CN114708266A - Tool, method and device for detecting card defects and medium - Google Patents
Tool, method and device for detecting card defects and medium Download PDFInfo
- Publication number
- CN114708266A CN114708266A CN202210632046.XA CN202210632046A CN114708266A CN 114708266 A CN114708266 A CN 114708266A CN 202210632046 A CN202210632046 A CN 202210632046A CN 114708266 A CN114708266 A CN 114708266A
- Authority
- CN
- China
- Prior art keywords
- card
- defects
- image
- detection model
- training sample
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Evolutionary Biology (AREA)
- Quality & Reliability (AREA)
- Signal Processing (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a tool, a method, a device and a medium for detecting card defects, which are applied to the field of image recognition. The method is applied to a detection tool for the defects of the card comprising a camera, a light source and a processor. The light source irradiates the card, and after the camera shoots the card image, the card image is input into a pre-trained card defect detection model to obtain the defects on the card. The card defect detection model is a target detection and segmentation algorithm based on deep learning; the training card defect detection model comprises: marking the defects on the training sample image of the defective card; inputting the training sample image into a card defect detection model to obtain the defect in the training sample image; and iteratively updating the weight of each layer of network according to the defect marked in the training sample image and the difference data between the defects obtained according to the card defect detection model so as to reduce the difference data. The method can detect the defects on the card so as to avoid influencing the use in daily life.
Description
Technical Field
The application relates to the field of image recognition, in particular to a tool, a method, a device and a medium for detecting card defects.
Background
The identity card is an important certificate for proving the identity of residents, the use frequency is very high, and if unqualified certificates occur, the difficulties of work, life, learning and the like can be caused to people, so the identity card manufacturing and quality inspection work is very important.
However, the existing automatic detection method for surface defects of cards such as identification cards is basically in a blank state, and the defective cards cannot be detected in time, so that the quality of the cards cannot be ensured.
Therefore, how to ensure the quality of the card is an urgent problem to be solved by those skilled in the art.
Disclosure of Invention
The application aims to provide a detection tool, a detection method, a detection device and a detection medium for card defects so as to ensure the quality of cards.
In order to solve the technical problem, the application provides a method for detecting a card defect, which is applied to a tool for detecting the card defect comprising a camera, a light source and a processor, and the method comprises the following steps:
controlling the light source to irradiate the card to be detected;
acquiring a card image shot by the camera;
inputting the card image into a pre-trained card defect detection model to obtain the defects on the card; the card defect detection model is a target detection and segmentation algorithm based on deep learning; training the card defect detection model comprises: marking the defects on the training sample image of the defective card; inputting the training sample image into the card defect detection model to obtain the defect in the training sample image; and iteratively updating the weight of each layer of network according to the difference data between the defects marked in the training sample image and the defects obtained according to the card defect detection model so as to reduce the difference data.
Preferably, before the inputting the training sample image into the card defect detection model to obtain the defect in the training sample image, the method further includes:
cutting and splicing the training sample images to obtain the training sample images with uniformly distributed defects;
the step of inputting the training sample image into the card defect detection model to obtain the defect in the training sample image comprises:
and inputting the training sample image with uniformly distributed defects into the card defect detection model to obtain the defects in the training sample image.
Preferably, the card defect detection model is a feature extraction backbone network based on Resnet 50;
inputting the card image and the training sample image into the card defect detection model to obtain the defect on the card comprises:
inputting the card image and the training sample image into Resnet50 to extract a plurality of feature information of different scales;
and inputting the characteristic information into the FPN for multi-scale information fusion so as to output the position coordinates of the defects, the probability values of the defects and semantic information of the defect regions.
Preferably, the light source includes a coaxial light source and a strip light source, and the controlling the light source to irradiate the card to be detected includes:
controlling the coaxial light source to irradiate from the right above the card;
controlling the strip light source to irradiate from each direction above the card;
the obtaining of the card image shot by the camera comprises:
and respectively acquiring the card image under the coaxial light source and the card image under the strip light source.
Preferably, before the inputting the card image into a pre-trained card defect detection model to obtain the defect on the card, the method further includes:
respectively cutting the card image under the coaxial light source and the card image under the strip light source to obtain a plurality of sub-images;
performing traditional morphological preprocessing on the subimages to obtain characteristic data;
the inputting the card image into a pre-trained card defect detection model to obtain the defects on the card comprises:
and inputting the characteristic data into the card defect detection model trained in advance to obtain the defects on the card.
Preferably, the conventional morphological preprocessing of the sub-images to obtain feature data comprises:
carrying out gray scale transformation on the sub-image;
performing gray gradient detection on the sub-image after gray conversion by using a Canny edge detection method based on gray gradient to extract gray gradient characteristics of the card image under the irradiation of the coaxial light source and the strip light source; wherein the gray gradient feature comprises a shading feature of the card and a defect feature of the card, and the gray gradient feature is the feature data.
For solving above-mentioned technical problem, this application still provides a detection frock of card defect, includes: the device comprises a camera, a light source and a processor;
the processor is connected with the light source and used for controlling the light source to irradiate the card to be detected;
the processor is also connected with the camera and used for acquiring a card image shot by the camera; the processor is also used for inputting the card image into a pre-trained card defect detection model so as to obtain the defects on the card; the card defect detection model is a target detection and segmentation algorithm based on deep learning; training the card defect detection model comprises: marking the defects on the training sample image of the defective card; inputting the training sample image into the card defect detection model to obtain the defect in the training sample image; and iteratively updating the weight of each layer of network according to the difference data between the defects marked in the training sample image and the defects obtained according to the card defect detection model so as to reduce the difference data.
For solving above-mentioned technical problem, this application still provides a detection device of card defect, is applied to the detection frock of the card defect including camera, light source and treater, the device includes:
the control module is used for controlling the light source to irradiate the card to be detected;
the acquisition module is used for acquiring the card image shot by the camera;
the input module is used for inputting the card image into a pre-trained card defect detection model so as to obtain the defects on the card; the card defect detection model is a target detection and segmentation algorithm based on deep learning; training the card defect detection model comprises: marking the defects on the training sample image of the defective card; inputting the training sample image into the card defect detection model to obtain the defect in the training sample image; and iteratively updating the weight of each layer of network according to the difference data between the defects marked in the training sample image and the defects obtained according to the card defect detection model so as to reduce the difference data.
Preferably, the apparatus for detecting a card defect further comprises: the splicing module is used for shearing and splicing the training sample images to obtain the training sample images with uniformly distributed defects before inputting the training sample images into the card defect detection model to obtain the defects in the training sample images; wherein, the defects in the training sample image are uniformly distributed; the step of inputting the training sample image into the card defect detection model to obtain the defects in the training sample image comprises: and inputting the training sample image with uniformly distributed defects into the card defect detection model to obtain the defects in the training sample image.
Preferably, the apparatus for detecting a card defect further comprises: the cutting module is used for respectively cutting the card image under the coaxial light source and the card image under the strip light source to obtain a plurality of sub-images before inputting the card image into a pre-trained card defect detection model to obtain defects on the card;
the preprocessing module is used for carrying out traditional morphological preprocessing on the sub-images to obtain characteristic data; the inputting the card image into a pre-trained card defect detection model to obtain the defects on the card comprises: and inputting the characteristic data into the pre-trained card defect detection model to obtain the defects on the card.
In order to solve the above technical problem, the present application further provides a device for detecting a card defect, including: a memory for storing a computer program;
and the processor is used for realizing the steps of the card defect detection method when executing the computer program.
In order to solve the above technical problem, the present application further provides a computer-readable storage medium, where a computer program is stored, and when the computer program is executed by a processor, the steps of the method for detecting a card defect are implemented.
The detection method for the card defects is applied to a detection tool for the card defects, wherein the detection tool comprises a camera, a light source and a processor. The processor controls the light source to irradiate the card to be detected, acquires a card image shot by the camera, and inputs the card image into a card defect detection model trained in advance to acquire defects on the card. The card defect detection model is a target detection and segmentation algorithm based on deep learning; the training card defect detection model comprises: marking the defects on the training sample image of the defective card; inputting the training sample image into a card defect detection model to obtain the defect in the training sample image; and iteratively updating the weight of each layer of network according to the defect marked in the training sample image and the difference data between the defects obtained according to the card defect detection model so as to reduce the difference data. The defect detection model of the card after training can accurately detect the defect of the card, and after the defect on the card is detected, the defective card can be screened out, so that the quality of the card is ensured, and the defect of the card is avoided, thereby influencing the use in daily life.
The application also provides a tool, a device and a computer readable storage medium for detecting the card defects, which correspond to the method, so that the tool, the device and the computer readable storage medium have the same beneficial effects as the method.
Drawings
In order to more clearly illustrate the embodiments of the present application, the drawings needed for the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings can be obtained by those skilled in the art without inventive effort.
FIG. 1 is a flowchart illustrating a method for detecting a card defect according to an embodiment of the present disclosure;
FIG. 2 is a sample number diagram of various defects provided by an embodiment of the present application;
FIG. 3 is a schematic diagram of a method for performing conventional morphological pre-processing on a sub-image according to an embodiment of the present application;
FIG. 4 is a diagram of an example neural network for segmentation and object detection based on yolact improvement according to an embodiment of the present disclosure;
FIG. 5 is a block diagram of a multi-scale information fusion phase;
fig. 6 is a structural diagram of a card defect detection apparatus according to an embodiment of the present application;
fig. 7 is a structural diagram of a device for detecting a card defect according to another embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all the embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application without any creative effort belong to the protection scope of the present application.
The core of the application is to provide a detection tool, a detection method, a detection device and a detection medium for the defects of the card so as to ensure the quality of the card.
In order that those skilled in the art will better understand the disclosure, the following detailed description will be given with reference to the accompanying drawings.
The card is often used as various certification cards to meet daily requirements, such as identity cards, social security cards, medical cards and the like, and is convenient to carry and use. However, when these documents are manufactured, defective products may occur, and if there is a defect in the card such as an identification card, the use of the identification card is inconvenient. In order to solve the above problems, an embodiment of the present application provides a method for detecting a card defect, which is applied to a tool for detecting a card defect, the tool comprising a camera, a light source and a processor, wherein in the tool for detecting a card defect, the processor is connected with the light source to control the light source to irradiate a card to be detected; the processor is also connected with the camera and used for controlling the camera to shoot the card and acquiring the card image shot by the camera. It should be noted that the type of the processor, the number and type of the light source and the camera, and the connection manner of the processor, the camera and the light source are not limited. Camera and light source generally set up in adjacent position, and set up in the top that the card was placed, and the card is shone to the light source, and the camera can acquire clear card image. In fact, in order to better recognize the image of the card, some special light sources are adopted to illuminate the card, and a coaxial light source and a strip light source are generally used; correspondingly, the camera can acquire card images under the coaxial light source and the strip light source respectively, and the processor inputs the acquired card images into a pre-trained card defect detection model so as to acquire defects on the card. The application scene of this application is generally after the card is produced, when carrying out quality inspection to the card, generally has a large amount of cards and need examine, and accessible transmission band places the card in proper order in the camera below, and after every card arrived assigned position, control transmission band pause a period so that the clear card image is shot to the camera. FIG. 1 is a flowchart illustrating a method for detecting a card defect according to an embodiment of the present disclosure; as shown in fig. 1, the method comprises the steps of:
s10: and controlling the light source to irradiate the card to be detected.
In the specific implementation, the type and number of the light sources are not limited, but as a preferable scheme, a combination of a coaxial light source and a strip light source can be adopted, and the processor controls the coaxial light source and the strip light source to respectively irradiate the card to be detected. Here, the order in which the coaxial light source and the bar light source irradiate the card is not limited, but a program for processing the card image in the processor needs to be associated, and the card image obtained when the coaxial light source and the bar light source irradiate the card is different, and therefore, the processing manner by the processor is also different. In addition, the position of the light source is not required, generally the coaxial light source is arranged right above the card placing position, and the strip-shaped light source is arranged in four directions of front, back, left and right above the card placing position.
S11: and acquiring the card image shot by the camera.
The quantity of camera, the position and the resolution ratio of installation do not do all do not limit yet, generally install directly over the card position of placing can, under some circumstances, for the defect of more accurate detection card, can set up the camera in order to obtain the card image of a plurality of different directions in a plurality of directions.
S12: and inputting the card image into a pre-trained card defect detection model to acquire the defects on the card.
The card defect detection model is a target detection and segmentation algorithm based on deep learning; the training card defect detection model comprises: marking the defects on the training sample image of the defective card; inputting the training sample image into a card defect detection model to obtain the defect in the training sample image; and iteratively updating the weight of each layer of network according to the defect marked in the training sample image and the difference data between the defects obtained according to the card defect detection model so as to reduce the difference data.
The defect detection of the identity card is taken as an example for explanation, because the size of the defects such as bubbles, edges, scratches and the like on the identity card is very small, in order to quickly detect the defects, firstly, a high-resolution industrial camera is used for respectively acquiring an identity card image of the identity card under the irradiation of a coaxial light source and an identity card image under the irradiation of a strip-shaped light source, then, an original large image is divided into sub-images with fixed sizes according to the size of the identity card image according to a certain proportion, then, after the traditional morphological pretreatment is carried out on all the sub-images, the sub-images of a coaxial lamp and an annular lamp are input into a pre-trained identity card defect detection model through a plurality of channels for detection, and the position of the defect in the identity card image, the probability value of the defect and the semantic information of all pixels in the defect area are obtained.
After the identity card defect detection model needs to be trained in advance and a deep learning-based target detection and segmentation algorithm (identity card defect detection model) is established, the specific training mode is as follows: the method comprises the steps of automatically collecting an identity card image through related equipment, manually marking the position of a defect in the collected identity card image, respectively marking coordinates of two vertexes of the upper left corner and the lower right corner of the defect, processing the coordinates into a rectangle, and acquiring semantic information of all regions of the rectangle. The labeled region can then be randomly copied and pasted to other positions in the ID card image to obtain the required training data (training sample image). Inputting the marked training data into a designed identity card defect detection model, performing multi-scale extraction on the input training data by using a neural network to obtain required features, and finally performing regression on the neural network to obtain the position coordinates of the defects in the image, the probability value of the defects and semantic information of the region. And comparing the regressed position, probability value and semantic information of the defect with the manually marked result to obtain difference data between the position, probability value and semantic information, and then reversely calculating the weight of each layer of network to perform iterative updating so as to finally obtain a neural network model which enables the difference between the two results to be minimum.
The method for detecting the card defects is applied to a detection tool for the card defects, wherein the detection tool comprises a camera, a light source and a processor. The processor controls the light source to irradiate the card to be detected, acquires a card image shot by the camera, and inputs the card image into a card defect detection model trained in advance to acquire defects on the card. The card defect detection model is a target detection and segmentation algorithm based on deep learning; the training card defect detection model comprises: marking the defects on the training sample image of the defective card; inputting the training sample image into a card defect detection model to obtain the defect in the training sample image; and iteratively updating the weight of each layer of network according to the defect marked in the training sample image and the difference data between the defects obtained according to the card defect detection model so as to reduce the difference data. The defect detection model of the card after training can accurately detect the defect of the card, and after the defect on the card is detected, the defective card can be screened out, so that the quality of the card is ensured, and the defect of the card is avoided, thereby influencing the use in daily life.
The above embodiment mentions that training sample images are needed to train the card defect detection model, and in the training stage, because the probability of each defect actually appearing on the card is different, the number of samples of partial defects is small, such as bubbles, scratches and the like. FIG. 2 is a sample number diagram of various defects provided by an embodiment of the present application; as shown in fig. 2, the number of samples of defects such as blank stains, character stains, etc. is large, and the number of samples of the remaining defects is small, such as bubbles, raised edges, scratches, fibers, white spots, etc., particularly scratches and white spots, which are small in number. This may cause the image data of each training sample to be unbalanced, which may easily cause over-fitting of the neural network, i.e. the type with large sample amount has high precision, and the precision with small amount is poor. Therefore, before inputting the training sample image into the card defect detection model to obtain the defect in the training sample image, the method provided by the embodiment of the application further includes: shearing and splicing all training sample images to obtain training sample images with uniformly distributed defects; inputting the training sample image into a card defect detection model to obtain the defects in the training sample image comprises the following steps: and inputting the training sample images with the defects distributed uniformly into the card defect detection model to obtain the defects in the training sample images.
In the scheme provided by the embodiment of the application, the defects in the images of the training samples are uniformly distributed, and the number of the samples can be increased through continuous shearing and splicing, for example, 4 samples are randomly selected from 100 samples and spliced into a new sample. The image size in the embodiment of the present application is 2592 × 1944 as an example, when a pre-trained deep neural network is used for Processing, the image is divided according to the size of 324 × 243 according to the principle of 8 × 8 equal division, 64 sub-images are totally divided, then the batch size is set to 64 by using the parallel computing function of a Graphics Processing Unit (GPU), and the 64 sub-images are simultaneously input to the neural network for computing.
The following three strategies are provided to solve the problem of overfitting, but the specific implementation is not limited to these three strategies, and any combination of these three strategies may be adopted. The first strategy is to copy the pixels of the whole defect area by using semantic information of the manually marked defect area, then take a random number P1 in the range of [0,1], and paste the defect area to the pixel position L of the card image by calculation, wherein the specific calculation of L is shown in formula (1).
L=（P1*W，P1*H） （1）
Wherein W is the width of the card image and H is the height of the card image.
The second strategy is also to take a random number P2 in the range of 0,1 and then calculate the center point M, as shown in equation (2).
M=（P2*W/2，P2*H/2） （2）
Assuming that N samples are provided, then taking three groups of samples X, Y and Z with random numbers; the number of the three groups of samples X, Y and Z is in a value range of [0, N ], then an original image is divided into four sub-images by using a central point M, and the width and height sizes of the four sub-images I1, I2, I3 and I4 are respectively shown in formula (3), formula (4), formula (5) and formula (6).
I1（P2*W/2，P2*H/2） （3）
I2（W-P2*W/2，P2*H/2） （4）
I3（W-P2*W/2，H-P2*H/2） （5）
I4（P2*W/2，H-P2*H/2） （6）
The sub-images corresponding to the I1 are reserved, then one image is randomly selected from the X, the Y and the Z images and respectively scaled to obtain the sub-images with the sizes corresponding to the I2, the I3 and the I4, and then the new four sub-images are combined into a new image, so that the number of samples is increased through continuous combination, and information with different scales is increased through scaling. It should be noted that the four sub-images obtained by cutting the original image are different from the 64 sub-images obtained by cutting the card image, which are processed to cut the card image into a plurality of small images, and the cut sub-images are finally spliced into a new image. In fact, the sub-image in the above may be taken as the original image in the present example.
The third strategy is based on the second strategy, four sub-images are randomly cut from the original image by taking the marking point S (i, j) as the center, and the sizes of the four sub-images are different when the marking point S (i, j) is different. Firstly, two random numbers RW and RH are taken; the card image was cut to obtain 64 images of 324 × 243 size, and the 324 × 243 size images were used as the original images. The value ranges of the random numbers RW and RH are [ -324/2, 324/2], [ -243/2 and 243/2], then a new annotation point position is calculated to be S1 (i + RW, j + RH), and then the image is cut again by taking S1 as a central point, so that the position of each cutting is different, the neural network can have translation invariance, does not depend on the position information of the defect, and has robustness.
The card defect detection model mentioned in the above embodiment is a target detection and segmentation algorithm based on deep learning, that is, a deep neural network, and in actual application, the card defect detection model may be a backbone network extracted based on features of Resnet 50; correspondingly, inputting the card image and the training sample image into the card defect detection model to acquire the defects on the card comprises: inputting the card image and the training sample image into Resnet50 to extract a plurality of feature information of different scales; and then inputting the Feature information into a Feature Pyramid (FPN) for multi-scale information fusion to output defect position coordinates, defect probability values and semantic information of defect regions. The embodiment of the application provides a method for acquiring defects, and finally, defect position coordinates, namely specific positions of the defects in a card, probability values of the defects and semantic information of defect regions are obtained; the probability value of the defect refers to the probability that the area detected by the card defect detection model is the defect, namely the area only has a certain probability of being the real defect, and the card defect detection model outputs the probability value; the semantic information, i.e. the mask image, is because the obtained defect area is generally a rectangle, and actually, the specific area of the defect is a part of the inside of the rectangle, and after the semantic information is obtained, the precise position of the defect can be found.
The embodiment of the application provides a preferred scheme of light source, and the light source includes coaxial light source and bar light source, and the control light source shines the card that detects includes: controlling a coaxial light source to irradiate from the right upper part of the card; controlling the strip light source to irradiate from each direction above the card; the card image that obtains the camera and shoot includes: and respectively acquiring a card image under the coaxial light source and a card image under the strip light source. According to the scheme provided by the embodiment of the application, the coaxial light source and the strip-shaped light source are used for irradiating the card, and the defects on the card can be more easily identified by the acquired card image.
Generally, the whole card image is not input into the card defect detection model, but the card image is cut into smaller sub-images, and then the sub-images are processed, so that the card image is input into the card defect detection model trained in advance in the embodiment of the present application, so as to obtain the defects on the card, the method further includes: respectively cutting the card image under the coaxial light source and the card image under the strip light source to obtain a plurality of sub-images; and then performing traditional morphological preprocessing on the sub-images to obtain characteristic data. Correspondingly, inputting the card image into a pre-trained card defect detection model to acquire the defects on the card comprises: and inputting the characteristic data into a pre-trained card defect detection model to acquire the defects on the card. The cutting mode in the embodiment of the application is not limited, taking an identity card as an example, the size of an image of the identity card is 2592 × 1944, when the pre-trained deep neural network is used for processing, the image is divided according to the size of 324 × 243 according to the principle of 8 × 8 equal division, 64 sub-images are divided in total, then the batch size is set to 64 by using the parallel computing function of the GPU, and the 64 sub-images are simultaneously input into the neural network for computing.
The conventional morphological preprocessing of the sub-images mentioned in the above embodiments to obtain the feature data includes: carrying out gray scale transformation on the sub-images; carrying out gray gradient detection on the sub-image subjected to gray scale conversion by a Canny edge detection method based on gray gradient so as to extract the gray gradient characteristics of the card image under the irradiation of a coaxial light source and a strip light source; the gray gradient feature comprises a shading feature of the card and a defect feature of the card, and the gray gradient feature is feature data.
According to the scheme provided by the embodiment of the application, the traditional morphological preprocessing of the sub-image comprises the steps of carrying out gray level conversion on the sub-image, carrying out gray level gradient detection on the sub-image by adopting a Canny edge detection method based on gray level gradient, and extracting the gray level gradient characteristics of the surface of the card under the irradiation of the coaxial light source and the strip light source. FIG. 3 is a schematic diagram of a method for performing conventional morphological pre-processing on a sub-image according to an embodiment of the present application; as shown in fig. 3, the method comprises the steps of:
s20: and (5) partitioning the color image.
S21: and (5) gray level transformation.
S22: and (4) Gaussian filtering.
S23: the gray scale gradients in the x and y directions are calculated.
S24: non-maxima suppression.
S25: and (4) double-threshold screening.
S26: the final result is obtained.
The method comprises the steps of partitioning a colorful card image to obtain sub-images, carrying out gray scale conversion on all the sub-images to obtain a single-channel gray scale image, and then calculating the gray scale gradient characteristic of the gray scale image by using a Canny algorithm. The Canny algorithm is an algorithm with ideal effect and efficiency in the conventional edge extraction algorithm at present, compared with an edge extraction method such as sobel, the Canny algorithm has a more robust result, and particularly, a better result can be obtained for an image with unobvious gray gradient change, so that the method is selected for preprocessing the sub-image.
FIG. 4 is a diagram of an example neural network for segmentation and object detection based on yolact improvement according to an embodiment of the present disclosure; the process is only one implementation scheme of the application, and is mainly divided into a feature extraction stage, a multi-scale information fusion stage, a classification stage, a regression position and an extraction mask. All convolution calculations can be detected quickly at the GPU terminal after tensorRT acceleration.
The characteristic extraction stage mainly extracts the image characteristics by using convolution operation and pooling operation. The convolution operation is to set a window with the size of K x K on an image area, perform translational sliding on the window according to step length, then calculate the product weighted sum of the weight value in the K x K window and each data of the image, then assign the result to the central point of the current sliding window, and so on, gradually complete the product weighted operation of the whole image and the sliding window; and the pooling operation is to set a window with the size of K x K to slide in the image area, then calculate the maximum value or the average value in the current window, and then assign the maximum value or the average value to the center point of the current window, wherein the weight is not required to be saved, and 5 scales of characteristic information such as C1, C2, C3, C4, C5 and the like are obtained through 3 times of convolution down-sampling with the step size of 2 and 1 time of pooling down-sampling with the step size of 2.
After the convolution operation is completed, all the operations are linear weighted sums, so that the effect of fitting the characteristics is poor, and in order to make the fitting effect of the neural network more accurate, an activation function can be added. The activation function can increase the nonlinearity of the neural network, enable the fitting capability of the neural network to be stronger, and accelerate the gradient convergence of the neural network.
FIG. 5 is a block diagram of a multi-scale information fusion phase; as shown in fig. 5, feature information of three different scales of C3, C4 and C5 is fused. The feature map of C5 is convolved by a convolution kernel with the window size of 1 × 1, the number of feature channels of C4 and C5 is compressed to be the same as that of C4, then the feature map is sampled to a P3 feature map with the same size as that of C4 through bilinear interpolation, C4 and P3 are subjected to the operation of adding the features of each channel to obtain P2, then P2 is subjected to bilinear interpolation to obtain the size of C3, and P2 is added to the feature map of C3 which is convolved by 1 × 1 to obtain P1. This results in a fused output of three different scales. Because each layer of convolution features has feature maps with different scales, the feature maps contain stronger semantic information. The multi-scale information fusion can fuse the feature map with strong semantic information under low resolution and the feature map with weak semantic information but strong spatial information under high resolution on the premise of less calculation amount, and the method not only can obtain spatial context information of different scales, but also is more effective for positioning and detecting small target defects.
In the classification stage, a classification result score of each class is obtained after the classification result score is output through a full connection layer and is subjected to a normalized exponential function (softmax function). score is used to distinguish whether the current object belongs to the background or to a defective object. In the regression stage, 4 × N characteristic information is output through convolution operation, the positions of N targets are calculated sequentially through preset N anchor target screening boxes according to the characteristic information, then a result that the classification result score is larger than t is screened out through setting a threshold value t, and masks of all pixels with defects are obtained by combining the output semantic information. The method provided by the embodiment of the application can obtain higher detection precision only by one-time training, and meanwhile, the model of the method can be continuously optimized by continuously accumulating samples.
For solving above-mentioned technical problem, this application embodiment provides a detection frock of card defect, and the detection frock of card defect includes: camera, light source, treater. The processor is connected with the light source and used for controlling the light source to irradiate the card to be detected; the processor is also connected with the camera and used for acquiring a card image shot by the camera; the processor is also used for inputting the card image into a pre-trained card defect detection model so as to obtain the defects on the card; the card defect detection model is a target detection and segmentation algorithm based on deep learning; the training card defect detection model comprises: marking the defects on the training sample image of the defective card; inputting the training sample image into a card defect detection model to obtain the defect in the training sample image; and iteratively updating the weight of each layer of network according to the defect marked in the training sample image and the difference data between the defects obtained according to the card defect detection model so as to reduce the difference data.
Since the embodiment of the tooling portion and the embodiment of the method portion correspond to each other, please refer to the description of the embodiment of the method portion for the embodiment of the tooling portion, which is not repeated here.
The tool for detecting the card defects provided by the embodiment corresponds to the method, so that the tool has the same beneficial effects as the method.
In the above embodiments, the method for detecting a card defect is described in detail, and the present application also provides embodiments corresponding to the apparatus for detecting a card defect. It should be noted that the present application describes the embodiments of the apparatus portion from two perspectives, one from the perspective of the function module and the other from the perspective of the hardware.
Based on the angle of the functional module, this embodiment provides a detection apparatus for card defects, fig. 6 is a structural diagram of the detection apparatus for card defects provided in this embodiment of the application, as shown in fig. 6, the apparatus includes:
the control module 10 is used for controlling the light source to irradiate the card to be detected;
the acquisition module 11 is used for acquiring a card image shot by a camera;
the input module 12 is used for inputting the card image into a pre-trained card defect detection model so as to acquire the defects on the card; the card defect detection model is a target detection and segmentation algorithm based on deep learning; the training card defect detection model comprises: marking the defects on the training sample image of the defective card; inputting the training sample image into a card defect detection model to obtain the defect in the training sample image; and iteratively updating the weight of each layer of network according to the defect marked in the training sample image and the difference data between the defects obtained according to the card defect detection model so as to reduce the difference data.
Since the embodiments of the apparatus portion and the method portion correspond to each other, please refer to the description of the embodiments of the method portion for the embodiments of the apparatus portion, which is not repeated here.
As a preferred embodiment, the apparatus for detecting a card defect further includes: the splicing module is used for shearing and splicing the training sample images to obtain training sample images with uniformly distributed defects before inputting the training sample images into the card defect detection model to obtain the defects in the training sample images; inputting the training sample image into a card defect detection model to obtain the defects in the training sample image comprises the following steps: and inputting the training sample image with the defects distributed uniformly into the card defect detection model to obtain the defects in the training sample image.
The cutting module is used for respectively cutting the card image under the coaxial light source and the card image under the strip light source to obtain a plurality of sub-images before inputting the card image into a pre-trained card defect detection model to obtain the defects on the card;
the preprocessing module is used for performing traditional morphological preprocessing on the sub-images to obtain characteristic data; inputting the card image into a pre-trained card defect detection model to obtain the defects on the card comprises: and inputting the characteristic data into a pre-trained card defect detection model to acquire the defects on the card.
The card defect detection device provided by the embodiment corresponds to the method, so that the card defect detection device has the same beneficial effects as the method.
Based on the hardware angle, the present embodiment provides another apparatus for detecting a card defect, fig. 7 is a structural diagram of the apparatus for detecting a card defect according to another embodiment of the present application, and as shown in fig. 7, the apparatus for detecting a card defect includes: a memory 20 for storing a computer program;
a processor 21 for implementing the steps of the method for detecting a defect of a card as mentioned in the above embodiments when executing the computer program.
The card defect detection apparatus provided in this embodiment may include, but is not limited to, a smart phone, a tablet computer, a notebook computer, a desktop computer, or the like.
The processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like. The Processor 21 may be implemented in hardware using at least one of a Digital Signal Processor (DSP), a Field-Programmable Gate Array (FPGA), and a Programmable Logic Array (PLA). The processor 21 may also include a main processor and a coprocessor, where the main processor is a processor for Processing data in an awake state, and is also called a Central Processing Unit (CPU); a coprocessor is a low power processor for processing data in a standby state. In some embodiments, the processor 21 may be integrated with a GPU, which is responsible for rendering and drawing the content that the display screen needs to display. In some embodiments, the processor 21 may further include an Artificial Intelligence (AI) processor for processing computational operations related to machine learning.
The memory 20 may include one or more computer-readable storage media, which may be non-transitory. Memory 20 may also include high speed random access memory, as well as non-volatile memory, such as one or more magnetic disk storage devices, flash memory storage devices. In this embodiment, the memory 20 is at least used for storing the following computer program 201, wherein after the computer program is loaded and executed by the processor 21, the relevant steps of the method for detecting a card defect disclosed in any one of the foregoing embodiments can be implemented. In addition, the resources stored in the memory 20 may also include an operating system 202, data 203, and the like, and the storage manner may be a transient storage manner or a permanent storage manner. Operating system 202 may include, among others, Windows, Unix, Linux, and the like. The data 203 may include, but is not limited to, data related to a method of detecting a defect of a card, and the like.
In some embodiments, the apparatus for detecting card defects may further include a display 22, an input/output interface 23, a communication interface 24, a power supply 25, and a communication bus 26.
It will be appreciated by those skilled in the art that the configurations shown in the figures do not constitute a limitation of the means for detecting defects in the card and may include more or less components than those shown.
The device for detecting the card defects comprises a memory and a processor, wherein when the processor executes a program stored in the memory, the following method can be realized: a method for detecting card defects.
The card defect detection device provided by the embodiment corresponds to the method, so that the card defect detection device has the same beneficial effects as the method.
Finally, the application also provides a corresponding embodiment of the computer readable storage medium. The computer-readable storage medium has stored thereon a computer program which, when being executed by a processor, carries out the steps as set forth in the above-mentioned method embodiments.
It is to be understood that if the method in the above embodiments is implemented in the form of software functional units and sold or used as a stand-alone product, it can be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be substantially or partially implemented in the form of a software product, which is stored in a storage medium and performs all or part of the steps of the methods described in the embodiments of the present application, or all or part of the technical solution. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
The computer-readable storage medium provided by the embodiment corresponds to the method, and therefore has the same beneficial effects as the method.
The tool, the method, the device and the medium for detecting the card defects are described in detail above. The embodiments are described in a progressive manner in the specification, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description. It should be noted that, for those skilled in the art, it is possible to make several improvements and modifications to the present application without departing from the principle of the present application, and such improvements and modifications also fall within the scope of the claims of the present application.
It is further noted that, in the present specification, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the same element.
Claims (10)
1. The method for detecting the defects of the card is applied to a tool for detecting the defects of the card, which comprises a camera, a light source and a processor, and comprises the following steps:
controlling the light source to irradiate the card to be detected;
acquiring a card image shot by the camera;
inputting the card image into a pre-trained card defect detection model to obtain the defects on the card; the card defect detection model is a target detection and segmentation algorithm based on deep learning; training the card defect detection model comprises: marking the defects on the training sample image of the defective card; inputting the training sample image into the card defect detection model to obtain the defect in the training sample image; and iteratively updating the weight of each layer of network according to the difference data between the defects marked in the training sample image and the defects obtained according to the card defect detection model so as to reduce the difference data.
2. The method for detecting card defects according to claim 1, wherein before inputting the training sample image into the card defect detection model to obtain the defects in the training sample image, the method further comprises:
cutting and splicing the training sample images to obtain the training sample images with uniformly distributed defects; the step of inputting the training sample image into the card defect detection model to obtain the defects in the training sample image comprises:
and inputting the training sample image with uniformly distributed defects into the card defect detection model to obtain the defects in the training sample image.
3. The method for detecting card defects according to claim 2, wherein the card defect detection model is a Resnet 50-based feature extraction backbone network;
inputting the card image and the training sample image into the card defect detection model to obtain the defect on the card comprises:
inputting the card image and the training sample image into Resnet50 to extract a plurality of feature information of different scales;
and inputting the characteristic information into the FPN for multi-scale information fusion so as to output the position coordinates of the defects, the probability value of the defects and the semantic information of the defect regions.
4. The method of claim 3, wherein the light source comprises a coaxial light source and a strip light source, and the controlling the light source to illuminate the card to be detected comprises:
controlling the coaxial light source to irradiate from the right upper part of the card;
controlling the strip-shaped light source to irradiate from each direction above the card;
the obtaining of the card image shot by the camera includes:
and respectively acquiring the card image under the coaxial light source and the card image under the strip light source.
5. The method for detecting card defects according to claim 4, wherein before inputting the card image into a pre-trained card defect detection model to obtain the defects on the card, the method further comprises:
respectively cutting the card image under the coaxial light source and the card image under the strip-shaped light source to obtain a plurality of sub-images;
performing traditional morphological preprocessing on the sub-images to obtain characteristic data;
the inputting the card image into a pre-trained card defect detection model to obtain the defects on the card comprises:
and inputting the characteristic data into the card defect detection model trained in advance to obtain the defects on the card.
6. The method for detecting card defects according to claim 5, wherein the conventional morphological preprocessing of the sub-images to obtain feature data comprises:
carrying out gray scale transformation on the sub-image;
performing gray gradient detection on the sub-image after gray conversion by using a Canny edge detection method based on gray gradient to extract gray gradient characteristics of the card image under the irradiation of the coaxial light source and the strip light source; wherein the gray gradient feature comprises a shading feature of the card and a defect feature of the card, and the gray gradient feature is the feature data.
7. The utility model provides a detection frock of card defect which characterized in that includes: the device comprises a camera, a light source and a processor;
the processor is connected with the light source and used for controlling the light source to irradiate the card to be detected;
the processor is also connected with the camera and used for acquiring a card image shot by the camera; the processor is also used for inputting the card image into a pre-trained card defect detection model so as to obtain the defects on the card; the card defect detection model is a target detection and segmentation algorithm based on deep learning; training the card defect detection model comprises: marking the defects on the training sample image of the defective card; inputting the training sample image into the card defect detection model to obtain the defect in the training sample image; and iteratively updating the weight of each layer of network according to the difference data between the defects marked in the training sample image and the defects obtained according to the card defect detection model so as to reduce the difference data.
8. The utility model provides a detection apparatus for card defect, its characterized in that is applied to the detection frock that includes the card defect of camera, light source and treater, the device includes:
the control module is used for controlling the light source to irradiate the card to be detected;
the acquisition module is used for acquiring the card image shot by the camera;
the input module is used for inputting the card image into a pre-trained card defect detection model so as to obtain the defects on the card; the card defect detection model is a target detection and segmentation algorithm based on deep learning; training the card defect detection model comprises: marking the defects on the training sample image of the defective card; inputting the training sample image into the card defect detection model to obtain the defect in the training sample image; and iteratively updating the weight of each layer of network according to the difference data between the defects marked in the training sample image and the defects obtained according to the card defect detection model so as to reduce the difference data.
9. An apparatus for detecting defects in a card, comprising a memory for storing a computer program;
processor for implementing the steps of the method for detecting defects of a card according to any one of claims 1 to 6 when executing said computer program.
10. A computer-readable storage medium, characterized in that a computer program is stored thereon, which computer program, when being executed by a processor, carries out the steps of the method of detection of card defects according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210632046.XA CN114708266A (en) | 2022-06-07 | 2022-06-07 | Tool, method and device for detecting card defects and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210632046.XA CN114708266A (en) | 2022-06-07 | 2022-06-07 | Tool, method and device for detecting card defects and medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114708266A true CN114708266A (en) | 2022-07-05 |
Family
ID=82178037
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210632046.XA Pending CN114708266A (en) | 2022-06-07 | 2022-06-07 | Tool, method and device for detecting card defects and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114708266A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115601363A (en) * | 2022-12-14 | 2023-01-13 | 中建科技集团有限公司（Cn） | Small-target detection algorithm-based assembly type building product defect detection method |
Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106548159A (en) * | 2016-11-08 | 2017-03-29 | 中国科学院自动化研究所 | Reticulate pattern facial image recognition method and device based on full convolutional neural networks |
| CN109871895A (en) * | 2019-02-22 | 2019-06-11 | 北京百度网讯科技有限公司 | The defect inspection method and device of circuit board |
| CN110018178A (en) * | 2019-04-28 | 2019-07-16 | 华南理工大学 | A kind of mobile phone bend glass typical defect on-line measuring device and method |
| CN110473178A (en) * | 2019-07-30 | 2019-11-19 | 上海深视信息科技有限公司 | A kind of open defect detection method and system based on multiple light courcess fusion |
| CN110702696A (en) * | 2019-10-14 | 2020-01-17 | 珠海博明视觉科技有限公司 | Magnetic shoe surface defect detection device |
| CN110728657A (en) * | 2019-09-10 | 2020-01-24 | 江苏理工学院 | Annular bearing outer surface defect detection method based on deep learning |
| CN111062961A (en) * | 2019-12-30 | 2020-04-24 | 天津大学 | Contact lens edge defect detection method based on deep learning |
| CN111103292A (en) * | 2019-12-31 | 2020-05-05 | 深圳市智信精密仪器有限公司 | Mobile phone defect inspection visual device based on deep learning |
| CN111754513A (en) * | 2020-08-07 | 2020-10-09 | 腾讯科技（深圳）有限公司 | Product surface defect segmentation method, defect segmentation model learning method and device |
| CN112085745A (en) * | 2020-09-07 | 2020-12-15 | 福建农林大学 | Retinal vessel image segmentation method of multi-channel U-shaped full convolution neural network based on balanced sampling splicing |
| CN112465746A (en) * | 2020-11-02 | 2021-03-09 | 新疆天维无损检测有限公司 | Method for detecting small defects in radiographic film |
| CN112489037A (en) * | 2020-12-15 | 2021-03-12 | 科大讯飞华南人工智能研究院（广州）有限公司 | Defect detection and related model training method, electronic equipment and storage device |
| CN112748120A (en) * | 2020-12-29 | 2021-05-04 | 中国联合网络通信集团有限公司 | Defect detection system, defect detection method, defect detection device, defect detection equipment and storage medium |
| CN113096098A (en) * | 2021-04-14 | 2021-07-09 | 大连理工大学 | Casting appearance defect detection method based on deep learning |
| CN113642576A (en) * | 2021-08-24 | 2021-11-12 | 凌云光技术股份有限公司 | Method and device for generating training image set in target detection and semantic segmentation task |
-
2022
- 2022-06-07 CN CN202210632046.XA patent/CN114708266A/en active Pending
Patent Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106548159A (en) * | 2016-11-08 | 2017-03-29 | 中国科学院自动化研究所 | Reticulate pattern facial image recognition method and device based on full convolutional neural networks |
| CN109871895A (en) * | 2019-02-22 | 2019-06-11 | 北京百度网讯科技有限公司 | The defect inspection method and device of circuit board |
| CN110018178A (en) * | 2019-04-28 | 2019-07-16 | 华南理工大学 | A kind of mobile phone bend glass typical defect on-line measuring device and method |
| CN110473178A (en) * | 2019-07-30 | 2019-11-19 | 上海深视信息科技有限公司 | A kind of open defect detection method and system based on multiple light courcess fusion |
| CN110728657A (en) * | 2019-09-10 | 2020-01-24 | 江苏理工学院 | Annular bearing outer surface defect detection method based on deep learning |
| CN110702696A (en) * | 2019-10-14 | 2020-01-17 | 珠海博明视觉科技有限公司 | Magnetic shoe surface defect detection device |
| CN111062961A (en) * | 2019-12-30 | 2020-04-24 | 天津大学 | Contact lens edge defect detection method based on deep learning |
| CN111103292A (en) * | 2019-12-31 | 2020-05-05 | 深圳市智信精密仪器有限公司 | Mobile phone defect inspection visual device based on deep learning |
| CN111754513A (en) * | 2020-08-07 | 2020-10-09 | 腾讯科技（深圳）有限公司 | Product surface defect segmentation method, defect segmentation model learning method and device |
| CN112085745A (en) * | 2020-09-07 | 2020-12-15 | 福建农林大学 | Retinal vessel image segmentation method of multi-channel U-shaped full convolution neural network based on balanced sampling splicing |
| CN112465746A (en) * | 2020-11-02 | 2021-03-09 | 新疆天维无损检测有限公司 | Method for detecting small defects in radiographic film |
| CN112489037A (en) * | 2020-12-15 | 2021-03-12 | 科大讯飞华南人工智能研究院（广州）有限公司 | Defect detection and related model training method, electronic equipment and storage device |
| CN112748120A (en) * | 2020-12-29 | 2021-05-04 | 中国联合网络通信集团有限公司 | Defect detection system, defect detection method, defect detection device, defect detection equipment and storage medium |
| CN113096098A (en) * | 2021-04-14 | 2021-07-09 | 大连理工大学 | Casting appearance defect detection method based on deep learning |
| CN113642576A (en) * | 2021-08-24 | 2021-11-12 | 凌云光技术股份有限公司 | Method and device for generating training image set in target detection and semantic segmentation task |
Non-Patent Citations (1)
| Title |
|---|
| 方叶祥等: "金属表面缺陷检测的改进YOLOv3算法研究", 《机械科学与技术》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115601363A (en) * | 2022-12-14 | 2023-01-13 | 中建科技集团有限公司（Cn） | Small-target detection algorithm-based assembly type building product defect detection method |
| CN115601363B (en) * | 2022-12-14 | 2023-05-23 | 中建科技集团有限公司 | Assembly type building product defect detection method based on small target detection algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106875381B (en) | Mobile phone shell defect detection method based on deep learning | |
| CN110414507B (en) | License plate recognition method and device, computer equipment and storage medium | |
| CN111178355B (en) | Seal identification method, device and storage medium | |
| CN110751154B (en) | Complex environment multi-shape text detection method based on pixel-level segmentation | |
| CN108830857B (en) | Self-adaptive Chinese character copy label image binarization segmentation method | |
| CN111680690A (en) | Character recognition method and device | |
| CN112614062A (en) | Bacterial colony counting method and device and computer storage medium | |
| CN115131283A (en) | Defect detection and model training method, device, equipment and medium for target object | |
| CN110599453A (en) | Panel defect detection method and device based on image fusion and equipment terminal | |
| CN115205247A (en) | Method, device and equipment for detecting defects of battery pole piece and storage medium | |
| CN115588024B (en) | Complex industrial image edge extraction method and device based on artificial intelligence | |
| CN113591719A (en) | Method and device for detecting text with any shape in natural scene and training method | |
| CN117392042A (en) | Defect detection method, defect detection apparatus, and storage medium | |
| CN113505702A (en) | Pavement disease identification method and system based on double neural network optimization | |
| CN113033558A (en) | Text detection method and device for natural scene and storage medium | |
| CN114708266A (en) | Tool, method and device for detecting card defects and medium | |
| US20240127404A1 (en) | Image content extraction method and apparatus, terminal, and storage medium | |
| CN110751225A (en) | Image classification method, device and storage medium | |
| CN113486715A (en) | Image reproduction identification method, intelligent terminal and computer storage medium | |
| CN116091503B (en) | Method, device, equipment and medium for discriminating panel foreign matter defects | |
| CN112393880A (en) | Screen replacement detection method and device | |
| CN115578362A (en) | Defect detection method and device for electrode coating, electronic device and medium | |
| CN116798041A (en) | Image recognition method and device and electronic equipment | |
| CN116245882A (en) | Circuit board electronic element detection method and device and computer equipment | |
| CN114511046A (en) | Object recognition method and device |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220705 |