CN117173444A - Edge banding board appearance defect detection method and system based on improved YOLOv7 network model - Google Patents
Edge banding board appearance defect detection method and system based on improved YOLOv7 network model Download PDFInfo
- Publication number
- CN117173444A CN117173444A CN202310679878.1A CN202310679878A CN117173444A CN 117173444 A CN117173444 A CN 117173444A CN 202310679878 A CN202310679878 A CN 202310679878A CN 117173444 A CN117173444 A CN 117173444A
- Authority
- CN
- China
- Prior art keywords
- yolov7
- network model
- improved
- edge sealing
- defect detection
- 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
- 230000007547 defect Effects 0.000 title claims abstract description 64
- 238000001514 detection method Methods 0.000 title claims abstract description 59
- 238000007789 sealing Methods 0.000 claims abstract description 60
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000012545 processing Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 14
- 238000002360 preparation method Methods 0.000 claims abstract description 3
- 230000006870 function Effects 0.000 claims description 31
- 238000004519 manufacturing process Methods 0.000 claims description 18
- 230000007246 mechanism Effects 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 16
- 238000011176 pooling Methods 0.000 claims description 14
- 239000003292 glue Substances 0.000 claims description 13
- 238000009966 trimming Methods 0.000 claims description 8
- 238000007373 indentation Methods 0.000 claims description 7
- 238000004458 analytical method Methods 0.000 claims description 6
- 238000013519 translation Methods 0.000 claims description 5
- 238000005520 cutting process Methods 0.000 claims description 4
- 238000013135 deep learning Methods 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 3
- 238000007689 inspection Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000003672 processing method Methods 0.000 claims description 3
- 210000000988 bone and bone Anatomy 0.000 abstract description 4
- 230000000052 comparative effect Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 239000000853 adhesive Substances 0.000 description 2
- 230000001070 adhesive effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 239000002023 wood Substances 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The application relates to a method and a system for detecting appearance defects of an edge sealing plate based on an improved YOLOv7 network model, wherein the system comprises an image acquisition unit, a defect detection unit and a result output unit; the detection method comprises the steps of defining defect types, sample preparation, network model training, defect detection and result output. Through adding the channel attention module and the space attention module after each convolution of the back bone layer of the YOLOv7 network model, the appearance defects of the edge sealing plate can be effectively detected, and the method has higher precision, recall rate and processing speed.
Description
Technical Field
The application belongs to the technical field of edge sealing plates, and particularly relates to an edge sealing plate appearance defect detection method and system based on an improved YOLOv7 network model.
Background
The edge sealing procedure is an important ring for manufacturing the plate furniture, plays an important role in beautifying the furniture and reducing pollution, but has great difficulty and a plurality of problems in the edge sealing process. The edge sealing quality effect can directly influence the overall appearance quality of furniture, and the problems of edge breakage, glue leakage, uneven trimming and the like occur. The statistics of the customer complaint rate of furniture quality shows that the feedback of the related edge sealing quality problem is high, and the problem of the edge sealing quality is solved for furniture enterprises.
The quality management of wood furniture in China is the original system, but the problems of insufficient standard coverage, insufficient timeliness and the like do exist. In modern furniture enterprise quality management, detection of edge sealing appearance quality is mostly dependent on manual work, namely sampling detection and batch detection after production, sampling detection items are more and comprehensive but only can reflect whether production quality is qualified or not in a period of time, batch detection is mainly used for rapidly detecting products produced by each production group after production, and unqualified products often occur due to short time consumption, incomplete detection items and the like. The customer needs to repair the order when complaining the product quality, which consumes time and labor, thus improving the product quality and reducing the generation of unqualified products, and being a necessary choice for improving the product qualification rate and the production efficiency and reducing the production cost. The key points of appearance quality control are more, and the control dimensions and requirements of single items are different, so that the establishment of a proper appearance detection system and method is more necessary for improving the appearance quality of the edge sealing, and the improvement of the speed and accuracy of the appearance quality detection of the plate is urgent.
As the machine vision technology is gradually applied to the defect detection of industrial production, new methods and means are rapidly developed, and the defect detection is changed from the initial machine vision and manual to the machine vision and machine learning and gradually updated to the real-time detection of defects generated in the production process of products by using the deep learning technology.
However, due to factors such as the size and shape of the edge sealing plates, the shape and distribution of the appearance defects of the edge sealing plates are also greatly different, which brings certain challenges to the task of detecting the appearance defects of the edge sealing plates. The traditional convolutional neural network is easy to take excessive attention to unimportant features, and the like, has poor model performance, and cannot meet the requirements of the current stage.
Disclosure of Invention
The application provides a method and a system for detecting the appearance defects of an edge sealing plate based on an improved YOLOv7 network model, which solve the technical problems of the prior art that the appearance defects of the edge sealing plate are detected inadequately intelligently and the detection effect is poor due to poor model performance, can effectively detect the appearance defects of the edge sealing plate, and have the technical effects of higher precision, recall rate, processing speed and the like.
In order to achieve the above purpose, the technical scheme adopted by the application is as follows:
the edge banding board appearance defect detection method based on the improved YOLOv7 network model comprises the following steps:
1) Defining defect types
The appearance defect types of the edge sealing plate are divided into: opening glue, shortage, collapse, uneven trimming, glue lines and edge sealing indentation;
2) Sample preparation
Collecting a plurality of edge sealing pictures in the edge sealing plate production process, and performing data processing and analysis to obtain a sample to be trained, wherein the data processing and analysis comprises cutting, data enhancement and marking processes;
3) Network model training
Inputting the sample to be trained into an improved YOLOv7 network model for training, so that the sample to be trained learns to identify the appearance defects of the edge sealing plate, and a trained network is obtained;
the improved YOLOv7 network model adds an attention mechanism after each convolution of a backlight layer of YOLOv7, wherein the attention mechanism comprises a channel attention module and a space attention module;
4) Defect detection
Acquiring an image of the edge sealing plate to be detected in real time, and processing, detecting and analyzing the image of the edge sealing plate to be detected by utilizing the network trained in the step 3);
5) Result output
Outputting the defect type of the edge sealing plate to be detected.
Further, in step 3), the channel attention module weights the feature map of each channel, and the function is:
the spatial attention module weights the feature map of each spatial location as a function of:
the function of the attentiveness mechanism is:
wherein x represents the input feature map, avePool (x) and MaxPool (x) represent the average pooling and maximum pooling operations for x, MLP_ { c } and MLP_ { s } represent a fully connected layer,representing a sigmoid function, fc { c } represents the function of the channel attention module, and fs { s } represents the function of the spatial attention module.
Further, in step 2), the clipping is: the picture is cropped and extraneous portions are removed so that the picture size is uniform to 640 x 640 pixels.
Further, in step 2), the data enhancement includes one or more of rotation, translation, and scaling.
Further, in step 2), the marking includes: and marking the pictures by using LabelImg target marking software, wherein each picture contains one or more defects, and the marking content contains defect types, position coordinates and defect ranges.
Further, in step 4), an industrial CCD camera is used to acquire an image of the edge sealing plate to be measured in the production process in real time.
Further, the YOLOv7 network model includes an input layer, a back layer and a back and head layer, the processing method includes preprocessing an input picture, dividing the input picture into 640 x 3 pictures, inputting the processed picture into the back layer, continuously processing the picture output by the back and head layer, outputting feature maps with different sizes, and finally analyzing the image through a RepVGG convolution architecture to obtain a final result.
Further, in step 3), the network model training is performed by using a Pytorch deep learning framework, specifically: the size of the input picture is 640 x 640 pixels, the initial learning rate is 0.001, the batch size is 20, the total iteration number is 500, the weight attenuation coefficient is 0.001, the optimization function and the activation function are an Adam optimizer and a leak ReLU function respectively, when the iteration number reaches 400 times, the learning rate is reduced to 0.0001, and after the iteration number reaches 500 times, the model tends to be stable.
Furthermore, the whole process and the result of the network training are visualized online by using a TensorBoard, and the shape, the position and other characteristics of the furniture edge sealing defect are analyzed through the visualized result.
Meanwhile, the application also provides a system for detecting the appearance defects of the edge sealing plate, which is used for realizing the method for detecting the appearance defects of the edge sealing plate based on the improved YOLOv7 network model, and comprises the following steps of:
comprising the following steps:
the image acquisition unit is used for acquiring the image of the edge sealing plate to be detected in real time;
the defect detection unit is used for processing, detecting and analyzing the image by adopting a trained network;
and the result output unit is used for outputting the defect types of the edge sealing plate to be tested.
Furthermore, the back bone layer of the YOLOv7 network model adopts a dark net53 network structure, the neg layer adopts an SPP structure, and the head layer adopts a YOLOv7 structure.
Compared with the prior art, the technical scheme of the application has the following beneficial effects:
1. the traditional manual inspection is replaced by a model learning mode, so that production automation is realized, and the production efficiency is improved;
2. the appearance defect types of the edge sealing plate are divided into glue opening, shortage, breakage, uneven trimming, glue lines and edge sealing indentation, so that various defect problems of the edge sealing plate in the actual process are fully reflected;
3. the data enhancement is carried out by adopting modes of rotation, translation, scaling and the like, so that the model is more robust in the training process, the condition of over fitting is prevented, and the accuracy of an output result is improved;
4. color marking and image annotation can be realized by using LabelImg target marking software, so that the recognition efficiency of the appearance quality of the edge sealing plate is effectively improved;
5. the backbone part adopts a dark net53 network structure, so that characteristic information of different layers can be extracted; the neg part adopts a SPP (Spatial Pyramid Pooling) structure, and can carry out pooling operation on the feature images with different scales, so that the expression capability of the feature images is improved; the head part adopts a YOLOv7 structure, so that the target can be detected and positioned;
6. the method comprises the steps of adding a channel attention function and a space attention function after each convolution of a back bone part of YOLOv7, calculating the weight of each channel through global average pooling and a full connection layer, applying the weight to a feature map of each channel to obtain channel attention representation, calculating the weight of each space position through global maximum pooling and the full connection layer, applying the weight to the feature map of each space position to obtain space attention representation, and multiplying the channel attention and the space attention to obtain final attention representation. The attention expression is applied to the feature map, resulting in an output of the attention mechanism. In this way, the attention mechanism of the application can adaptively learn the importance weight of each channel and space position and apply the importance weight to the feature map, thereby improving the attention capability and performance of the model and finally improving the detection precision and efficiency of the edge sealing plate appearance defects;
7. the detection precision, recall rate and F1 value of the model in the edge banding plate appearance defect detection task are respectively 96.3%, 95.7% and 95.9%, and the average detection time is 122.3ms. In contrast, detection accuracy, recall and F1 values in the same task using the conventional YOLOv7 model were 83.8%, 82.2% and 82.7%, respectively, with an average detection time of 137.1ms. The application greatly improves the detection precision.
Drawings
FIG. 1 is a schematic diagram of defect types in embodiment 1 of the present application;
FIG. 2 is a schematic diagram of a portion of a sample after clipping and data enhancement according to embodiment 1 of the present application;
FIG. 3 is a schematic illustration of the labeling process according to embodiment 1 of the present application;
FIG. 4 is a diagram of a YOLOv7 network model architecture;
FIG. 5 is a diagram of a modified Yolov7 network model architecture according to embodiment 1 of the present application;
FIG. 6 is a graph showing the comparison of the P-R curves of the embodiment 1 and the comparative example 1 of the present application;
FIG. 7 is a graph showing mAP curves of specific example 1 of the present application and comparative example 1;
FIG. 8 is a graph showing the comparison of the results of the present application in example 1 and comparative example 1.
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth such as the particular system architecture, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. It will be apparent, however, to one skilled in the art that the present application may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.
Example 1
As shown in fig. 1 to 8, embodiment 1 specifically includes:
the appearance defect types of the edge sealing plate are divided into: opening glue, shortage, collapse, uneven trimming, glue lines and edge sealing indentation; specifically, as shown in fig. 1, a) is adhesive opening, b) is shortage, c) is collapse, d) is uneven trimming, e is adhesive line, and f) is edge sealing indentation;
collecting a plurality of edge sealing pictures in the edge sealing plate production process, performing data processing and analysis to obtain samples to be trained, collecting 1887 pictures altogether, wherein the data processing and analysis comprises cutting, data enhancement and marking processes, firstly cutting the pictures, removing irrelevant parts to ensure that the sizes of the pictures are unified to 640 x 640 pixels, enabling the models to be more robust in the training process through data enhancement processes such as rotation, translation, scaling and the like, preventing fitting, and the processed partial results are shown in figure 2;
marking the picture by using LabelImg target marking software, and using color marking and image annotation, wherein the processing process is shown in figure 3, each picture comprises one or more defects, and marking content comprises defect types, position coordinates, defect ranges and the like; finally obtaining a sample to be trained;
inputting the sample to be trained into an improved YOLOv7 network model for training, so that the sample to be trained learns to identify the appearance defects of the edge sealing plate, and a trained network is obtained;
as shown in fig. 4 and 5, the YOLOv7 network model includes an input layer, a back layer and a back and head layer, the processing method includes preprocessing an input picture, dividing the input picture into 640 x 3 pictures, inputting the processed picture into the back layer, continuously processing the picture output by the back and head layer, outputting feature maps with different sizes, and finally analyzing the image through a RepVGG convolution architecture to obtain a final result.
The improved YOLOv7 network model adds an attention mechanism after each convolution of a backlight layer of YOLOv7, wherein the attention mechanism comprises a channel attention module and a space attention module;
the channel attention module weights the feature map of each channel, and the function is as follows:
the spatial attention module weights the feature map of each spatial location as a function of:
the function of the attentiveness mechanism is:
wherein x represents the input feature map, avePool (x) and MaxPool (x) represent the average pooling and maximum pooling operations for x, MLP_ { c } and MLP_ { s } represent a fully connected layer,representing a sigmoid function, fc { c } represents the function of the channel attention module, and fs { s } represents the function of the spatial attention module.
Test configuration: the experiment adopts windows operating system, RTX 3060Ti-12G display card is used for operation, the experiment environment is Pytorch architecture, and the development language is Python.
The network model training is carried out by using a Pytorch deep learning framework, and specifically comprises the following steps: the size of the input picture is 640 x 640 pixels, the initial learning rate is 0.001, the batch size is 20, the total iteration number is 500, the weight attenuation coefficient is 0.001, the optimization function and the activation function are an Adam optimizer and a leak ReLU function respectively, when the iteration number reaches 400 times, the learning rate is reduced to 0.0001, and after the iteration number reaches 500 times, the model tends to be stable. The whole process and the result of the network training are visualized online by using a TensorBoard, and the shape, the position and other characteristics of the furniture edge sealing defect are analyzed through the visualized result.
As can be seen from table 1, fig. 6 and fig. 7, the detection accuracy, recall and F1 value of the modified YOLOv7 network model in example 1 were 96.3%, 95.7% and 95.9%, respectively, and the average detection time was 122.3ms.
And acquiring an image of the edge sealing plate to be detected in the production process in real time by using an industrial CCD camera, processing and detecting and analyzing the image by using a trained neural network, and setting the confidence threshold to be 0.50. The result output schematic diagram is shown in the lower pattern of fig. 8, wherein in fig. 8, a) is the comparison before and after the shortage detection, b) is the comparison before and after the trimming misalignment detection, c) is the comparison before and after the glue line detection, d) is the comparison before and after the glue opening detection, e) is the comparison before and after the edge sealing indentation detection, and f) is the comparison before and after the collapse detection.
Comparative example 1:
the unmodified YOLOv7 network model was tested using the same experimental configuration as in example 1, and as can be seen from table 1, fig. 6 and fig. 7, the detection accuracy, recall and F1 value of the unmodified YOLOv7 network model corresponding to comparative example 1 were 83.8%, 82.2% and 82.7%, respectively, with an average detection time of 137.1ms. The result output is schematically shown in the upper diagram of fig. 8.
TABLE 1 evaluation Table before and after YOLOv7 improvement
| Model comparison | F1 value | Recall rate of recall | mAP@0.5 |
| YOLOv7 | 0.827 | 0.822 | 0.838 |
| Improved YOLOv7 | 0.959 | 0.957 | 0.963 |
From tables 1, 6 and 7, the improvement of the attention mechanism of the application is obviously improved on the basis of the model index of the original YOLOv7, and compared with the YOLOv7, the improved YOLOv7 is improved on mAP@0.5 and P-R curve indexes, wherein mAP@0.5 is most obvious, and reaches 0.963, which indicates that the improved YOLOv7 has better performance on the accuracy of target detection. The attention mechanism may help the model better understand key information in the input sequence, thereby improving the accuracy and generalization ability of the model. In addition, the attention mechanism of the application can also help the model to process long-sequence input, and avoid the problems of information loss, gradient disappearance and the like.
As can be seen from fig. 8, for example, as shown in d) of fig. 8, the detection result of the unmodified YOLOv7 of comparative example 1 is shown at the upper part, the detection result of the modified YOLOv7 of specific example 1 is shown at the lower part, and two positions of the sample to be tested are actually glued, but only one position is detected in comparative example 1, so that the detection accuracy of the modified YOLOv7 is obviously improved.
Based on the comparison experiment, compared with the prior art, the application has the following beneficial effects:
1. the traditional manual inspection is replaced by a model learning mode, so that production automation is realized, and the production efficiency is improved;
2. the appearance defect types of the edge sealing plate are divided into glue opening, shortage, breakage, uneven trimming, glue lines and edge sealing indentation, so that various defect problems of the edge sealing plate in the actual process are fully reflected;
3. the data enhancement is carried out by adopting modes of rotation, translation, scaling and the like, so that the model is more robust in the training process, the condition of over fitting is prevented, and the accuracy of an output result is improved;
4. color marking and image annotation can be realized by using LabelImg target marking software, so that the recognition efficiency of the appearance quality of the edge sealing plate is effectively improved;
5. the backbone part adopts a dark net53 network structure, so that characteristic information of different layers can be extracted; the neg part adopts a SPP (Spatial Pyramid Pooling) structure, and can carry out pooling operation on the feature images with different scales, so that the expression capability of the feature images is improved; the head part adopts a YOLOv7 structure, so that the target can be detected and positioned;
6. the method comprises the steps of adding a channel attention function and a space attention function after each convolution of a back bone part of YOLOv7, calculating the weight of each channel through global average pooling and a full connection layer, applying the weight to a feature map of each channel to obtain channel attention representation, calculating the weight of each space position through global maximum pooling and the full connection layer, applying the weight to the feature map of each space position to obtain space attention representation, and multiplying the channel attention and the space attention to obtain final attention representation. The attention expression is applied to the feature map, resulting in an output of the attention mechanism. In this way, the attention mechanism of the application can adaptively learn the importance weight of each channel and space position and apply the importance weight to the feature map, thereby improving the attention capability and performance of the model and finally improving the detection precision and efficiency of the edge sealing plate appearance defects;
7. the detection precision, recall rate and F1 value of the model in the edge banding plate appearance defect detection task are respectively 96.3%, 95.7% and 95.9%, and the average detection time is 122.3ms. In contrast, detection accuracy, recall and F1 values in the same task using the conventional YOLOv7 model were 83.8%, 82.2% and 82.7%, respectively, with an average detection time of 137.1ms. The application greatly improves the detection precision.
The above embodiments are only for illustrating the technical solution of the present application, and not for limiting the same; although the application has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present application, and are intended to be included in the scope of the present application.
Claims (10)
1. The edge banding board appearance defect detection method based on the improved YOLOv7 network model is characterized by comprising the following steps of:
1) Defining defect types
The appearance defect types of the edge sealing plate are divided into: opening glue, shortage, collapse, uneven trimming, glue lines and edge sealing indentation;
2) Sample preparation
Collecting a plurality of edge sealing pictures in the edge sealing plate production process, and performing data processing and analysis to obtain a sample to be trained, wherein the data processing and analysis comprises cutting, data enhancement and marking processes;
3) Network model training
Inputting the sample to be trained into an improved YOLOv7 network model for training, so that the sample to be trained learns to identify the appearance defects of the edge sealing plate, and a trained network is obtained;
the improved YOLOv7 network model adds an attention mechanism after each convolution of a backlight layer of YOLOv7, wherein the attention mechanism comprises a channel attention module and a space attention module;
4) Defect detection
Acquiring an image of the edge sealing plate to be detected in real time, and processing, detecting and analyzing the image of the edge sealing plate to be detected by utilizing the network trained in the step 3);
5) Result output
Outputting the defect type of the edge sealing plate to be detected.
2. The edge banding panel appearance defect detection method based on the improved YOLOv7 network model of claim 1, wherein in step 3), the channel attention module weights a feature map of each channel as a function of:
f c (x)＝σ(MLP c (AvePool(x))+MLP c (MaxPool(x)))
the spatial attention module weights the feature map of each spatial location as a function of:
f s (x)＝σ(MLP s (AvePool(x))+MLP s (MaxPool(x)))
the function of the attentiveness mechanism is:
f(x)＝f c (f s (x)·x)·x
where x represents the input feature map, avePool (x) and MaxPool (x) represent the operations of average pooling and maximum pooling of x, MLP_ { c } and MLP_ { s } each represent a fully connected layer, σ represents a sigmoid function, fc { c } represents the function of the channel attention module, and fs { s } represents the function of the spatial attention module, respectively.
3. The edge banding panel appearance defect detection method based on the improved YOLOv7 network model of claim 2, wherein in step 2), the clipping is: the picture is cropped and extraneous portions are removed so that the picture size is uniform to 640 x 640 pixels.
4. The improved YOLOv7 network model-based edge banding pattern inspection method of claim 2, wherein in step 2), the data enhancement includes one or more of rotation, translation, and scaling.
5. The edge banding panel appearance defect detection method based on the improved YOLOv7 network model of claim 2, wherein in step 2), the marking comprises: and marking the pictures by using LabelImg target marking software, wherein each picture contains one or more defects, and the marking content contains defect types, position coordinates and defect ranges.
6. The edge banding plate appearance defect detection method based on the improved YOLOv7 network model of claim 2, wherein in step 4), an industrial CCD camera is used to acquire images of the edge banding plate to be detected in the production process in real time.
7. The edge banding board appearance defect detection method based on the improved YOLOv7 network model of claim 2, wherein the YOLOv7 network model comprises an input layer, a back box layer and a back & head layer, the processing method is to preprocess an input picture, divide the input picture into 640 x 3 pictures, input the processed picture into the back box layer, the back & head layer continuously processes the picture output by the back box layer and outputs feature maps with different sizes, and finally the image is analyzed through a RepVGG convolution architecture to obtain a final result.
8. The edge banding pattern defect detection method based on the improved YOLOv7 network model according to claim 2, wherein in step 3), a Pytorch deep learning framework is used for training the network model, specifically: the size of the input picture is 640 x 640 pixels, the initial learning rate is 0.001, the batch size is 20, the total iteration number is 500, the weight attenuation coefficient is 0.001, the optimization function and the activation function are an Adam optimizer and a leak ReLU function respectively, when the iteration number reaches 400 times, the learning rate is reduced to 0.0001, and after the iteration number reaches 500 times, the model tends to be stable.
9. A banding panel appearance defect detection system based on the improved YOLOv7 network model of any one of claims 1-8, comprising:
the image acquisition unit is used for acquiring the image of the edge sealing plate to be detected in real time;
the defect detection unit is used for processing, detecting and analyzing the image by adopting a trained network;
and the result output unit is used for outputting the defect types of the edge sealing plate to be tested.
10. The edge banding panel appearance defect detection system of claim 9, wherein a backbox layer of the YOLOv7 network model adopts a dark net53 network structure, a neg layer adopts an SPP structure, and a head layer adopts a YOLOv7 structure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310679878.1A CN117173444A (en) | 2023-06-08 | 2023-06-08 | Edge banding board appearance defect detection method and system based on improved YOLOv7 network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310679878.1A CN117173444A (en) | 2023-06-08 | 2023-06-08 | Edge banding board appearance defect detection method and system based on improved YOLOv7 network model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117173444A true CN117173444A (en) | 2023-12-05 |
Family
ID=88945676
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310679878.1A Pending CN117173444A (en) | 2023-06-08 | 2023-06-08 | Edge banding board appearance defect detection method and system based on improved YOLOv7 network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117173444A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115294109A (en) * | 2022-09-30 | 2022-11-04 | 南通恒祥木业有限公司 | Real wood board production defect identification system based on artificial intelligence, and electronic equipment |
| CN115880558A (en) * | 2023-03-03 | 2023-03-31 | 北京市农林科学院信息技术研究中心 | Farming behavior detection method and device, electronic equipment and storage medium |
| CN116206185A (en) * | 2023-02-27 | 2023-06-02 | 山东浪潮科学研究院有限公司 | Lightweight small target detection method based on improved YOLOv7 |
-
2023
- 2023-06-08 CN CN202310679878.1A patent/CN117173444A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115294109A (en) * | 2022-09-30 | 2022-11-04 | 南通恒祥木业有限公司 | Real wood board production defect identification system based on artificial intelligence, and electronic equipment |
| CN116206185A (en) * | 2023-02-27 | 2023-06-02 | 山东浪潮科学研究院有限公司 | Lightweight small target detection method based on improved YOLOv7 |
| CN115880558A (en) * | 2023-03-03 | 2023-03-31 | 北京市农林科学院信息技术研究中心 | Farming behavior detection method and device, electronic equipment and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 邓成: "基于深度学习的板材封边缺陷检测系统和方法", 中国优秀硕士学位论文全文数据库 工程科技Ⅰ辑, 15 January 2023 (2023-01-15), pages 13 - 21 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111553929B (en) | Mobile phone screen defect segmentation method, device and equipment based on converged network | |
| CN106875381B (en) | Mobile phone shell defect detection method based on deep learning | |
| WO2022236876A1 (en) | Cellophane defect recognition method, system and apparatus, and storage medium | |
| CN109767422A (en) | Pipe detection recognition methods, storage medium and robot based on deep learning | |
| CN112381175A (en) | Circuit board identification and analysis method based on image processing | |
| CN111696079B (en) | Surface defect detection method based on multitask learning | |
| WO2024002187A1 (en) | Defect detection method, defect detection device, and storage medium | |
| CN114066848B (en) | FPCA appearance defect visual detection system | |
| CN111783616B (en) | Nondestructive testing method based on data-driven self-learning | |
| CN111461121A (en) | Electric meter number identification method based on YO L OV3 network | |
| CN110020691A (en) | LCD screen defect inspection method based on the training of convolutional neural networks confrontation type | |
| CN116071315A (en) | Product visual defect detection method and system based on machine vision | |
| CN113506239A (en) | Strip steel surface defect detection method based on cross-stage local network | |
| CN111507249A (en) | Transformer substation nest identification method based on target detection | |
| CN116721104A (en) | Live three-dimensional model defect detection method and device, electronic equipment and storage medium | |
| CN118329910A (en) | Printing paper defect detection system and method based on machine vision | |
| CN112381794B (en) | Printing defect detection method based on deep convolution generation network | |
| CN113327243A (en) | PAD light guide plate defect visualization detection method based on AYOLOv3-Tiny new framework | |
| CN117173444A (en) | Edge banding board appearance defect detection method and system based on improved YOLOv7 network model | |
| CN117314829A (en) | Industrial part quality inspection method and system based on computer vision | |
| CN116123040A (en) | Fan blade state detection method and system based on multi-mode data fusion | |
| CN115543820A (en) | Tesseract-orc-based software automated testing method and system | |
| CN113592850A (en) | Defect detection method and device based on meta-learning | |
| CN114299323A (en) | Printed matter defect detection method combining machine vision and deep learning technology | |
| CN116542931A (en) | Training method of image processing model, image processing method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |