion description array, and the convolution description array is obtained by performing convolution processing on the target integration description array. For example, the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to each separated image are sequentially connected on a channel to obtain a target integration description array corresponding to each separated image, and then the target integration description array corresponding to each separated image is subjected to convolution processing to obtain convolution description arrays corresponding to a plurality of separated images.
Step 153, determining the maximum value of the description array and the average value of the description array corresponding to each order in the convolution description array through the convolution description arrays corresponding to the multiple separated images.
Step 154, summing the maximum value of the description array and the average value of the description array to obtain the paraphrase extraction description array value corresponding to each order in the convolution description array, and obtaining the paraphrase extraction description array corresponding to the multiple separated images through the paraphrase extraction description array value corresponding to each order in the convolution description array.
The maximum value of the description array is the maximum description array in all the description arrays corresponding to the corresponding order (which can be understood as the dimension of the description array), the average value of the description arrays is the average result of all the description arrays corresponding to the corresponding order, and the value of the paraphrase extraction description array is the description array which is obtained by extraction and represents the paraphrase description array of the image. For example, the paraphrase extraction description array corresponding to each separated image is determined in sequence, the convolution description array corresponding to the separated image to be acquired is determined, and then the description array maximum value and the description array average value corresponding to each order in the convolution description array are acquired, in other words, the description array average value and the description array maximum value of all the description arrays corresponding to each order are determined. And then summing the maximum value of the description array and the equal value of the description array to obtain the paraphrase extraction description array value corresponding to each order in the convolution description array, and taking the paraphrase extraction description array value corresponding to each order as the paraphrase extraction description array of the current separation image.
Step 155, performing linear transformation on the paraphrase extraction description arrays corresponding to the multiple separated images to obtain image paraphrase description arrays corresponding to the multiple separated images.
And 156, detecting and judging the defects and non-defects of the flexible printed circuit board through the image definition description arrays corresponding to the multiple separated images to obtain the credibility coefficients of the defects of the flexible printed circuit board corresponding to the multiple separated images.
For example, the paraphrase extraction description arrays corresponding to the separated images are subjected to linear transformation through a Sigmoid function in sequence to obtain image paraphrase description arrays corresponding to the separated images, and then a classifier (normalized) is selected through the image paraphrase description arrays to perform detection and judgment on defects and non-defects of the flexible printed circuit board, so that the credibility coefficients of the defects of the flexible printed circuit board corresponding to the separated images are obtained. The description array maximum value and the description array equivalent value are determined, the paraphrasing extraction description array is obtained through the description array maximum value and the description array equivalent value, the description array maximum value can reflect the most prominent description information, the description array equivalent value can reflect the description information of the whole image, the extracted image paraphrasing description array is helped to contain more excellent accuracy, defect identification is carried out through the image paraphrasing description array, and the accuracy of the obtained defect credibility coefficient of the flexible circuit board is improved conveniently.
In an implementation of the embodiment of the present application, the method for detecting defects of a flexible printed circuit based on image processing further includes:
and step 10, putting the image sequence to be identified into an image defect detection model, and separating the image sequence to be identified through the image defect detection model to obtain a plurality of separated images.
Step 20, performing first analysis space description array extraction on the multiple separated images through the image defect detection model to obtain first analysis space description arrays corresponding to the multiple separated images, wherein the first analysis space description arrays comprise a first analysis space temporary description array and a first analysis space final description array; and respectively extracting the second analysis space description arrays of the multiple separated images to obtain second analysis space description arrays corresponding to the multiple separated images, wherein the second analysis space description arrays comprise a second analysis space temporary description array and a second analysis space final description array.
And step 30, performing description array fusion on the first analysis space temporary description array and the second analysis space temporary description array corresponding to the multiple separated images through the image defect detection model to obtain target fusion description arrays corresponding to the multiple separated images.
And step 40, extracting the image paraphrase description array from the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images through the image defect detection model to obtain the image paraphrase description array corresponding to the multiple separated images, and detecting the image defects through the image paraphrase description array to obtain the defect reliability coefficients of the flexible circuit board corresponding to the multiple separated images.
In this embodiment, the image defect detection model is configured to identify defects and non-defects in the image sequence, and the image defect detection model may be established by selecting a machine learning or deep learning network, such as a convolutional neural network, without limitation. In practical application, firstly obtaining an image sequence to be recognized, putting the image sequence to be recognized into an image defect detection model, wherein the image defect detection model can comprise two sub-networks, extracting a second analysis space final description array and a first analysis space final description array corresponding to the image sequence to be recognized based on two sub-modules, fusing the description arrays of the second analysis space temporary description array and the first analysis space temporary description array obtained by extraction to obtain a target fusion description array, extracting a definition array through the obtained second analysis space final description array, the obtained first analysis space final description array and the target fusion description array, and detecting image defects through the extracted definition array. By adopting the process, the image defect detection is executed through the image defect detection model, the credibility coefficients of the defects of the flexible circuit board corresponding to the multiple separated images are obtained, and the process of the image defect detection is accelerated.
In an implementation scheme of the embodiment of the application, the image defect detection model includes a first analysis space description array extraction network, a second analysis space description array extraction network, a description array fusion network, an image paraphrasing description array extraction network and a defect discrimination network; the method for detecting the defects of the flexible circuit board based on the image processing can further comprise the following steps:
step 101, putting an image sequence to be identified into an image defect detection model, and separating the image sequence to be identified through the image defect detection model to obtain a plurality of separated images.
And 102, putting the plurality of separated images into a first analysis space description array extraction network to perform second analysis space description array extraction, so as to obtain a first analysis space temporary description array and a first analysis space final description array.
And 103, putting the plurality of separated images into a second analysis space description array extraction network to perform second analysis space description array extraction, so as to obtain a second analysis space temporary description array and a second analysis space final description array.
And 104, putting the first analysis space temporary description array and the second analysis space temporary description array corresponding to the multiple separated images into a description array fusion network for description array fusion to obtain target fusion description arrays corresponding to the multiple separated images.
And 105, putting the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images into an image paraphrase description array extraction network for image paraphrase description array extraction to obtain image paraphrase description arrays corresponding to the multiple separated images, and putting the image paraphrase description arrays into a defect judgment network for image defect detection to obtain the defect credibility coefficients of the flexible printed circuit corresponding to the multiple separated images.
The first analysis space description array extraction network is configured to extract a first analysis space description array of the image, the second analysis space description array extraction network is configured to extract a second analysis space description array of the image, the description array fusion network is configured to perform description array fusion on the second analysis space temporary description array and the first analysis space temporary description array, the image definition description array extraction network is configured to extract a definition array of the image, and the defect judgment network is configured to judge defects and non-defects.
For example, a plurality of separated images are put into a first analysis space description array extraction network to perform second analysis space description array extraction, that is, a convolution module of the first analysis space description array extraction network generates a first analysis space description array, a first analysis space final description array is obtained based on a last convolution module, a previous convolution module generates a first analysis space temporary description array, a plurality of separated images are put into a second analysis space description array extraction network to perform second analysis space description array extraction, that is, a convolution module of the second analysis space description array extraction network generates a second analysis space description array, a last convolution module generates a second analysis space final description array, a previous convolution module obtains the second analysis space temporary description array, and the convolution times of the first analysis space description array extraction network and the second analysis space description array extraction network are equal. And performing description array fusion on the first analysis space temporary description array and the second analysis space temporary description array through a description array fusion network to obtain a target fusion description array, then performing image definition description array extraction based on an image definition description array extraction network, and performing image defect detection through a defect judgment network to obtain defect credibility coefficients of the flexible printed circuit corresponding to a plurality of separated images.
In an implementation of the embodiment of the present application, the image defect detection model may include a shared network and two subnetworks respectively corresponding to different analysis spaces, where the image defect detection model puts the image sequence to be identified into the two subnetworks respectively, and obtains the first analysis space final description array and the second analysis space final description array including the same dimension and channel by performing a plurality of convolution and pooling operations.
In an implementation of the embodiment of the present application, the image defect detection model is optimally trained by the following steps:
step 100, obtaining an image sequence sample and a matched indicator.
The image sequence sample is an image sequence required by model training, and the indicator is a real record corresponding to the image sequence sample, such as whether the image sequence sample is defective or non-defective.
And 200, putting the image sequence samples into a quasi-training image defect detection model, and separating the image sequence samples by using the quasi-training image defect detection model to obtain a plurality of separated image samples.
Step 300, performing first analysis space description array extraction on each separated image sample by using a to-be-trained image defect detection model to obtain an original first analysis space description array corresponding to a plurality of separated image samples, wherein the original first analysis space description array comprises an original first analysis space temporary description array and an original first analysis space final description array; and respectively extracting the second analysis space description array of each separated image sample to obtain an original second analysis space description array corresponding to the plurality of separated image samples, wherein the original second analysis space description array comprises an original second analysis space temporary description array and an original second analysis space final description array.
And 400, performing description array fusion on the original first analysis space temporary description array and the original second analysis space temporary description array corresponding to each separated image sample by using a quasi-training image defect detection model to obtain original fusion description arrays corresponding to a plurality of separated image samples.
Step 500, performing image paraphrase description array extraction on the original first analysis space final description array, the original second analysis space final description array and the original fusion description array corresponding to each separated image sample by using a quasi-training image defect detection model to obtain original image paraphrase description arrays corresponding to a plurality of separated image samples, and performing image defect detection through the original image paraphrase description arrays to obtain original flexible printed circuit board defect confidence coefficients corresponding to the plurality of separated image samples.
The method comprises the steps of obtaining a separation image sample through separation in a training process, obtaining an original first analysis space description array through parameter extraction to be optimized, obtaining an original second analysis space description array through parameter extraction to be optimized, obtaining a defect credibility coefficient of the original flexible printed circuit board through parameter inference to be optimized, building a quasi-training image defect detection model based on a neural network, carrying out first image defect detection inference on an image sequence sample through the quasi-training image defect detection model, obtaining the defect credibility coefficient of the original flexible printed circuit board corresponding to each separation image sample, and keeping the image defect detection inference of the quasi-training image defect detection model and the inference of the trained image defect detection model the same.
Step 600, obtaining training costs through the original flexible printed circuit board defect confidence coefficients corresponding to the separated image samples and the indication marks matched with the image sequence samples to obtain cost values, and debugging the image defect detection model to be trained through the cost values to obtain a debugging image defect detection model.
And 700, determining the debugging image defect detection model as a to-be-trained image defect detection model, and repeatedly training until the model converges to finally obtain the image defect detection model.
The image sequence sample and the matched indication mark are used for training the image defect detection model to be trained to obtain the image defect detection model, and the image defect detection model is independently deployed, so that the optimization result is accurate and low in cost after the optimization is completed, and the inference precision of the image defect detection model is improved.
Specifically, the embodiment of the application deploys a quasi-training image sequence processing network, optimizes the quasi-training image sequence processing network through training samples to obtain an image sequence processing network, separates an image sequence to be identified based on the image sequence processing network to obtain a plurality of separated images, performs first analysis spatial description array extraction on each of the plurality of separated images to obtain a first analysis spatial description array corresponding to the plurality of separated images, the first analysis spatial description array comprises a first analysis spatial temporary description array and a first analysis spatial final description array, performs second analysis spatial description array extraction on each of the plurality of separated images to obtain a second analysis spatial description array corresponding to the plurality of separated images, the second analysis spatial description array comprises a second analysis spatial temporary description array and a second analysis spatial final description array, performs description array fusion through the first analysis spatial temporary description array and the second analysis spatial temporary description array corresponding to the plurality of separated images to obtain a target fusion description array corresponding to the plurality of separated images, performs defect image identification on a plurality of defect chains corresponding to the plurality of separated images through a soft-analysis spatial description array, determines defect chain corresponding to the defect chain of the plurality of separated images by the image chain, determines a plurality of defect chain corresponding to be identified through the defect chain, and determines a plurality of the defect chain through the image chain, a unified defect image chain combination is obtained.
In addition, in an implementation of the embodiment of the present application, the method for detecting defects of a flexible printed circuit board based on image processing is performed by a defect detection system, and the method includes the following steps:
step I, obtaining an image sequence to be identified, putting the image sequence to be identified into an image defect detection model, and separating the image sequence to be identified through the image defect detection model to obtain a plurality of separated images, wherein the image defect detection model comprises a first analysis space description array extraction network, a second analysis space description array extraction network, a description array fusion network, an image definition description array extraction network and a defect judgment network.
And step II, putting the multiple separated images into a first analysis space description array extraction network to perform first analysis space convolution operation to obtain a temporary convolution description array and a final convolution description array corresponding to the multiple separated images, and performing dimension transformation operation on the temporary convolution description array and the final convolution description array to obtain the first analysis space temporary description array and the first analysis space description array corresponding to the multiple separated images.
And III, extracting the original image description arrays corresponding to the multiple separated images, and putting the original image description arrays corresponding to the multiple separated images into a second analysis space description array extraction network to perform second analysis space convolution operation, so as to obtain a second analysis space temporary description array and a second analysis space final description array corresponding to the multiple separated images. And integrating the first analysis space temporary description array and the second analysis space temporary description array to obtain a first integration description array, and performing convolution processing through the first integration description array to obtain a target fusion description array.
And IV, putting the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images into an image definition description array extraction network for integration to obtain target integration description arrays corresponding to the multiple separated images, performing convolution processing on the target integration description arrays corresponding to the multiple separated images to obtain convolution description arrays corresponding to the multiple separated images, determining description array maximum values and description array average values corresponding to all the orders in the convolution description arrays through the convolution description arrays corresponding to the multiple separated images, summing the description array maximum values and the description array average values to obtain definition extraction description array values corresponding to all the orders in the convolution description arrays, and extracting the description array values corresponding to all the orders in the convolution description arrays to obtain definition extraction description arrays corresponding to the multiple separated images.
And V, putting the image definition description array into a defect judgment network to detect and judge the defects and the non-defects of the flexible printed circuit board, and obtaining the credibility coefficients of the defects of the flexible printed circuit board corresponding to the multiple separated images. Determining a plurality of defect image chains from the image sequence to be identified through the defect credibility coefficients of the flexible circuit board corresponding to the plurality of separated images, and determining a defect paraphrase description array corresponding to the plurality of defect image chains through the image paraphrase description array.
And VI, putting the defect paraphrase description arrays corresponding to the plurality of defect image chains into a description array extraction network of a description array processing model to perform specified array extraction to obtain selected description arrays corresponding to the plurality of defect image chains, and putting the selected description arrays corresponding to the plurality of defect image chains and the corresponding flexible circuit board defect credibility coefficients into an analysis reduction network of the description array processing model to perform analysis reduction to obtain target defect paraphrase description arrays corresponding to the plurality of defect image chains.
And step VII, determining the matching degree among the plurality of defect image chains through the target defect paraphrase description arrays corresponding to the plurality of defect image chains, and performing image barreling according to the matching degree among the plurality of defect image chains to obtain a unified defect image chain combination.
To sum up, in the method for detecting defects of a flexible printed circuit board based on image processing provided by the embodiment of the present application, a plurality of separation images are obtained by separating an image sequence to be identified, then the first analysis space description array is extracted from the plurality of separation images, the first analysis space temporary description array and the first analysis space final description array are obtained, the second analysis space description array is extracted from the plurality of separation images, the second analysis space temporary description array and the second analysis space final description array are obtained, the description array fusion is performed on the first analysis space temporary description array and the second analysis space temporary description array corresponding to the plurality of separation images, the target fusion description array corresponding to the plurality of separation images is obtained, and the target fusion description array obtained through the description array fusion order contains nesting information of different analysis spaces. And then, extracting the image paraphrase description array through the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images to obtain the image paraphrase description array corresponding to the multiple separated images, so that the extracted image paraphrase description array simultaneously contains the information of the two analysis spaces, and simultaneously the extracted image paraphrase description array can restore the basic properties of the images to the utmost extent. Then, image defect detection is carried out through the image paraphrase description array to obtain the defect credibility coefficients of the flexible circuit board corresponding to the multiple separated images, so that the accuracy of image defect detection can be improved, then, multiple defect image chains are determined from the image sequence to be identified through the defect credibility coefficients of the flexible circuit board, and the defect paraphrase description arrays corresponding to the multiple defect image chains are determined through the image paraphrase description array; and identifying the defects of the defect image chains through the defect paraphrase description arrays corresponding to the defect image chains to obtain a unified defect image chain combination, so that the accuracy of identifying the defects of the defect image chains is improved, and the accuracy of the obtained unified defect image chain combination is further ensured.
Based on the same principle as the method shown in fig. 1, the embodiment of the present application further provides a defect detecting apparatus 10, as shown in fig. 3, where the apparatus 10 includes:
the separation module 11 is configured to acquire an image sequence to be identified, and separate the image sequence to be identified to obtain a plurality of separated images; the image sequence to be identified comprises a plurality of images of the flexible circuit board to be detected.
The first extraction module 12 is configured to perform first analysis space description array extraction on the multiple separated images respectively to obtain first analysis space description arrays corresponding to the multiple separated images, where the first analysis space description array includes a first analysis space temporary description array and a first analysis space final description array.
And a second extraction module 13, configured to perform second analysis space description array extraction on the multiple separated images respectively to obtain second analysis space description arrays corresponding to the multiple separated images, where the second analysis space description array includes a second analysis space temporary description array and a second analysis space final description array.
And the fusion module 14 is configured to perform description array fusion through the first analysis space temporary description array and the second analysis space temporary description array corresponding to the multiple separation images to obtain a target fusion description array corresponding to the multiple separation images.
And the credibility coefficient determining module 15 is configured to extract the image paraphrase description array through the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images to obtain the image paraphrase description array corresponding to the multiple separated images, and perform image defect detection through the image paraphrase description array to obtain the credibility coefficients of the defects of the flexible printed circuit board corresponding to the multiple separated images.
And the defect determining module 16 is configured to determine a plurality of defect image chains from the image sequence to be identified through the defect confidence coefficients of the flexible printed circuit board, and determine a defect paraphrase description array corresponding to the plurality of defect image chains through the image paraphrase description array.
And the combination module 17 is used for identifying the defects of the defect image chains through the defect paraphrase description arrays corresponding to the defect image chains to obtain a unified defect image chain combination.
The above embodiment introduces the defect detecting apparatus 10 from the perspective of a virtual module, and the following introduces a defect detecting system from the perspective of a physical module, as follows:
an embodiment of the present application provides a defect detection system, as shown in fig. 4, a defect detection system 100 includes: a processor 101 and a memory 103. Wherein the processor 101 is coupled to the memory 103, such as via a bus 102. Optionally, the defect detection system 100 may further include a transceiver 104. It should be noted that the transceiver 104 is not limited to one in practical applications, and the structure of the defect detection system 100 is not limited to the embodiment of the present application.
The processor 101 may be a CPU, general purpose processor, GPU, DSP, ASIC, FPGA or other programmable logic device, transistor logic device, hardware component, or any combination thereof. Which may implement or perform the various illustrative logical blocks, modules, and circuits described in connection with the disclosure. The processor 101 may also be a combination of computing functions, e.g., comprising one or more microprocessors, DSPs, and microprocessors.
Bus 102 may include a path that carries information between the aforementioned components. The bus 102 may be a PCI bus or an EISA bus, etc. The bus 102 may be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, only one thick line is shown in FIG. 4, but this does not indicate only one bus or one type of bus.
The memory 103 may be, but is not limited to, a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, an EEPROM, a CD-ROM or other optical disk storage, optical disk storage (including compact disk, laser disk, optical disk, digital versatile disk, blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer.
The memory 103 is used for storing application program codes for executing the scheme of the application, and is controlled by the processor 101 to execute. The processor 101 is configured to execute application program code stored in the memory 103 to implement the aspects illustrated in any of the method embodiments described above.
The embodiment of the present application provides a defect detecting system, and the defect detecting system in the embodiment of the present application includes: one or more processors; a memory; one or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more programs are executed by the processors, the method for detecting defects of a flexible printed circuit based on image processing is realized. According to the technical scheme, a plurality of separation images are obtained by separating an image sequence to be recognized, then, first analysis space description array extraction is carried out on the plurality of separation images respectively, a first analysis space temporary description array and a first analysis space final description array are obtained, second analysis space description array extraction is carried out on the plurality of separation images respectively, a second analysis space temporary description array and a second analysis space final description array are obtained, description array fusion is carried out on the first analysis space temporary description array and the second analysis space temporary description array corresponding to the plurality of separation images, target fusion description arrays corresponding to the plurality of separation images are obtained, and the target fusion description arrays obtained through the description array fusion include nesting information of different analysis spaces. And then, extracting the image paraphrase description array through the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images to obtain the image paraphrase description array corresponding to the multiple separated images, so that the extracted image paraphrase description array simultaneously contains the information of the two analysis spaces, and simultaneously the extracted image paraphrase description array can restore the basic properties of the images to the utmost extent. Then, image defect detection is carried out through the image paraphrase description array to obtain the defect credibility coefficients of the flexible circuit board corresponding to the multiple separated images, so that the accuracy of image defect detection can be improved, then, multiple defect image chains are determined from the image sequence to be identified through the defect credibility coefficients of the flexible circuit board, and the defect paraphrase description arrays corresponding to the multiple defect image chains are determined through the image paraphrase description array; and identifying the defects of the defect image chains through the defect paraphrase description arrays corresponding to the defect image chains to obtain a unified defect image chain combination, so that the accuracy of identifying the defects of the defect image chains is improved, and the accuracy of the obtained unified defect image chain combination is further ensured.
The present application provides a computer-readable storage medium, on which a computer program is stored, which, when running on a processor, enables the processor to execute the corresponding content in the foregoing method embodiments.
It should be understood that, although the steps in the flowcharts of the figures are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and may be performed in other orders unless explicitly stated herein. Moreover, at least a portion of the steps in the flow chart of the figure may include multiple sub-steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed alternately or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
The foregoing is only a partial embodiment of the present application, and it should be noted that, for those skilled in the art, several modifications and decorations can be made without departing from the principle of the present application, and these modifications and decorations should also be regarded as the protection scope of the present application.

Claims (10)

1. A method for detecting defects of a flexible printed circuit based on image processing is applied to a defect detection system, and comprises the following steps:
acquiring an image sequence to be identified, and separating the image sequence to be identified to obtain a plurality of separated images; the image sequence to be identified comprises a plurality of images of the flexible circuit board to be detected;
respectively extracting a first analysis space description array from the multiple separated images to obtain a first analysis space description array corresponding to the multiple separated images, wherein the first analysis space description array comprises a first analysis space temporary description array and a first analysis space final description array;
respectively extracting a second analysis space description array from the multiple separation images to obtain second analysis space description arrays corresponding to the multiple separation images, wherein the second analysis space description arrays comprise a second analysis space temporary description array and a second analysis space final description array;
performing description array fusion through a first analysis space temporary description array and a second analysis space temporary description array corresponding to the multiple separation images to obtain target fusion description arrays corresponding to the multiple separation images;
extracting image definition description arrays through a first analysis space final description array, a second analysis space final description array and a target fusion description array corresponding to the multiple separated images to obtain image definition description arrays corresponding to the multiple separated images, and performing image defect detection through the image definition description arrays to obtain defect credibility coefficients of the flexible circuit board corresponding to the multiple separated images;
determining a plurality of defect image chains from the image sequence to be identified through the credibility coefficients of the defects of the flexible printed circuit board, and determining defect paraphrase description arrays corresponding to the plurality of defect image chains through the image paraphrase description arrays;
and identifying the defects of the defect image chains through the defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain a unified defect image chain combination.
2. The method of claim 1, wherein the identifying defects in the defect image chain by the defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain a unified defect image chain combination comprises:
performing specified array extraction through the defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain a selected description array;
analyzing and restoring through the selected description array and the defect credibility coefficients of the flexible circuit board corresponding to the multiple separated images to obtain target defect paraphrase description arrays corresponding to the multiple defect image chains;
and identifying the defects of the plurality of defect image chains through the target defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain the unified defect image chain combination.
3. The method of claim 2, wherein the performing specified array extraction on the defect paraphrase description arrays corresponding to the plurality of defect image chains to obtain a selected description array comprises:
extracting original image description arrays corresponding to the multiple separated images respectively, and determining defect image chain original description arrays corresponding to the multiple defect image chains from the original image description arrays corresponding to the multiple separated images respectively;
integrating original description arrays of the defect image chains corresponding to the defect image chains with corresponding defect paraphrase description arrays respectively to obtain target integrated description arrays corresponding to the defect image chains;
and putting the target integration description array corresponding to the plurality of defect image chains into a description array extraction network of a description array processing model for processing to obtain a target selection description array.
4. The method of claim 2, wherein the identifying the defects of the defect image chains by the target defect paraphrase descriptor arrays corresponding to the defect image chains to obtain the unified defect image chain combination comprises:
determining the matching degree among the plurality of defect image chains through target defect paraphrase description arrays corresponding to the plurality of defect image chains;
and performing image bucket division according to the matching degree among the plurality of defect image chains to obtain the unified defect image chain combination.
5. The method according to claim 1, wherein the extracting the first analysis space description array for the plurality of separated images respectively to obtain the first analysis space description array corresponding to the plurality of separated images, the first analysis space description array including a first analysis space temporary description array and a first analysis space final description array includes:
respectively carrying out first analysis space convolution operation on the multiple separated images to obtain multiple temporary convolution description arrays and ending convolution description arrays corresponding to the multiple separated images;
performing dimension transformation operation on the plurality of temporary convolution description arrays to obtain a plurality of first analysis space temporary description arrays corresponding to the plurality of separation images;
carrying out dimension transformation operation on the ending convolution description array to obtain a first analysis space final description array corresponding to the multiple separation images;
the extracting of the second analysis space description array is respectively carried out on the multiple separated images to obtain the second analysis space description array corresponding to the multiple separated images, and the second analysis space description array comprises a second analysis space temporary description array and a second analysis space final description array, and the extracting of the second analysis space description array comprises the following steps:
extracting original image description arrays corresponding to the multiple separated images respectively;
and performing second analysis space convolution operation on the original image description arrays corresponding to the multiple separated images to obtain multiple second analysis space temporary description arrays and second analysis space final description arrays corresponding to the multiple separated images.
6. The method according to claim 1, wherein the number of the first analysis space temporary description arrays is plural, and the number of the second analysis space temporary description arrays is plural;
the obtaining of the target fusion description array corresponding to the multiple separation images by performing description array fusion through the first analysis space temporary description array and the second analysis space temporary description array corresponding to the multiple separation images includes:
integrating a first analysis space temporary description array tensor1-1 in the plurality of first analysis space temporary description arrays with a corresponding second analysis space temporary description array tensor2-1 in the plurality of second analysis space temporary description arrays to obtain a first integrated description array, and performing convolution processing through the first integrated description array to obtain a first fusion description array;
integrating the first fusion description array, a first analysis space temporary description array tensor1-2 in the plurality of first analysis space temporary description arrays and a corresponding second analysis space temporary description array tensor2-2 in the plurality of second analysis space temporary description arrays to obtain a second integration description array, and performing convolution processing through the second integration description array to obtain a second fusion description array;
and when the plurality of first analysis space temporary description arrays and the plurality of second analysis space temporary description arrays are finished, obtaining a target fusion description array.
7. The method of claim 1, wherein the extracting image definition description arrays through the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the plurality of separated images to obtain the image definition description arrays corresponding to the plurality of separated images, and performing image defect detection through the image definition description arrays to obtain the defect confidence coefficients of the flexible printed circuit board corresponding to the plurality of separated images comprises:
integrating a first analysis space final description array, a second analysis space final description array and a target fusion description array corresponding to the multiple separation images to obtain a target integration description array corresponding to the multiple separation images;
performing convolution processing through the target integration description arrays corresponding to the multiple separated images to obtain convolution description arrays corresponding to the multiple separated images;
obtaining a description array maximum value and a description array average value corresponding to each order in the convolution description array through the convolution description arrays corresponding to the multiple separated images;
summing the maximum value of the description array and the average value of the description array to obtain paraphrase extraction description array values corresponding to all orders in the convolution description array, and obtaining the paraphrase extraction description array values corresponding to the multiple separated images through the paraphrase extraction description array values corresponding to all orders in the convolution description array;
performing linear transformation on the paraphrase extraction description arrays corresponding to the multiple separated images to obtain image paraphrase description arrays corresponding to the multiple separated images;
and detecting and judging the defects and non-defects of the flexible circuit board through the image definition description arrays corresponding to the multiple separated images to obtain the credibility coefficients of the defects of the flexible circuit board corresponding to the multiple separated images.
8. The method of claim 1, further comprising:
putting the image sequence to be identified into an image defect detection model, and separating the image sequence to be identified through the image defect detection model to obtain a plurality of separated images;
respectively performing first analysis space description array extraction on the multiple separated images through the image defect detection model to obtain first analysis space description arrays corresponding to the multiple separated images, wherein the first analysis space description arrays comprise a first analysis space temporary description array and a first analysis space final description array;
respectively extracting a second analysis space description array from the multiple separation images to obtain second analysis space description arrays corresponding to the multiple separation images, wherein the second analysis space description arrays comprise a second analysis space temporary description array and a second analysis space final description array;
performing description array fusion on a first analysis space temporary description array and a second analysis space temporary description array corresponding to the multiple separated images through the image defect detection model to obtain target fusion description arrays corresponding to the multiple separated images;
and extracting image definition description arrays from the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images through the image defect detection model to obtain the image definition description arrays corresponding to the multiple separated images, and detecting image defects through the image definition description arrays to obtain the defect reliability coefficients of the flexible printed circuit board corresponding to the multiple separated images.
9. The method of claim 8, wherein the image defect detection model comprises a first analysis space description array extraction network, a second analysis space description array extraction network, a description array fusion network, an image paraphrasing description array extraction network, and a defect discrimination network;
the method further comprises the following steps:
putting the image sequence to be identified into an image defect detection model, and separating the image sequence to be identified through the image defect detection model to obtain a plurality of separated images;
putting the plurality of separated images into the first analysis space description array extraction network to extract a second analysis space description array to obtain a first analysis space temporary description array and a first analysis space final description array;
putting the plurality of separated images into a second analysis space description array extraction network to extract a second analysis space description array so as to obtain a second analysis space temporary description array and a second analysis space final description array;
putting a first analysis space temporary description array and a second analysis space temporary description array corresponding to a plurality of separation images into the description array fusion network for description array fusion to obtain target fusion description arrays corresponding to the plurality of separation images;
and putting the first analysis space final description array, the second analysis space final description array and the target fusion description array corresponding to the multiple separated images into the image paraphrase description array extraction network for image paraphrase description array extraction to obtain image paraphrase description arrays corresponding to the multiple separated images, and putting the image paraphrase description arrays into the defect judgment network for image defect detection to obtain the defect credibility coefficients of the flexible printed circuit corresponding to the multiple separated images.
10. A defect detection system comprising a processor and a memory, the memory storing a computer program that, when executed by the processor, performs the method of claim 1~9.
CN202211440566.7A 2022-11-17 2022-11-17 Flexible circuit board defect detection method and system based on image processing Active CN115641330B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211440566.7A CN115641330B (en) 2022-11-17 2022-11-17 Flexible circuit board defect detection method and system based on image processing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211440566.7A CN115641330B (en) 2022-11-17 2022-11-17 Flexible circuit board defect detection method and system based on image processing

Publications (2)

Publication Number Publication Date
CN115641330A true CN115641330A (en) 2023-01-24
CN115641330B CN115641330B (en) 2023-06-30

Family

ID=84948196

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211440566.7A Active CN115641330B (en) 2022-11-17 2022-11-17 Flexible circuit board defect detection method and system based on image processing

Country Status (1)

Country Link
CN (1) CN115641330B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020125229A1 (en) * 2018-12-20 2020-06-25 深圳云天励飞技术有限公司 Feature fusion method and apparatus, and electronic device and storage medium
CN113516661A (en) * 2021-09-15 2021-10-19 深圳新视智科技术有限公司 Defect detection method and device based on feature fusion
CN113554611A (en) * 2021-07-19 2021-10-26 广东电网有限责任公司 Insulator self-explosion defect detection method and device, terminal and storage medium
CN113989600A (en) * 2021-09-28 2022-01-28 上海师范大学 Image foreign matter detection method
CN114612411A (en) * 2022-03-04 2022-06-10 腾讯科技（深圳）有限公司 Image tampering detection method, device, equipment and storage medium

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020125229A1 (en) * 2018-12-20 2020-06-25 深圳云天励飞技术有限公司 Feature fusion method and apparatus, and electronic device and storage medium
CN113554611A (en) * 2021-07-19 2021-10-26 广东电网有限责任公司 Insulator self-explosion defect detection method and device, terminal and storage medium
CN113516661A (en) * 2021-09-15 2021-10-19 深圳新视智科技术有限公司 Defect detection method and device based on feature fusion
CN113989600A (en) * 2021-09-28 2022-01-28 上海师范大学 Image foreign matter detection method
CN114612411A (en) * 2022-03-04 2022-06-10 腾讯科技（深圳）有限公司 Image tampering detection method, device, equipment and storage medium

Also Published As

Publication number Publication date
CN115641330B (en) 2023-06-30

Similar Documents

Publication Publication Date Title
CN112199375A (en) Cross-modal data processing method and device, storage medium and electronic device
CN112446870B (en) Pipeline damage detection method, device, equipment and storage medium
CN111210412B (en) Packaging detection method and device, electronic equipment and storage medium
CN113870254B (en) Target object detection method and device, electronic equipment and storage medium
WO2023124278A1 (en) Image processing model training method and apparatus, and image classification method and apparatus
CN113724219A (en) Building surface disease detection method and system based on convolutional neural network
CN117011260A (en) Automatic chip appearance defect detection method, electronic equipment and storage medium
CN113139549A (en) Parameter self-adaptive panorama segmentation method based on multitask learning
CN113191235A (en) Sundry detection method, device, equipment and storage medium
CN114596273B (en) Intelligent detection method for multiple defects of ceramic substrate by using YOLOV4 network
CN115546211A (en) Welding spot defect classification method, terminal and computer storage medium
Wang et al. A transformer-based mask R-CNN for tomato detection and segmentation
CN111091122A (en) Training and detecting method and device for multi-scale feature convolutional neural network
Shang et al. Instance segmentation by learning deep feature in embedding space
CN115641330A (en) Flexible circuit board defect detection method and system based on image processing
Huo Development of a real-time Printed Circuit board (PCB) visual inspection system using You Only Look Once (YOLO) and fuzzy logic algorithms
CN115937492A (en) Transformer equipment infrared image identification method based on feature identification
CN115063628A (en) Fruit picking prediction method based on visual semantic segmentation
Li et al. Focus on local: transmission line defect detection via feature refinement
CN117333493B (en) Machine vision-based detection system and method for production of display base
CN117935174B (en) Intelligent management system and method for vacuum bag film production line
CN113407439B (en) Detection method for software self-recognition type technical liabilities
Wang et al. An Improved YOLOv5x-Based Algorithm for IC Pin Welding Defects Detection
CN118822989A (en) PCB surface defect detection method and system based on RT-DETR detector
Chiu et al. Development and validation of a real-time vision-based automatic HDMI wire-split inspection system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant