CN114429445B - A PCB defect detection and identification method based on MAIRNet - Google Patents

A PCB defect detection and identification method based on MAIRNet Download PDF

Info

Publication number
CN114429445B
CN114429445B CN202111318512.9A CN202111318512A CN114429445B CN 114429445 B CN114429445 B CN 114429445B CN 202111318512 A CN202111318512 A CN 202111318512A CN 114429445 B CN114429445 B CN 114429445B
Authority
CN
China
Prior art keywords
image
pcb
component
polarity
detected
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.)
Active
Application number
CN202111318512.9A
Other languages
Chinese (zh)
Other versions
CN114429445A (en
Inventor
谢非
章悦
张瑞
杨嘉乐
夏光圣
吴佳豪
郑鹏飞
张培彪
王慧敏
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.)
Nanjing Normal University
Original Assignee
Nanjing Normal University
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 Nanjing Normal University filed Critical Nanjing Normal University
Priority to CN202111318512.9A priority Critical patent/CN114429445B/en
Publication of CN114429445A publication Critical patent/CN114429445A/en
Application granted granted Critical
Publication of CN114429445B publication Critical patent/CN114429445B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • 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
    • 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
    • G06N3/048—Activation functions
    • 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
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/20—Image enhancement or restoration using local operators
    • G06T5/30—Erosion or dilatation, e.g. thinning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/90—Dynamic range modification of images or parts thereof
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
    • G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20036—Morphological image processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G06T2207/30141—Printed circuit board [PCB]
    • 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

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a method for detecting and identifying PCB defects based on MAIRNet, which comprises the steps of collecting PCB template images and images to be detected, constructing a data set, dividing the data set into test set images and training set images, detecting whether the situation of missing components exists, finally outputting missing component positioning information and marking the missing component positioning information in the images to be detected, outputting positioning conditions, color ring resistance types and positioning information of polar components, welding spot defect types and positioning information of the polar components, outputting the cut PCB polar component images to be detected and template images, constructing a PCB component polarity judging method, outputting component polarity plugging conditions, and summarizing and displaying the obtained information. The invention can detect and identify the common problems of component deficiency, component polarity connection error, welding spot defect and the like on the surface of the PCB, output the position information and the category information of the defect area, and detect and identify the color ring resistance category, thereby having remarkable advantages compared with a manual detection mode.

Description

MAIRNet-based PCB defect detection and identification method
Technical Field
The invention relates to the technical field of machine vision and deep learning, in particular to a MAIRNet-based PCB defect detection and identification method.
Background
With the rapid development of machine vision and artificial intelligence technology, deep learning, neural networks and other technologies are gradually applied to the process of industrial detection. The PCB has the characteristics of high density, light weight, high integration level and the like, so that the PCB defect detection process also faces the problems of diversification of target components, small size of detection objects and difficult identification.
The main modes of PCB detection are mainly divided into three types of manual visual inspection, contact detection and non-contact detection. The problem of low efficiency exists in manual visual inspection, contact detection is extremely vulnerable to damage to components on the surface of the PCB, along with the development of an image processing technology and a visual sensor, visual-based PCB detection is gradually applied to actual production, but a conventional visual detection mode cannot simultaneously identify and detect various components, and still needs a relatively complex manual operation process, so that the method has limitations.
The detection and identification method based on the traditional image processing technology does not need complex and expensive hardware, but has certain limitation, has lower detection and identification precision on the complex background of PCBs with various target types, and can not accurately judge the types of components and defects. The current target detection and recognition technology based on deep learning does not have a specific neural network model built for PCB defects, and the detection and recognition technology for the common defects of the PCB, such as welding spot continuous welding, cold joint, component deficiency, polarity reverse connection and the like, is relatively not mature enough, and is difficult to detect and recognize the defects of the PCB in a targeted manner, and meanwhile, the deep neural network is complicated in model, occupies increased computing resources and is difficult to be transplanted to an embedded development board to be applied to industrial production detection due to the fact that the network structure is complex and the number of layers is deepened.
Therefore, a new solution is needed to solve these problems.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, provides a method for detecting and identifying PCB defects based on MAIRNet, adopts a detection method based on deep learning and machine vision to replace the traditional detection mode, has relatively low cost in an actual application scene, does not need an expensive automatic optical detection (AOI) system, greatly improves the efficiency compared with manual detection and contact detection, is not easy to cause damage to PCB surface components, is built based on a neural network model, has lower complexity, is convenient to be transplanted to an embedded development board to be applied to actual industrial production detection, can realize the detection function of various defect targets in single detection operation, improves the production efficiency and reliability, meets the production requirements of high-performance and high-complexity products, can greatly improve the overall detection level of the PCB on a plate production assembly line of intelligent power grid equipment, and has important practical application value.
The invention provides a MAIRNet-based PCB defect detection and identification method for achieving the purposes, which comprises the following steps:
s1, respectively collecting template images and images to be detected on the front surface and the back surface of a PCB, constructing a data set, dividing the data set into a test set image and a training set image, marking the training set image with component categories and defect conditions, and generating a corresponding label file;
S2, inputting the images to be detected and the template images in the training set into a component existence detection module, detecting whether the condition of component missing exists, and finally outputting the missing component positioning information and marking in the images to be detected;
S3, inputting the PCB image in the testing set into a built and trained MAIRNet-based PCB defect detection and identification model to perform detection and identification operation, and outputting positioning conditions, color ring resistance types and positioning information, welding spot defect types and positioning information for the polar components;
S4, cutting and screening the image to be detected and the template image by utilizing the positioning condition of the polar component output in the step S3, outputting the cut image to be detected of the PCB polar component and the cut template image, constructing a method for judging the polarity of the PCB component, operating by utilizing the component existence detection module, independently comparing the polar component areas of the image to be detected and the template image by dividing the polar component existence area, inputting the comparison condition into the polarity detection module for judging, and finally outputting the polarity plugging condition of the component;
and S5, summarizing and displaying the obtained positioning information of the missing component, the labeling condition on the PCB image, the polarity plugging condition of the component, the resistance type and positioning information of the color ring, the defect type and positioning information of the welding spot.
Further, the detecting method of the component existence detecting module in the step S2 is as follows:
preprocessing an input image to be detected, performing image registration operation on the image to be detected and an original image by using ORB characteristics, and ensuring that the relative characteristics contained in the image to be detected and the template image are as close as possible;
D2, performing image graying operation on the image to be detected and the template image after image registration to generate a gray level image to be detected and a template gray level image;
d3, carrying out image difference on the gray level image to be detected and the template gray level image;
D4, performing image binarization processing on the differential gray level image obtained in the step D3 to obtain a differential binary level image, performing image morphology processing, and further reducing noise and influence on the positioning of the missing area of the component caused by the PCB background environment by adopting corrosion expansion operation;
D5, carrying out pixel screening on the differential binary image subjected to the image morphology processing in the step D4, and screening out the maximum connected region in the image;
and D6, marking the rectangular frame of the largest connected region screened in the step D5, mapping the rectangular frame of the missing region of the marked component into the input image to be detected, and further outputting the locating information of the missing component and realizing marking.
Further, the method for building the PCB defect detection and identification model based on MAIRNet in the step S3 includes the following steps:
A1, constructing a feature map based on MAIRNet for extracting the features of the PCB image in the training set, outputting feature maps with various scales, wherein MAIRNet is a multidimensional attention mechanism inverse residual error network (Multidimensional Attention Inverse Residual Network);
A2, detecting and identifying the targets of the PCB components on the feature graphs with various scales, wherein the category information corresponds to the position information targets of the bounding boxes of the targets, and the category information comprises specific categories and defect conditions of the target components.
Further, the step A1 specifically includes the following steps:
MAIRNet is composed of an inverse residual error network and a multidimensional attention module, and a multiscale feature fusion mechanism is used in the network construction process;
b2, adding a multi-dimensional attention module on the basis of an inverse residual error network, and optimizing each inverse residual error module in the inverse residual error network by using the multi-dimensional attention module to obtain an inverse residual error module optimized by the multi-dimensional attention module;
and B3, constructing MAIRNet according to a multi-scale feature fusion mechanism.
Further, the step A2 specifically includes the following steps:
Inputting the 5-layer feature images output by the multidimensional attention-enhancing neural network into a PCB component target detection and identification module to respectively perform three types of operations of polar component positioning, chromatic circle resistance identification and welding spot detection, wherein the 5-layer feature images in each type of operation pass through 4 groups of convolutional neural networks, each group of convolutional neural networks comprises convolution with a convolution kernel of 3 multiplied by 3 and a step length of 1, and the group normalization operation is performed, and finally, a ReLU activation function is used for activation;
In each type of operation, carrying out bounding box regression on the feature images output through the convolution of 4 groups of convolution neural networks, determining the area where the component targets are located, and using a generalized cross-over comparison function GIoU (Generalized Intersection over Union) as a bounding box regression loss function;
and C3, classifying the feature images output through the convolution of the 4 groups of convolution neural networks in each type of operation, determining the types and defect types of the targets of the components and the devices, and using a focus loss function FocalLoss as a classification loss function.
Further, the specific calculation process of the generalized cross ratio function GIoU in the step C2 is as follows:
LGIoU=1-GIoU (1)
Where A represents a prediction bounding box, B represents a real bounding box, C represents a minimum closure rectangular region of the prediction bounding box and the real bounding box, L GIoU represents a bounding box regression loss function, ioU represents an intersection ratio function between the prediction bounding box and the real bounding box, and GIoU represents a generalized intersection ratio function between the prediction bounding box and the real bounding box.
Further, the specific calculation process of the focus loss function FocalLoss in the step C3 is:
FL(p,y)=-y(1-p)γlog(p)-(1-y)pγlog(1-p) (4)
Where p is a predicted value predicted as a certain type label, and between 0 and 1, y is an actual type label, and is 0 or 1, and γ is a constant set by man.
Further, the method for discriminating the polarity of the PCB component in the step S4 includes the following steps:
E1, preprocessing an input PCB polarity component to-be-detected image, performing image registration operation on the to-be-detected image and an original image by using ORB characteristics, and ensuring that the relative characteristics contained in the PCB polarity component to-be-detected image and the PCB polarity component template image are as close as possible;
E2, performing image graying operation on the PCB polarity component to-be-detected image and the PCB polarity component template image obtained after image registration to generate a PCB polarity component to-be-detected gray level image and a PCB polarity component template gray level image;
e3, carrying out image difference on the gray level image to be detected of the PCB polarity component and the gray level image of the PCB polarity component template;
e4, performing image binarization processing on the PCB polarity component differential gray level image obtained in the step E3 to obtain a PCB polarity component differential binary level image, performing image morphology processing, and adopting corrosion expansion operation to further reduce noise and influence of the PCB background environment on the positioning of the polarity component differential region;
E5, carrying out pixel screening on the differential binary image subjected to the image morphology processing in the step E4, and screening out the maximum communication area in the differential binary image of the PCB polar component;
and E6, inputting the maximum communication area screened in the step E5 into a polarity detection module to further judge, setting a threshold S t, if the maximum communication area is larger than the set threshold S t, indicating that the polarity is wrong, and if the maximum communication area is not larger than the set threshold S t, indicating that the polarity is correct.
Further, the main body of MAIRNet in the step B1 is an inverse residual network constructed by 17 inverse residual modules, the head end of each inverse residual module is 1×1 convolution, the 1×1 convolution is used for expanding the feature matrix channel, the feature quantity is enriched, the middle part of the inverse residual module is composed of a convolution kernel which is 3×3 and is used for generating a feature matrix consistent with the input feature matrix channel quantity, so that the parameter quantity and the operation cost are reduced, and the expression capability of the network is finally enhanced by using a ReLU6 activation function;
The specific operation of the multidimensional attention module in the step B2 is that the characteristic matrix of each inverse residual module is divided into two channels for carrying out average pooling operation, the characteristic matrices of the two channels after average pooling are respectively connected and convolved by 1 multiplied by 1, the Batch Norm normalization operation is used, the activated characteristic matrix is activated by Swish activation functions, the activated characteristic matrix is decomposed into the characteristic matrices of the two channels again, the characteristic matrices of the two channels are activated by 1 multiplied by 1 convolution and Sigmoid activation functions respectively, the obtained characteristic matrices of the two channels are combined with the original characteristic matrix output by the inverse residual module, so that a multidimensional attention-enhanced characteristic diagram is obtained, and the enhanced characteristic diagram can more effectively express information in an original image;
The step B3 specifically comprises dividing 17 multi-dimensional attention module optimized inverse residual modules into 7 layers of networks, wherein each layer comprises 1,2,3,4,3,3,1 multi-dimensional attention optimized inverse residual modules in sequence, respectively defining characteristic graphs processed by the 2 nd layer, the 3 rd layer, the 5 th layer and the 7 th layer of networks as C 2、C3、C5、C7 to carry out 1X 1 convolution to obtain F 2、F3、F5、F7, carrying out step length 2 convolution on F 7 again to obtain F 7 ', fusing characteristic graphs with different scales, and finally fusing and outputting 5 layers of characteristic graphs F 2、F3、F5、F7、F7'.
Further, in the step S5, a user software system interface is set up for summary display.
Compared with the prior art, the invention has the following advantages:
1. According to the invention, the PCB image in industrial production is collected, the image to be detected and the template image are marked, a data set is constructed, the image is collected through the industrial camera, the software running environment can be detected and identified after configuration, expensive detection equipment and fixtures are not needed, meanwhile, the damage to the PCB components caused by contact detection is avoided, the cost is lower, less manual auxiliary work is needed, the implementation mode is quick and convenient, and the operation is easy.
2. The invention adopts the neural network based on multidimensional attention enhancement to detect, and has better detection and identification effects on defects of small targets such as welding spots and the like.
3. The invention can detect and identify various common PCB defects at one time without complex manual operation and repeated detection, and simultaneously constructs the interactive interface to output defect detection and identification results, and has the advantages of high identification precision, high detection speed, simple and convenient operation and visual display result.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a diagram of a PCB defect detection and identification model based on MAIRNet in the present invention;
FIG. 3 is a block diagram of the inverse residual module of the multidimensional attention module optimization used in the present invention;
FIG. 4 is a schematic diagram of a generalized cross-ratio function GIoU;
FIG. 5 is a graph of detection and identification results of common defects of a PCB based on MAIRNet;
FIG. 6 is a diagram of detecting missing components from the component presence detection module of the present invention;
FIG. 7 is a diagram showing the result of judging the polarity of the component output by the polarity detection module;
FIG. 8 is a schematic diagram of a user software system interface constructed in accordance with the present invention.
Detailed Description
The present application is further illustrated in the accompanying drawings and detailed description which are to be understood as being merely illustrative of the application and not limiting of its scope, and various modifications of the application, which are equivalent to those skilled in the art upon reading the application, will fall within the scope of the application as defined in the appended claims.
The invention provides a PCB defect detection and identification method based on a multidimensional attention-enhancing neural network (MAIRNet), which is shown in figure 1 and comprises the following steps:
S1, collecting images to be detected on the front side and the back side of a PCB and template images, wherein the images to be detected have the conditions of component missing, component polarity connection error, chromatic circle resistance connection error and welding spot defect. Dividing a test set image and a training set image according to the ratio of 3:7, marking the component category and the defect condition of the training set image, and generating a corresponding label file;
S2, constructing a MAIRNet-based PCB defect detection and identification model;
and S3, inputting the images to be detected and the template image in the training set into a component existence detection module, detecting whether the condition of component deletion exists, and finally outputting the component deletion positioning information and marking in the images to be detected.
S4, inputting the PCB image in the test set into a trained MAIRNet-based PCB defect detection and identification model for detection and identification operation, and outputting positioning conditions, color ring resistance types and positioning information, welding spot defect types and positioning information for the polar components;
And S5, cutting and screening the image to be detected and the template image by utilizing the positioning condition of the polar component output in the step S4, and outputting the cut PCB polar component image to be detected and the template image. Constructing a polarity judging method of the PCB component, utilizing the operation in the component existence detecting module in the step S3, independently comparing the polar component areas of the image to be detected and the template image by dividing the polar component existence areas, inputting the comparison condition into the polarity detecting module for judging, and finally outputting the polarity plugging condition of the component;
And S6, constructing a user software system interface, and summarizing and displaying the missing component positioning information output in the step S3 and the labeling condition, the component polarity plugging condition, the color ring resistance type and positioning information, the welding spot defect type and the positioning information on the PCB image in the user software system interface.
The specific process of step S2 in this embodiment is as follows:
A1, constructing a feature map based on MAIRNet for extracting the features of the PCB image in the training set, outputting feature maps with various scales, wherein MAIRNet is a multidimensional attention mechanism inverse residual error network (Multidimensional Attention Inverse Residual Network);
a2, carrying out target detection and identification on the PCB components of the feature graphs with various different scales, wherein the category information corresponds to the bounding box position information target of the bounding box where the target is located, and the category information comprises specific categories and defect conditions of the target components;
the specific process of step A1 in this embodiment is as follows:
B1: MAIRNet is composed of an inverse residual error network and a multidimensional attention module, and a multiscale feature fusion mechanism is used in the network construction process. The main body of MAIRNet is an inverse residual network constructed by 17 inverse residual modules, the head end of each inverse residual module is 1 multiplied by 1 convolution, the 1 multiplied by 1 convolution is used for expanding characteristic matrix channels and enriching the characteristic quantity, the middle part of the inverse residual module is composed of a convolution kernel which is 3 multiplied by 3 and is used for generating a characteristic matrix consistent with the input characteristic matrix channels, so that the parameter quantity and the operation cost are reduced, and the expression capacity of the network is finally enhanced by using a ReLU6 activation function;
And B2, adding a multi-dimensional attention module on the basis of the inverse residual error network, and optimizing each inverse residual error module in the inverse residual error network by using the multi-dimensional attention module to obtain an inverse residual error module optimized by the multi-dimensional attention module. The specific operation of the multidimensional attention module is shown as follows, the characteristic matrix of each inverse residual module is divided into two channels for carrying out average pooling operation, the characteristic matrices of the two channels after average pooling are respectively connected and subjected to 1X 1 convolution, the Batch Norm normalization operation is used for activating through Swish activation functions, the activated characteristic matrix is decomposed into the characteristic matrices of the two channels again, the characteristic matrices of the two channels are respectively activated through 1X 1 convolution and Sigmoid activation functions, the obtained characteristic matrices of the two channels are combined with the original characteristic matrix output by the inverse residual module, so that a multidimensional attention-enhanced characteristic diagram is obtained, and the enhanced characteristic diagram can more effectively express information in the original image;
and B3, constructing MAIRNet according to a multi-scale feature fusion mechanism. The 17 multi-dimensional attention module optimized inverse residual modules are divided into 7 layers of networks, wherein each layer from 1 st layer to 7 th layer sequentially comprises 1,2,3,4,3,3,1 multi-dimensional attention optimized inverse residual modules, feature images processed by the 2 nd layer, 3 rd layer, 5 th layer and 7 th layer networks are respectively defined as C 2、C3、C5、C7 to be subjected to 1X 1 convolution to obtain F 2、F3、F5、F7, F 7 is subjected to convolution with a step length of 2 again to obtain F 7 ', feature images with different scales are fused, and finally 5 layers of feature images F 2、F3、F5、F7、F7' are fused and output.
The specific process of step A2 in this embodiment is as follows:
And C1, inputting the 5-layer feature images output by the multidimensional attention-enhancing neural network into a target detection and identification module of the PCB component to respectively perform three types of operations of polar component positioning, chromatic circle resistance identification and welding spot detection, wherein the 5-layer feature images in each type of operation pass through 4 groups of convolutional neural networks. Each group of convolution neural network comprises convolution with convolution kernel of 3 multiplied by 3 and step length of 1, and group normalization operation is carried out, and a ReLU activation function is finally used for activation;
In each type of operation, carrying out bounding box regression on the feature images output through the convolution of 4 groups of convolution neural networks, determining the area where the component targets are located, and using a generalized cross-over comparison function GIoU (Generalized Intersection over Union) as a bounding box regression loss function;
And C3, classifying the feature images output through the convolution of the 4 groups of convolution neural networks in each type of operation, determining the types and defect types of the targets of the components and the devices, and using a Focal Loss function Focal Loss as a classification Loss function.
It should be noted that, the schematic diagram of GIoU functions in step C2 is shown in fig. 4, and it can be seen specifically that:
LGIoU=1-GIoU (1)
Where A represents a prediction bounding box, B represents a real bounding box, C represents a minimum closure rectangular region of the prediction bounding box and the real bounding box, L GIoU represents a bounding box regression loss function, ioU represents an intersection ratio function between the prediction bounding box and the real bounding box, and GIoU represents a generalized intersection ratio function between the prediction bounding box and the real bounding box.
It should be noted that, in the step C3, the calculation process of the focus loss function FocalLoss is specifically as follows:
FL(p,y)=-y(1-p)γlog(p)-(1-y)pγlog(1-p) (4)
Where p is a predicted value predicted as a certain type label, and between 0 and 1, y is an actual type label, and is 0 or 1, and γ is a constant set by man.
The specific process of step S3 in this embodiment is as follows:
preprocessing an input image to be detected, performing image registration operation on the image to be detected and an original image by using ORB characteristics, and ensuring that the relative characteristics contained in the image to be detected and the template image are as close as possible;
D2, performing image graying operation on the image to be detected and the template image after image registration to generate a gray level image to be detected and a template gray level image;
d3, carrying out image difference on the gray level image to be detected and the template gray level image;
d4, performing image binarization processing on the differential gray level image obtained in the step D3 to obtain a differential binary level image, performing image morphology processing, and adopting corrosion expansion operation to further reduce noise and influence on the positioning of the missing region of the component caused by the PCB background environment;
D5, carrying out pixel screening on the differential binary image subjected to the image morphology processing in the step D4, and screening out the maximum connected region in the image;
and D6, marking the rectangular frame of the largest connected region screened in the step D5, mapping the rectangular frame of the missing region of the marked component into the input image to be detected, and further outputting the locating information of the missing component and realizing marking.
The specific process of step S5 in this embodiment is as follows:
E1, preprocessing an input PCB polarity component to-be-detected image, performing image registration operation on the to-be-detected image and an original image by using ORB characteristics, and ensuring that the relative characteristics contained in the PCB polarity component to-be-detected image and the PCB polarity component template image are as close as possible;
E2, performing image graying operation on the PCB polarity component to-be-detected image and the PCB polarity component template image obtained after image registration to generate a PCB polarity component to-be-detected gray level image and a PCB polarity component template gray level image;
e3, carrying out image difference on the gray level image to be detected of the PCB polarity component and the gray level image of the PCB polarity component template;
E4, performing image binarization processing on the PCB polarity component differential gray level image obtained in the step F3 to obtain a PCB polarity component differential binary level image, performing image morphology processing, and adopting corrosion expansion operation to further reduce noise and influence of the PCB background environment on the positioning of the polarity component differential area;
E5, performing pixel screening on the differential binary image subjected to the image morphology processing in the step F4, and screening out the maximum communication area in the differential binary image of the PCB polar component;
And E6, inputting the maximum communication area screened in the step F5 into a polarity detection module to further judge, setting a threshold S t, if the maximum communication area is larger than the set threshold S t, indicating that the polarity is wrong, and if the maximum communication area is not larger than the set threshold S t, indicating that the polarity is correct.
In this embodiment, in order to verify the actual effect of the method of the present invention, the actual detection and identification conditions of the PCB defect detection and identification model based on MAIRNet on the resistor and the solder joint are shown in fig. 5 (a) and fig. b), respectively, it can be seen that the type of the resistor target and the region where the solder joint has a defect can be clearly and efficiently identified in fig. 5 (a) and fig. b), the result of the component presence detection module on the common component is shown in fig. 6, the component presence detection module can accurately locate the region where the component is missing, the result of the judgment on the plugging condition of the polar component is shown in fig. 7, and finally the judgment result is output on the interface of the user software system, and the result of the detection and identification of fig. 5, fig. 6 and fig. 7 can verify the actual effect of the method of the present invention, and the interface of the user software system is shown in fig. 8.
The invention can detect and identify the common problems of missing components, wrong polarity connection of components, welding spot defects and the like on the surface of the PCB, output the position information and the category information of the defect area, detect and identify the color ring resistance category, and has the detection precision up to 95 percent, and the average detection and identification speed of a single image is not more than 2 seconds, thus having remarkable advantages compared with the manual detection mode.

Claims (7)

1.一种基于MAIRNet的PCB缺陷检测与识别方法，其特征在于，包括如下步骤：1. A PCB defect detection and identification method based on MAIRNet, characterized in that it includes the following steps: S1：分别采集PCB正反两面的模板图像与待检测图像，构建数据集，并且划分成测试集图像和训练集图像，对训练集图像进行元器件类别以及缺陷情况的标注，生成对应的标签文件；S1: Collect template images and images to be tested on both sides of the PCB respectively, build a data set, and divide it into test set images and training set images. Label the training set images with component categories and defects, and generate corresponding label files. S2：将训练集中的待检测图像和模板图像全部输入到元器件存在性检测模块，检测是否存在元器件缺失的情况，最终输出缺失元器件定位信息并在待检测图像中进行标注；S2: Input all the images to be detected and the template images in the training set into the component existence detection module to detect whether there are any missing components, and finally output the location information of the missing components and mark them in the images to be detected; S3：将测试集中的PCB图像输入搭建和训练好的基于MAIRNet的PCB缺陷检测与识别模型进行检测和识别操作，输出对于极性元器件的定位情况、色环电阻种类及定位信息和焊点缺陷类别及定位信息；S3: Input the PCB images in the test set into the built and trained MAIRNet-based PCB defect detection and recognition model for detection and recognition operations, and output the positioning of polar components, color ring resistor types and positioning information, and solder joint defect types and positioning information; S4：利用步骤S3中输出的极性元器件定位情况对待检测图像与模板图像进行裁剪筛选，输出裁剪后的PCB极性元器件待检测图像与模板图像，构建一种PCB元器件极性判别方法，利用元器件存在性检测模块进行操作，通过对极性元器件存在区域进行分割，单独对待检测图像与模板图像的极性元器件区域进行对比，将对比情况输入到极性检测模块中进行判别，最终输出元器件极性接插情况；S4: using the polarity component positioning situation output in step S3 to crop and screen the image to be detected and the template image, output the cropped PCB polarity component image to be detected and the template image, construct a PCB component polarity discrimination method, use the component existence detection module to operate, by segmenting the polarity component existence area, separately compare the polarity component area of the image to be detected and the template image, input the comparison situation into the polarity detection module for discrimination, and finally output the component polarity connection situation; S5：对获取的缺失元器件定位信息与PCB图像上的标注情况、元器件极性接插情况、色环电阻种类及定位信息、焊点缺陷类别及定位信息进行汇总显示；S5: The acquired missing component location information and the marking conditions on the PCB image, component polarity connection conditions, color ring resistor types and location information, solder joint defect types and location information are summarized and displayed; 所述步骤S3中基于MAIRNet的PCB缺陷检测与识别模型的搭建方法包括如下步骤：The method for building a PCB defect detection and identification model based on MAIRNet in step S3 includes the following steps: A1：构建基于MAIRNet用于提取训练集中PCB图像的特征，输出多种不同尺度的特征图；A1: Build a MAIRNet-based framework to extract features of PCB images in the training set and output feature maps of various scales; A2：对多种不同尺度的特征图进行PCB元器件目标检测与识别，目标所处的边界框boundingbox位置信息目标对应的类别信息，类别信息包含目标元器件的具体类别以及缺陷情况；A2: PCB component target detection and recognition are performed on feature maps of various scales. The bounding box location information of the target contains the category information corresponding to the target. The category information includes the specific category of the target component and the defect situation. 所述步骤A1具体包括如下步骤：The step A1 specifically includes the following steps: B1：MAIRNet由逆残差网络、多维注意力模块所组成，并且在网络搭建过程中使用了多尺度特征融合机制；B1: MAIRNet consists of an inverse residual network and a multi-dimensional attention module, and uses a multi-scale feature fusion mechanism in the network construction process; B2：在逆残差网络的基础上增加了一种多维注意力模块，使用多维注意力模块对于逆残差网络中的每一个逆残差模块进行优化，得到多维注意力模块优化的逆残差模块；B2: A multidimensional attention module is added on the basis of the inverse residual network. The multidimensional attention module is used to optimize each inverse residual module in the inverse residual network to obtain an inverse residual module optimized by the multidimensional attention module. B3：按照多尺度特征融合机制搭建MAIRNet；B3: Build MAIRNet according to the multi-scale feature fusion mechanism; 所述步骤A2具体包括如下步骤：The step A2 specifically includes the following steps: C1：将经过多维注意力增强神经网络输出的5层特征图输入到PCB元器件目标检测与识别模块中分别进行极性元器件定位、色环电阻识别、焊点检测三类操作，每一类操作中5层特征图均通过了4组卷积神经网络；每组卷积神经网络包含卷积核为3×3、步长为1的卷积，群组归一化操作，最终使用ReLU激活函数进行激活；C1: The 5-layer feature map output by the multi-dimensional attention enhanced neural network is input into the PCB component target detection and recognition module to perform three types of operations, namely polar component positioning, color ring resistor recognition, and solder joint detection. In each type of operation, the 5-layer feature map passes through 4 groups of convolutional neural networks; each group of convolutional neural networks includes a convolution with a convolution kernel of 3×3 and a stride of 1, a group normalization operation, and finally an activation using the ReLU activation function; C2：每一类操作中，对通过4组卷积神经网络卷积输出的特征图进行边界框回归，确定元器件目标所在区域，使用广义交并比函数GIoU作为边界框回归损失函数；C2: In each type of operation, bounding box regression is performed on the feature map output by the convolution of the four groups of convolutional neural networks to determine the area where the component target is located, and the generalized intersection-over-union function GIoU is used as the bounding box regression loss function; C3：每一类操作中，对通过4组卷积神经网络卷积输出的特征图进行分类，确定元器件目标的种类及缺陷类别，使用焦点损失函数Focal Loss作为分类损失函数。C3: In each type of operation, the feature maps output by the convolution of four groups of convolutional neural networks are classified to determine the type of component target and the defect category, and the focal loss function is used as the classification loss function. 2.根据权利要求1所述的一种基于MAIRNet的PCB缺陷检测与识别方法，其特征在于，所述步骤S2中元器件存在性检测模块的检测方法为：2. According to a PCB defect detection and identification method based on MAIRNet according to claim 1, it is characterized in that the detection method of the component existence detection module in step S2 is: D1：对输入的待检测图像进行预处理，使用ORB特征对待检测图像与原始图像进行图像配准操作；D1: Preprocess the input image to be detected and use ORB features to perform image registration between the image to be detected and the original image; D2：对于图像配准后的待检测图像与模板图像进行图像灰度化操作，生成待检测灰度图与模板灰度图；D2: Perform image grayscale operation on the image to be detected and the template image after image registration to generate a grayscale image to be detected and a grayscale image of the template; D3：对于待检测灰度图与模板灰度图进行图像差分；D3: Perform image difference between the grayscale image to be detected and the template grayscale image; D4：对于步骤D3中得到的差分灰度图进行图像二值化处理，得到差分二值图，并进行图像形态学处理，采用腐蚀膨胀操作进一步减少PCB背景环境对元器件缺失区域定位带来的噪声及影响；D4: performing image binarization processing on the differential grayscale image obtained in step D3 to obtain a differential binary image, and performing image morphological processing, using corrosion and expansion operations to further reduce the noise and influence of the PCB background environment on the positioning of the missing component area; D5：对于步骤D4中进行图像形态学处理后的差分二值图进行像素筛选，筛选出图像中的最大连通区域；D5: Filter pixels of the differential binary image after the image morphological processing in step D4 to filter out the largest connected area in the image; D6：对于步骤D5中筛选出的最大连通区域进行矩形框标注，并将标注元器件缺失区域的矩形框映射到输入的待检测图像中，进而输出缺失元器件定位信息并实现标注。D6: Mark the largest connected area selected in step D5 with a rectangular frame, and map the rectangular frame of the marked component missing area to the input image to be detected, thereby outputting the missing component positioning information and achieving marking. 3.根据权利要求1所述的一种基于MAIRNet的PCB缺陷检测与识别方法，其特征在于，所述步骤C2中广义交并比函数GIoU的具体计算过程为：3. According to a MAIRNet-based PCB defect detection and identification method according to claim 1, it is characterized in that the specific calculation process of the generalized intersection-over-union function GIoU in step C2 is: LGIoU＝1-GIoU (1)L GIoU = 1-GIoU (1) 式中，A表示预测边界框，B表示真实边界框，C表示预测边界框和真实边界框的最小闭包矩形区域，LGIoU表示边界框回归损失函数，IoU表示预测边界框与真实边界框之间的交并比函数，GIoU表示预测边界框与真实边界框之间的广义交并比函数。Where A represents the predicted bounding box, B represents the true bounding box, C represents the minimum enclosing rectangular area of the predicted bounding box and the true bounding box, L GIoU represents the bounding box regression loss function, IoU represents the intersection-over-union ratio function between the predicted bounding box and the true bounding box, and GIoU represents the generalized intersection-over-union ratio function between the predicted bounding box and the true bounding box. 4.根据权利要求1所述的一种基于MAIRNet的PCB缺陷检测与识别方法，其特征在于，所述步骤C3中焦点损失函数Focal Loss的具体计算过程为：4. A PCB defect detection and identification method based on MAIRNet according to claim 1, characterized in that the specific calculation process of the focal loss function Focal Loss in step C3 is: FL(p,y)＝-y(1-p)γlog(p)-(1-y)pγlog(1-p) (4)FL(p,y)＝-y(1-p) γ log(p)-(1-y)p γ log(1-p) (4) 式中，p表示预测为某一类别标签的预测值，在0到1之间，y表示实际的类别标签，为0或是1，γ表示人为设定的常数。In the formula, p represents the predicted value of a certain category label, which is between 0 and 1, y represents the actual category label, which is 0 or 1, and γ represents an artificially set constant. 5.根据权利要求1所述的一种基于MAIRNet的PCB缺陷检测与识别方法，其特征在于，所述步骤S4中PCB元器件极性判别方法包括如下步骤：5. According to a PCB defect detection and identification method based on MAIRNet according to claim 1, it is characterized in that the PCB component polarity discrimination method in step S4 comprises the following steps: E1：对输入的PCB极性元器件待检测图像进行预处理，使用ORB特征对待检测图像与原始图像进行图像配准操作；E1: Preprocess the input image of the PCB polarity components to be inspected, and use the ORB feature to perform image registration between the image to be inspected and the original image; E2：对于图像配准后得到的PCB极性元器件待检测图像与PCB极性元器件模板图像进行图像灰度化操作，生成PCB极性元器件待检测灰度图与PCB极性元器件模板灰度图；E2: performing image grayscale operation on the PCB polarity component to be detected image and the PCB polarity component template image obtained after image registration, and generating a grayscale image of the PCB polarity component to be detected and a grayscale image of the PCB polarity component template; E3：对于PCB极性元器件待检测灰度图与PCB极性元器件模板灰度图进行图像差分；E3: Perform image difference between the grayscale image of the PCB polarity component to be detected and the grayscale image of the PCB polarity component template; E4：对于步骤E3中得到的PCB极性元器件差分灰度图进行图像二值化处理，得到PCB极性元器件差分二值图，并进行图像形态学处理，采用腐蚀膨胀操作进一步较少PCB背景环境对极性元器件差异区域的定位带来的噪声及影响；E4: performing image binarization processing on the PCB polarity component differential grayscale image obtained in step E3 to obtain a PCB polarity component differential binary image, and performing image morphology processing, using corrosion and dilation operations to further reduce the noise and influence of the PCB background environment on the positioning of the polarity component difference area; E5：对于步骤E4中进行图像形态学处理后的差分二值图进行像素筛选，筛选出PCB极性元器件差分二值图中的最大连通区域；E5: Pixel screening is performed on the differential binary image after image morphology processing in step E4, and the maximum connected area in the differential binary image of PCB polarity components is screened out; E6：将步骤E5中筛选出的最大连通区域输入到极性检测模块中进行进一步判别，设定阈值St，若最大连通区域大于设定阈值St则说明可能存在极性错误的情况，若最大连通区域不超过设定阈值St则说明极性正确。E6: Input the maximum connected area screened in step E5 into the polarity detection module for further determination, set a threshold S t , if the maximum connected area is greater than the set threshold S t, it indicates that there may be a polarity error, if the maximum connected area does not exceed the set threshold S t, it indicates that the polarity is correct. 6.根据权利要求1所述的一种基于MAIRNet的PCB缺陷检测与识别方法，其特征在于，所述步骤B1中MAIRNet的主体是一个由17个逆残差模块构建的逆残差网络，每一个逆残差模块首端为1×1卷积，1×1卷积用于扩张特征矩阵通道，丰富特征数量，逆残差模块中间部分由卷积核为3×3的深度卷积组成，用于生成与输入特征矩阵通道数一致的特征矩阵，从而减少参数量以及运算成本，最终使用ReLU6激活函数增强网络的表达能力；6. According to a PCB defect detection and identification method based on MAIRNet according to claim 1, it is characterized in that the main body of MAIRNet in step B1 is an inverse residual network constructed by 17 inverse residual modules, each inverse residual module has a 1×1 convolution at the head end, and the 1×1 convolution is used to expand the feature matrix channel and enrich the number of features. The middle part of the inverse residual module is composed of a deep convolution with a convolution kernel of 3×3, which is used to generate a feature matrix consistent with the number of channels of the input feature matrix, thereby reducing the number of parameters and computing costs, and finally using the ReLU6 activation function to enhance the expression ability of the network; 所述步骤B2中多维注意力模块的具体操作为：通过每一个逆残差模块的特征矩阵被分为两个通道进行平均池化操作，平均池化后的两个通道的特征矩阵分别进行连接及1×1卷积，使用Batch Norm归一化操作，并通过Swish激活函数进行激活，将激活后的特征矩阵再次分解为两个通道的特征矩阵，对两个通道的特征矩阵分别使用1×1卷积和Sigmoid激活函数进行激活，得到的两个通道的特征矩阵与逆残差模块输出的原始特征矩阵进行合并，从而得到多维注意力增强的特征图；The specific operation of the multidimensional attention module in step B2 is as follows: the feature matrix of each inverse residual module is divided into two channels for average pooling operation, the feature matrices of the two channels after average pooling are respectively connected and 1×1 convolved, Batch Norm normalization operation is used, and activated by Swish activation function, the activated feature matrix is decomposed into feature matrices of two channels again, and the feature matrices of the two channels are respectively activated by 1×1 convolution and Sigmoid activation function, and the obtained feature matrices of the two channels are merged with the original feature matrix output by the inverse residual module, so as to obtain a feature map enhanced by multidimensional attention; 所述步骤B3具体为：将17个多维注意力模块优化的逆残差模块分为7层网络，其中第1层至第7层，每层依次包含：1，2，3，4，3，3，1个多维注意力优化的逆残差模块，将经过第2层、第3层、第5层、第7层网络处理的特征图分别定义为C2、C3、C5、C7进行1×1卷积得到F2、F3、F5、F7，并将F7再次进行步长为2的卷积得到F7’，对不同尺度的特征图进行融合，最终融合输出5层特征图F2、F3、F5、F7、F7’。The step B3 is specifically as follows: the inverse residual modules optimized by 17 multidimensional attention modules are divided into a 7-layer network, wherein each layer from the 1st layer to the 7th layer comprises: 1, 2, 3, 4, 3, 3, 1 multidimensional attention optimized inverse residual modules, and the feature maps processed by the 2nd, 3rd, 5th and 7th layers of the network are defined as C 2 , C 3 , C 5 , C 7 respectively, and 1×1 convolution is performed on F 2 , F 3 , F 5 , F 7 , and F 7 is obtained, and F 7 is convolved again with a step size of 2 to obtain F 7 ', and feature maps of different scales are fused, and finally the fusion outputs 5 layers of feature maps F 2 , F 3 , F 5 , F 7 , F 7 '. 7.根据权利要求1所述的一种基于MAIRNet的PCB缺陷检测与识别方法，其特征在于，所述步骤S5中通过搭建用户软件系统界面进行汇总显示。7. A PCB defect detection and identification method based on MAIRNet according to claim 1, characterized in that in step S5, a user software system interface is built for summary display.
CN202111318512.9A 2021-11-09 2021-11-09 A PCB defect detection and identification method based on MAIRNet Active CN114429445B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111318512.9A CN114429445B (en) 2021-11-09 2021-11-09 A PCB defect detection and identification method based on MAIRNet

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111318512.9A CN114429445B (en) 2021-11-09 2021-11-09 A PCB defect detection and identification method based on MAIRNet

Publications (2)

Publication Number Publication Date
CN114429445A CN114429445A (en) 2022-05-03
CN114429445B true CN114429445B (en) 2025-05-23

Family

ID=81311239

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111318512.9A Active CN114429445B (en) 2021-11-09 2021-11-09 A PCB defect detection and identification method based on MAIRNet

Country Status (1)

Country Link
CN (1) CN114429445B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116703926B (en) * 2023-08-08 2023-11-03 苏州思谋智能科技有限公司 Defect detection method, defect detection device, computer equipment and computer readable storage medium
CN117409261B (en) * 2023-12-14 2024-02-20 成都数之联科技股份有限公司 Element angle classification method and system based on classification model
CN118226236B (en) * 2024-05-23 2024-08-06 联觉(深圳)科技有限公司 Intelligent recognition method and device for PCB defects

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108921846A (en) * 2018-07-17 2018-11-30 北京航空航天大学 A kind of rail tread defect identification method combined based on gray level image and depth image
CN111738994A (en) * 2020-06-10 2020-10-02 南京航空航天大学 A Lightweight PCB Defect Detection Method

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108846841A (en) * 2018-07-02 2018-11-20 北京百度网讯科技有限公司 Display screen quality determining method, device, electronic equipment and storage medium
CN110766011B (en) * 2019-12-26 2020-04-28 南京智莲森信息技术有限公司 Contact net nut abnormity identification method based on deep multistage optimization
CN112232240B (en) * 2020-10-21 2024-08-27 南京师范大学 Road casting object detection and identification method based on optimized cross-over ratio function
CN112766195B (en) * 2021-01-26 2022-03-29 西南交通大学 Electrified railway bow net arcing visual detection method
CN112818871B (en) * 2021-02-04 2024-03-29 南京师范大学 Target detection method of full fusion neural network based on half-packet convolution
CN113469177B (en) * 2021-06-30 2024-04-26 河海大学 Deep learning-based drainage pipeline defect detection method and system

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108921846A (en) * 2018-07-17 2018-11-30 北京航空航天大学 A kind of rail tread defect identification method combined based on gray level image and depth image
CN111738994A (en) * 2020-06-10 2020-10-02 南京航空航天大学 A Lightweight PCB Defect Detection Method

Also Published As

Publication number Publication date
CN114429445A (en) 2022-05-03

Similar Documents

Publication Publication Date Title
CN114429445B (en) A PCB defect detection and identification method based on MAIRNet
CN111179251B (en) Defect detection system and method using template comparison based on Siamese neural network
CN111862064B (en) Silver wire surface flaw identification method based on deep learning
CN111368690B (en) Deep learning-based video image ship detection method and system under influence of sea waves
CN101915769B (en) An automatic optical inspection method for resistive components in printed circuit boards
CN111738994B (en) Lightweight PCB defect detection method
CN109285139A (en) A deep learning-based X-ray imaging weld inspection method
CN113222913B (en) Circuit board defect detection positioning method, device and storage medium
CN105574550A (en) Vehicle identification method and device
CN112037219A (en) Metal surface defect detection method based on two-stage convolution neural network
CN109117836A (en) Text detection localization method and device under a kind of natural scene based on focal loss function
CN110929795B (en) Method for quickly identifying and positioning welding spot of high-speed wire welding machine
CN113344041B (en) A PCB defect image recognition method based on multi-model fusion convolutional neural network
CN112183639B (en) A Mineral Image Recognition and Classification Method
CN115456955A (en) A method for detecting internal burr defects of ball cage dust cover
CN117036243A (en) Method, device, equipment and storage medium for detecting surface defects of shaving board
CN116071315A (en) Product visual defect detection method and system based on machine vision
CN111680691B (en) Text detection method, text detection device, electronic equipment and computer readable storage medium
CN117152080A (en) A PCB defect image detection method that combines GAM and SimAM attention mechanisms to improve YOLOv5
Chen et al. EEE-Net: Efficient edge enhanced network for surface defect detection of glass
CN113642473A (en) Mining coal machine state identification method based on computer vision
CN113762162A (en) Fire early warning method and system based on semantic segmentation and recognition
CN118297926A (en) Online detection method and system for metal processing multi-mode defects
CN112529835A (en) Image splicing tampering detection and positioning method based on source camera identification
CN111738264A (en) An intelligent collection method of display panel data of equipment room equipment

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