CN117455835A - Improved PCB surface defect detection method based on YOLOv5 - Google Patents
Improved PCB surface defect detection method based on YOLOv5 Download PDFInfo
- Publication number
- CN117455835A CN117455835A CN202311212657.XA CN202311212657A CN117455835A CN 117455835 A CN117455835 A CN 117455835A CN 202311212657 A CN202311212657 A CN 202311212657A CN 117455835 A CN117455835 A CN 117455835A
- Authority
- CN
- China
- Prior art keywords
- model
- pcb
- network
- detection
- yolov5
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 62
- 230000007547 defect Effects 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 claims abstract description 33
- 230000007246 mechanism Effects 0.000 claims abstract description 18
- 230000006872 improvement Effects 0.000 claims abstract description 8
- 238000012545 processing Methods 0.000 claims abstract description 5
- 238000000605 extraction Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 52
- 230000006870 function Effects 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 5
- 229910052802 copper Inorganic materials 0.000 claims description 5
- 239000010949 copper Substances 0.000 claims description 5
- 238000002372 labelling Methods 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 4
- 101000863875 Mus musculus Tyrosine-protein phosphatase non-receptor type substrate 1 Proteins 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 150000003071 polychlorinated biphenyls Chemical class 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 2
- 238000002474 experimental method Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 10
- 238000010801 machine learning Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 2
- 238000013523 data management Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 241000555745 Sciuridae Species 0.000 description 1
- 208000003464 asthenopia Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
- G06V10/763—Non-hierarchical techniques, e.g. based on statistics of modelling distributions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/7715—Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
A PCB surface defect improvement detection method based on YOLOv5 relates to a computer image processing and industrial control detection method, which comprises the steps of firstly, utilizing a Mini-Batch K-means method to conduct re-clustering, selecting an anchor frame more suitable for a data set, additionally, adding an SE attention mechanism behind a backbone network CBS module, thereby highlighting a useful space channel, improving the extraction capacity of small target features, reducing interference between the background and the defect in an image, and finally adopting an EIOU loss function to improve the detection precision of a prediction frame. The model is used for carrying out experiments on a PCB_DATASET data set, and the experimental results are compared with different scale models of YOLOv5, so that the results show that the method is more accurate and faster than many other methods in the aspect of detecting the surface defects of the product in real time and high precision. The method solves the problems of low detection speed, low network precision and missing error detection in the industrial detection of the printed circuit board.
Description
Technical Field
The invention relates to a computer image processing and industrial control detection method, in particular to a PCB surface defect improvement detection method based on YOLOv 5.
Technical Field
With the development of the electronic industry, the electronic industry plays an important role in modern manufacturing. As an important electronic component, a Printed Circuit Board (PCB) is a carrier to which various electronic components are connected, providing line connection and hardware support for the device. As small as electronic watches, calculators, as large as computers, communications electronics, military weapon systems, and as long as there are electronic components such as integrated circuits, almost every electronic device requires a PCB. However, the printed circuit board manufacturing process is complex and is prone to weeping, squirrel biting, open circuits, short circuits and other small defects. In order to ensure the safety and reliability of the electronic device, it is necessary to detect surface defects of the PCB.
Traditional manual detection is easily interfered by external environmental factors, so that the defect detection efficiency is affected. In addition, detection of microscopic defects can cause visual fatigue, resulting in misclassification. To address these issues, some students have introduced machine learning into PCB inspection and have made great progress. Wang et al have combined machine learning knowledge to propose an automatic PCB pinhole detection method. Yuk et al use accelerated robust features and random forest methods to enable detection of PCB defects. By considering the density of features, a Weighted Kernel Density Estimation (WKDE) map is generated using weighted probabilities, thereby enabling detection of defect concentration areas. V et al use the similarity index to detect surface defects of the PCB.
This method can be used to detect and locate localized defects in complex component mounted PCB images. Also, scholars have proposed machine learning-based methods for detecting surface defects of PCBs, none of which are real-time methods. Although machine learning-based methods can realize the identification of surface defects of PCBs, most methods still require manual setting of image features through a priori knowledge, resulting in a lack of generalization capability of the method.
Disclosure of Invention
The invention aims to provide a PCB surface defect improvement detection method based on YOLOv5, which adopts a YOLOv5 target detection method to process a Printed Circuit Board (PCB) image, extracts picture characteristics and transmits information data required in the picture to a PCB detection module, and the module obtains the PCB defect condition through the data and classifies defects.
The invention adopts the following technical scheme:
the method for improving and detecting the surface defects of the PCB based on the YOLOv5 comprises the following steps of:
1) Collecting an intelligent robot open laboratory open source Dataset PCB_Dataset;
2) Adopting a deep learning pyrach framework to configure the environment of the network, and completing the model construction of yolov5 in the environment;
3) Adding an SE attention mechanism into the original model framework of yolov5 to optimize the model;
4) Taking the preprocessed data set as the input of the network and training, loading yolov5 pre-training weight, and taking EIOU as a loss function;
5) Putting the shot PCB picture into a Yolov5 network model for quality detection;
6) The PCB system detection and identification module sorts the PCB conditions detected by the system, and images according to six types of defects are included: holes (missing_hole), copper (residual_cap), short (short), mouse bite (mouse_bit), open_circuit (open_circuit), and burrs (spike) are classified.
The improved detection method of the PCB surface defects based on the YOLOv5 comprises the steps of preprocessing a picture data set of an input model; the data set adopts an intelligent robot open laboratory open source data set PCB_Dataset; collecting and sorting pictures in a data set, converting the pictures into a JPG format, then manually labeling the picture data by a labeling tool, and outputting a label format in an xml format; during network training, data needs to be divided into a training set and a testing set, wherein 80% of the data set is used as the training set of the network, and the remaining 20% of the data set is used as the testing set.
According to the PCB surface defect improvement detection method based on the YOLOv5, the image data set is preprocessed by the method, and the construction of an image processing target detection model is completed; the model architecture is divided into four parts, namely an input end, a Backbone network, a Neck network layer and a Head output end, the input end processes an input picture, the speed of model training and the precision of the network are improved by adopting modes of Mixup data enhancement, self-adaptive anchor frame calculation, self-adaptive picture scaling and the like, firstly, the Mini-Batch K-means method is utilized for reclustering, an anchor frame which is more suitable for the data set is selected, in addition, an SE attention mechanism is added behind a Backbone network CBS module, thereby highlighting a useful space channel, improving the extraction capacity of small target features, reducing the interference between the background and the defect in the image, and finally, the EIOU loss function is adopted for improving the detection precision of a prediction frame; model loading yolov5 pre-training weight, initial learning rate of 0.0005, momentum setting of 0.937, loss function of EIOU, and other parameters of EIOU.
According to the improved PCB surface defect detection method based on the YOLOv5, the target detection model is built, the SE attention mechanism is introduced to optimize the network model, the attention mechanism is added into the built YOLOv5 main network, and the method of combining the self-attention mechanism and convolution is used for the network, so that the PCB detection precision is improved.
According to the PCB surface defect improvement detection method based on the YOLOv5, optimization of a network model is completed, parameters such as the number of data input into the network at one time, the number of training wheels, working threads and the like are set, and training of the model is started; after model training is completed, checking whether the performance index of the model is reasonable, detecting input data, and checking the confidence coefficient of the prediction frame.
According to the improved detection method for the PCB surface defects based on the YOLOv5, model training is completed and testing is carried out, a PCB detection system extracts defects of a printed circuit board by using a target detection network, and the system passes through the six types of defect images: the defects, residual copper, short circuits, rat bites, opens and burrs are classified.
The invention has the following beneficial effects:
the invention adopts the method based on target detection, adds the image recognition into the PCB detection system, and detects the system through the PCB system detection recognition module, so that the image condition of the Printed Circuit Board (PCB) can be obtained in time, and compared with the traditional time sequence prediction method, the method has extremely strong authenticity and timeliness. In addition, the scheduling method can be optimized by detecting the presence of a defect-free printed circuit board. And an SE attention mechanism is added in the yolov5 target detection method, so that the original method is optimized, the network precision is improved, and the detection of small targets in a printed circuit board is facilitated.
Drawings
FIG. 1 is a diagram of a Printed Circuit Board (PCB) inspection configuration of the present invention;
FIG. 2 is a diagram of a yolov5 model architecture of the present invention;
FIG. 3 is a diagram of the network architecture of the SE attention mechanism of the present invention;
FIG. 4 is a flow chart of the model training of the present invention;
FIG. 5 is a map precision map of the present invention;
FIG. 6 is a graph of the loss function of the present invention;
fig. 7 is a schematic diagram of the system configuration of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in further detail with reference to the accompanying drawings.
The whole flow of the invention is as follows:
1. collecting an intelligent robot open laboratory open source Dataset PCB_Dataset;
2. adopting a deep learning pyrach framework to configure the environment of the network, and completing the model construction of yolov5 in the environment;
3. adding an SE attention mechanism into the original model framework of yolov5 to optimize the model;
4. taking the preprocessed data set as the input of the network and training, loading yolov5 pre-training weight, and taking EIOU as a loss function;
5. putting the shot PCB picture into a Yolov5 network model for quality detection;
and 6, the PCB detection and identification module sorts the PCB conditions detected by the system, and the system comprises six types of defect images: holes (missing_hole), copper (residual_cap), short (short), mouse bite (mouse_bit), open_circuit (open_circuit), and burrs (spike) are classified.
The specific implementation steps of the step 1 are as follows:
1. collecting pictures with six different defects of the PCB, and marking the pictures by labelimg;
2. converting the marked data into a txt format of yolo;
3. the data set is divided into a training set and a test set.
The specific implementation steps of the step 2 are as follows:
a configuration environment required for importing the project;
and constructing a yolov5 model, wherein the main body comprises an Input end, a Backbone network, a Neck network layer and a Head output end.
The specific implementation steps of the step 3 are as follows:
the network architecture diagram of the SE attention mechanism module is introduced, and after the feature diagram is input, the feature diagram is subjected to global average pooling based on the width and the height of the feature diagram, so that the dimension of the spatial feature is reduced to 1 multiplied by 1, as shown in a formula 1. The connection between the channels is then established using two full connection layers and a nonlinear activation function such that each channel is represented by the following formula:
the method is completed through two full-connection layers again, weight information required by the user is generated through weight W, wherein W is obtained through learning and used for displaying modeling of characteristic correlation required by the user, and a calculation formula is as follows
s＝F ex (z,W)＝σ(g(z,W))＝σ(W 2 δ(W 1 z)) (2)
The vector z obtained in the previous step is processed through the two full-connection layers W1 and W2 to obtain a channel weight value s which is wanted by us, and after the two full-connection layers are passed, different values in the s represent weight information of different channels, and different weights are given to the channels. Finally, will
Thirdly, generating a weight vector s to carry out weight assignment on the feature map U to obtain the feature map which is wanted by usThe size of the SE module is identical to the size of the feature map U, the SE module does not change the size of the feature map, and the calculation formula is as follows:
the specific implementation steps of the step 4 are as follows:
loading a pre-training weight of yolov5, and improving the speed and precision of network training;
the preprocessed data set is divided into a training set and a testing set and sent to a network;
setting the base size as 16, setting the epochs as 300, setting the input size as 640, and selecting the super-parameters as default settings;
the Precision and recall rate ReECALL are selected as indexes of a measurement model, and the formulas are respectively as follows:
wherein TP is correctly identified, FP is positively identified as negatively identified, FN is positively identified; in order to solve the problem of divergence of the IOU in the training process, CIOU is adopted as a loss function of the training box, the EIOU considers the distance between the target and the anchor, the overlapping rate, the scale and the punishment factor, the regression stability of the target frame is enhanced, the regression convergence speed of the prediction frame and the real frame is accelerated, and the formula is as follows:
the specific implementation steps of the step 5 are as follows:
detecting performance indexes such as Precision, reECAll, F score, map, loss function and the like of the training model; a Printed Circuit Board (PCB) image is input into the network, and the confidence of the network identification is detected.
The specific implementation steps of the step 6 are as follows:
the PCB system detection and identification module sorts the PCB conditions detected by the system to obtain the PCB quality conditions, the data are transmitted to the data management module, and the data management module sorts the data and transmits the data to the corresponding data modules respectively; and performing more timely and accurate classification tasks.
The invention is applied to detection and identification of a Printed Circuit Board (PCB) system by using a target detection method, taking a defective PCB as an example, the system is divided into two parts, namely a target detection method network and PCB mode identification, and a PCB detection structure diagram is shown in figure 1. The method comprises the steps that preprocessing operation is needed to be carried out on a picture data set of an input model before a network is built; the data set adopts an open source data set PCB_Dataset of an intelligent robot open laboratory of Beijing university; collecting and sorting pictures in a data set, converting the pictures into a JPG format, then manually labeling the picture data by a labelimg tool, and outputting a label format in an xml format; during network training, data needs to be divided into a training set and a testing set, wherein 80% of the data set is used as the training set of the network, and the remaining 20% of the data set is used as the testing set.
Completing the construction of an image processing target detection model; the model architecture is divided into four parts, namely an input end, a Backbone network, a Neck network layer and a Head output end, the input end processes an input picture, the speed of model training and the precision of the network are improved by adopting modes of Mixup data enhancement, self-adaptive anchor frame calculation, self-adaptive picture scaling and the like, firstly, the Mini-Batch K-means method is utilized for reclustering, an anchor frame which is more suitable for the data set is selected, in addition, an SE attention mechanism is added behind a Backbone network CBS module, thereby highlighting a useful space channel, improving the extraction capacity of small target features, reducing the interference between the background and the defect in the image, and finally, the EIOU loss function is adopted for improving the detection precision of a prediction frame; model loading yolov5 pre-training weight, initial learning rate of 0.0005, momentum setting of 0.937, loss function of EIOU, and other parameters of EIOU.
And after the initial model of yolov5 is built, adding an SE attention mechanism into the backbone network to optimize the network model. SE is a novel mobile network attention mechanism that can improve network accuracy. A block diagram of the SE attention machine is shown in fig. 3. After model improvement, the preprocessed picture data set needs to be input into the network for training. Pre-training weights of yolov5 are added before training the model to optimize the training of the model, shorten the time of network training and improve the accuracy of model prediction. Setting the path, the type and the number of the training data set when parameters are changed for the training model; loading a pre-training weight type and a path of the model; selecting super parameters and activation functions which accord with the training type of the data set; setting a training working thread, the number of pictures of an input network and the number of training rounds. After the model parameter setting is completed, training the model to obtain model weights conforming to the system, wherein the training process of the model is shown in fig. 4.
After model training, performance indexes such as Precision, reECAll, F score and map are adopted to check whether the model training is good or not, the map precision of the model is shown in fig. 5, and the loss function is shown in fig. 6. And taking the PCB picture as the input of the model to detect the accuracy of the model for identifying six defects.
Claims (6)
1. The improved PCB surface defect detection method based on YOLOv5 is characterized by comprising the following steps of constructing a network model for PCB quality detection, extracting and detecting and classifying input images by a PCB quality detection and identification module:
1) Collecting an intelligent robot open laboratory open source Dataset PCB_Dataset;
2) Adopting a deep learning pyrach framework to configure the environment of the network, and completing the model construction of yolov5 in the environment;
3) Adding an SE attention mechanism into the original model framework of yolov5 to optimize the model;
4) Taking the preprocessed data set as the input of the network and training, loading yolov5 pre-training weight, and taking EIOU as a loss function;
5) Putting the shot PCB picture into a Yolov5 network model for quality detection;
6) The PCB system detection and identification module sorts the PCB conditions detected by the system, and images according to six types of defects are included: holes (missing_hole), copper (residual_cap), short (short), mouse bite (mouse_bit), open_circuit (open_circuit), and burrs (spike) are classified.
2. The YOLOv 5-based PCB surface defect improvement detection method of claim 1, wherein the method performs a preprocessing operation on a picture dataset of an input model; the data set adopts an intelligent robot open laboratory open source data set PCB_Dataset; collecting and sorting pictures in a data set, converting the pictures into a JPG format, then manually labeling the picture data by a labeling tool, and outputting a label format in an xml format; during network training, data needs to be divided into a training set and a testing set, wherein 80% of the data set is used as the training set of the network, and the remaining 20% of the data set is used as the testing set.
3. The YOLOv 5-based PCB surface defect improvement detection method of claim 1, wherein the method performs preprocessing on a picture dataset to complete construction of an image processing target detection model; the model architecture is divided into four parts, namely an input end, a Backbone network, a Neck network layer and a Head output end, the input end processes an input picture, the speed of model training and the precision of the network are improved by adopting modes of Mixup data enhancement, self-adaptive anchor frame calculation, self-adaptive picture scaling and the like, firstly, the Mini-Batch K-means method is utilized for reclustering, an anchor frame which is more suitable for the data set is selected, in addition, an SE attention mechanism is added behind a Backbone network CBS module, thereby highlighting a useful space channel, improving the extraction capacity of small target features, reducing the interference between the background and the defect in the image, and finally, the EIOU loss function is adopted for improving the detection precision of a prediction frame; model loading yolov5 pre-training weight, initial learning rate of 0.0005, momentum setting of 0.937, loss function of EIOU, and other parameters of EIOU.
4. The improved detection method of surface defects of a PCB based on YOLOv5 of claim 1, wherein the method of constructing a target detection model, introducing an SE attention mechanism to optimize a network model, adding the attention mechanism in a constructed YOLOv5 backbone network, and combining the self-attention mechanism and convolution is used for the network, so that the detection accuracy of the PCB is improved.
5. The improved detection method of PCB surface defects based on YOLOv5 of claim 1, wherein the optimization of the network model is completed, parameters such as the number of data input into the network at one time, the number of training rounds, working threads and the like are set, and training of the model is started; after model training is completed, checking whether the performance index of the model is reasonable, detecting input data, and checking the confidence coefficient of the prediction frame.
6. The YOLOv 5-based improved detection method of surface defects of PCBs according to claim 1, wherein model training is completed and tested, the PCB detection system extracts defects of the printed circuit board using a target detection network, and the system passes through the six types of defect images: the defects, residual copper, short circuits, rat bites, opens and burrs are classified.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311212657.XA CN117455835A (en) | 2023-09-20 | 2023-09-20 | Improved PCB surface defect detection method based on YOLOv5 |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311212657.XA CN117455835A (en) | 2023-09-20 | 2023-09-20 | Improved PCB surface defect detection method based on YOLOv5 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117455835A true CN117455835A (en) | 2024-01-26 |
Family
ID=89582516
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311212657.XA Pending CN117455835A (en) | 2023-09-20 | 2023-09-20 | Improved PCB surface defect detection method based on YOLOv5 |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117455835A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117952977A (en) * | 2024-03-27 | 2024-04-30 | 山东泉海汽车科技有限公司 | Pavement crack identification method, device and medium based on improvement yolov s |
-
2023
- 2023-09-20 CN CN202311212657.XA patent/CN117455835A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117952977A (en) * | 2024-03-27 | 2024-04-30 | 山东泉海汽车科技有限公司 | Pavement crack identification method, device and medium based on improvement yolov s |
| CN117952977B (en) * | 2024-03-27 | 2024-06-04 | 山东泉海汽车科技有限公司 | Pavement crack identification method, device and medium based on improvement yolov s |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111179251B (en) | Defect detection system and method based on twin neural network and by utilizing template comparison | |
| CN114627383B (en) | Small sample defect detection method based on metric learning | |
| CN105574550A (en) | Vehicle identification method and device | |
| CN110610197A (en) | Method and device for mining difficult sample and training model and electronic equipment | |
| CN109284779A (en) | Object detection method based on deep full convolution network | |
| CN117455835A (en) | Improved PCB surface defect detection method based on YOLOv5 | |
| CN112651966A (en) | Printed circuit board micro-defect detection method based on ACYOLOV4_ CSP | |
| CN113409250A (en) | Solder joint detection method based on convolutional neural network | |
| Liao et al. | Solder joint defect inspection method based on ConvNeXt-YOLOX | |
| CN114418980A (en) | Deep learning method, system, equipment and medium for printed circuit board defect identification | |
| Luan et al. | A Siamese network utilizing image structural differences for cross-category defect detection | |
| CN114708518A (en) | Bolt defect detection method based on semi-supervised learning and priori knowledge embedding strategy | |
| CN112184679A (en) | YOLOv 3-based wine bottle flaw automatic detection method | |
| CN113724233B (en) | Transformer equipment appearance image defect detection method based on fusion data generation and transfer learning technology | |
| CN113447902A (en) | Sea surveillance radar target identification method based on machine learning | |
| CN114926702B (en) | Small sample image classification method based on depth attention measurement | |
| CN116520154A (en) | Performance evaluation method and system for lithium battery pole piece detection | |
| CN116152169A (en) | Printed circuit board defect detection method based on improved YOLOv5s model | |
| CN109543571A (en) | A kind of intelligent recognition and search method of Complex Product abnormity machining feature | |
| CN115620083A (en) | Model training method, face image quality evaluation method, device and medium | |
| Austria et al. | Defect detection and classification in printed circuit boards using Convolutional Neural Networks | |
| CN108846421B (en) | Image classification method based on deep learning | |
| CN114692887A (en) | Semi-supervised learning system and semi-supervised learning method | |
| Poxi et al. | Bushing Surface Defect Detection Method Based on Improved YOLOX | |
| CN113887567B (en) | Vegetable quality detection method, system, medium and 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 |