Adhesive tape edge defect detection method, electronic equipment, storage medium and device

Abstract

The invention relates to the technical field of printers, in particular to an adhesive tape edge defect detection method, a storage medium, electronic equipment and device.The adhesive tape edge defect detection method comprises the steps that local images of edge faces of N adhesive tapes are obtained, an image A is obtained, the image A comprises images B of the N adhesive tapes in the same area, and N is an integer larger than or equal to 2; comparing and analyzing the N images B, and determining whether a defective image B1 exists or not; and if yes, inputting the image B1 into a defect identification model for identification to obtain a defect type. The automatic image acquisition and processing technology is adopted, so that the requirement of manual visual inspection is reduced, the detection time is greatly shortened, and the detection efficiency is improved. And the defect identification model is used, so that various types of defects can be identified more accurately, and the detection precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of printers, and in particular to a method, electronic equipment, storage medium and device for detecting edge defects of a rubber strip. Background Art

[0002] In modern office automation equipment, the drum cleaning system of printers and copiers plays a vital role. The scraper rubber strip in the drum cleaning system is one of the key components. Its main function is to remove the residual toner on the drum surface through friction to ensure the quality of printing or copying. Therefore, it is required to have high wear resistance (to ensure long-term and stable friction with the drum surface), good elasticity (to enable it to generate sufficient pressure when contacting the drum to effectively remove the residual toner without causing damage to the drum) and precise size (to ensure perfect coordination with the drum and other components). However, due to the complexity and uncertainty of production during the production process, the existing scraper rubber strips are prone to defects such as knife marks, knife lines, ripples, and notches on the edges. These defects will directly affect the cleaning effect and service life of the scraper film.

[0003] At present, the commonly used method for detecting defects on the edges of rubber strips in the industry is manual visual inspection. This method relies on the experience and vision of the operator, has low detection efficiency, and is prone to missed detection and false detection. Especially in mass production environments, manual visual inspection is not only time-consuming, but also costly, and cannot meet the needs of modern production. In addition, manual visual inspection has the following problems: 1. Limited detection accuracy: Human vision is limited, and it is difficult to accurately identify tiny defects. 2. Poor consistency: The detection standards and experience of different operators vary greatly, resulting in inconsistent detection results. 3. High labor intensity: Long-term visual inspection can easily cause operator fatigue, further reducing detection efficiency and accuracy.

[0004] Therefore, how to improve the efficiency of defect detection on the edges of rubber strips and realize automated and high-precision detection is a technical problem that technicians currently need to solve. Summary of the invention

[0005] The present invention provides a method, electronic equipment, storage medium and device for detecting edge defects of rubber strips. The method is used to solve the technical problem of low detection efficiency caused by the detection method of rubber strip edge defects by manual visual inspection in the prior art.

[0006] A first aspect of the present invention provides a method for detecting edge defects of a rubber strip, comprising:

[0007] S1: Obtaining local images of the edge surfaces of N rubber strips to obtain image A, where image A includes image B of the same area of ​​the N rubber strips, where N is an integer greater than or equal to 2;

[0008] S2: Compare and analyze the N images B to determine whether there is a defective image B1;

[0009] S3: If it exists, input the image B1 into the defect recognition model for recognition to obtain the defect type. The training samples of the defect recognition model include one or more of the following annotated images: knife marks, knife lines, ripples, and gaps.

[0010] In a first possible method for detecting edge defects of a rubber strip according to the first aspect, S2 includes:

[0011] S21: Segment image A to obtain N images B;

[0012] S22: Compare and analyze any two images B to obtain a difference K;

[0013] S23: Compare the difference K with the difference threshold K1;

[0014] S24: If K>K1, then compare and analyze the corresponding image B with the standard image to obtain the difference degree K2;

[0015] S25: Compare the difference K2 with the difference threshold K3, where K3 < K1;

[0016] S26: If K2>K3, the corresponding image B is determined to be a defective image B1.

[0017] In combination with the first possible implementation of the rubber strip edge defect detection method of the first aspect, in a second possible implementation of the rubber strip edge defect detection method of the first aspect, after S23, the following steps are further included:

[0018] S231: if K≤K1, then loop through steps S1 to S23 to detect the undetected area of ​​the corresponding rubber strip until the undetected area is cleared;

[0019] S25 and later also include:

[0020] S251: If K2≤K3, then loop through steps S1 to S25 to detect the undetected area of ​​the corresponding rubber strip until the undetected area is cleared.

[0021] In combination with the first possible implementation of the rubber strip edge defect detection method of the first aspect, in a third possible implementation of the rubber strip edge defect detection method of the first aspect, S22 is replaced by:

[0022] Compare and analyze any two images B taken from the same adhesive strip to obtain the difference K. The areas corresponding to the two images B are located on the two edge surfaces of the same adhesive strip.

[0023] In combination with the first possible implementation of the rubber strip edge defect detection method of the first aspect, in a fourth possible implementation of the rubber strip edge defect detection method of the first aspect, after S1 and before S21, the following steps are further included:

[0024] S201: Compare and analyze any two images A to obtain a difference k;

[0025] S202: Subtract k from the difference threshold k1. If k>k1, execute S21.

[0026] In a fifth possible method for detecting rubber strip edge defects according to the first aspect, the defect recognition model also has a defect grading function.

[0027] In a sixth possible method for detecting edge defects of a rubber strip according to the first aspect, before S1, the method further includes:

[0028] S0: Arrange N rubber strips in such a way that their edges are coplanar and flush.

[0029] A second aspect of the present invention provides an electronic device, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any possible method for detecting edge defects of rubber strips provided in the first aspect.

[0030] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any possible method for detecting edge defects of rubber strips provided in the first aspect.

[0031] A fourth aspect of the present invention provides a device for detecting edge defects of a rubber strip, comprising:

[0032] An image acquisition component, used for acquiring images of film edges;

[0033] Image analysis component, used to analyze and process images.

[0034] It can be seen from the above technical solutions that the present invention has the following advantages:

[0035] ① The rubber strip edge defect detection method provided by the present invention obtains a local image of the edge surface of N rubber strips to obtain image A, which includes image B of the same area of ​​the N rubber strips, where N is an integer greater than or equal to 2; N images B are compared and analyzed to determine whether there is a defective image B1; if so, image B1 is input into a defect recognition model for identification to obtain the defect type. The use of automated image acquisition and processing technology reduces the need for manual visual inspection, greatly shortens the detection time, and improves the detection efficiency. In addition, the use of a defect recognition model can more accurately identify multiple types of defects, thereby improving the accuracy of detection.

[0036] ② By comparing and analyzing multiple images first to determine whether there are defective images, only images with defects will be input into the defect recognition model. This method avoids the need for comprehensive and in-depth analysis of each image as in traditional methods, especially for a large number of defect-free images, without the need for complex defect recognition operations, greatly reducing the amount of calculation and analysis time, and only processing the defective parts, significantly improving the detection efficiency and meeting the needs of high-speed production.

[0037] ③ From obtaining partial images of the edges of multiple rubber strips to comparative analysis to determine whether there are defects, and then to identifying defective images, the entire process is more efficient and orderly. Compared with the traditional manual inspection method, this automated hierarchical inspection process reduces redundant steps and improves the overall fluency and speed of inspection. In an automated production line, this efficient inspection process can better match the production rhythm and will not affect the operating efficiency of the entire production line due to delays in the inspection link.

[0038] ④First, determine the images that may have defects through comparative analysis, so that the images input into the defect recognition model are all potential problems. This helps the defect recognition model focus more on the features that may have defects, reduce interference factors, and improve the accuracy of defect recognition. Clearly distinguish between defective and non-defective products. Non-defective products can directly enter the next process, and for defective products, accurately identify the defect type to facilitate the adoption of corresponding treatment measures. This targeted treatment method not only improves the detection efficiency, but also can more specifically solve quality problems and improve the level of product quality control. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 A schematic flow chart of a method for detecting edge defects of a rubber strip provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The embodiments of the present invention provide a method, electronic device, storage medium and device for detecting edge defects of rubber strips, which are used to solve the technical problem that the detection method of the edge defects of rubber strips by manual visual inspection in the prior art leads to low detection efficiency.

[0042] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] In the description of the embodiments of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, which are only for the convenience of describing the embodiments of the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the embodiments of the present application. In addition, the terms "first", "second", and "third" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0044] In the description of the embodiments of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a replaceable connection, or an integral connection, it can be a mechanical connection, it can be an electrical connection, it can be a direct connection, it can be indirectly connected through an intermediate medium, and it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the embodiments of the present application can be understood according to specific circumstances.

[0045] At present, the commonly used method for detecting defects on the edges of rubber strips in the industry is manual visual inspection. This method relies on the experience and vision of the operator, has low detection efficiency, and is prone to missed detection and false detection. Especially in mass production environments, manual visual inspection is not only time-consuming, but also costly, and cannot meet the needs of modern production. In addition, manual visual inspection has the following problems: 1. Limited detection accuracy: Human vision is limited, and it is difficult to accurately identify tiny defects. 2. Poor consistency: The detection standards and experience of different operators vary greatly, resulting in inconsistent detection results. 3. High labor intensity: Long-term visual inspection can easily cause operator fatigue, further reducing detection efficiency and accuracy.

[0046] Embodiment 1

[0047] See also Figure 1 The method for detecting edge defects of rubber strips provided by the embodiment of the present invention includes:

[0048] S1: Obtaining local images of the edge surfaces of N rubber strips to obtain image A, where image A includes image B of the same area of ​​the N rubber strips, where N is an integer greater than or equal to 2;

[0049] Specifically, the function of this step is to collect data. Arrange N rubber strips to be tested at a fixed spacing and position to ensure that the same local area of ​​each rubber strip (such as a specific length or specific area of ​​the edge surface) can be photographed at the same time, and ensure that the edge surface of each rubber strip is clean and free of foreign objects. Use an industrial camera or other high-resolution image acquisition equipment for image acquisition. The camera's field of view should be large enough to cover the same local area of ​​all rubber strips, and should have sufficient resolution to capture tiny defects, such as knife marks, knife marks, ripples and gaps, and ensure that the shooting angle and distance are consistent to ensure image consistency and comparability. Take a picture of the same local area of ​​the arranged N rubber strips once to obtain an image A containing the local areas of all rubber strips. The image A obtained by shooting is an image of the same area of ​​the edge surfaces of multiple rubber strips. The edge surface refers to the surface with a smaller area where the two edges are located on the rectangular rubber strip, or the surface where the edges parallel to the length direction and the edges parallel to the thickness direction are located.

[0050] For example: On a production line, the rubber strips pass through the inspection area at a speed of 100 strips per minute, with 5 strips in a group. The 5 strips are fixed side by side on the inspection platform to ensure that the edges of each strip are on the same horizontal line and the spacing is uniform. The interval between each group of strips is 1 second, and the strips in each group pass through the inspection area at the same time in the same posture. Set the camera to shoot the edges of each group of strips, and each shot contains an image of one-quarter (four equal divisions based on length) of the area of ​​the 5 strips. In this way, we obtain 2 images A per second (each image contains the same area of ​​5 strips).

[0051] Optimally, N rubber strips are arranged in a way that their edges are coplanar and flush, ensuring that the edges of each rubber strip are on the same horizontal line and arranged neatly so that their edges are coplanar and flush, which facilitates subsequent image acquisition.

[0052] For further optimization, N rubber strips are arranged in an interleaved and stacked manner, that is, there is a certain interval between each rubber strip, and they are stacked so that they have a certain interval in the vertical direction, so that the edge surface of the rubber strip appears as a vertical surface. When the rubber strip flows along its own length direction (front and back direction) on the production line, the two edge surfaces of the rubber strip appear as the left surface and the right surface respectively, so that images of the two edge surfaces can be collected at the same time, thereby improving the image collection efficiency, thereby further improving the defect detection efficiency. In addition, in order to avoid mutual influence between the image collection on both sides, the image collection devices on both sides can be designed so that the spacing in the front and back direction is greater than or equal to an image collection spacing.

[0053] S2: Compare and analyze the N images B to determine whether there is a defective image B1;

[0054] Specifically, when there are no defects on the edge of the rubber strip, the images B corresponding to different rubber strips in image A are highly consistent, that is, the difference between the images B in image A is very small, or it can be said that they are basically the same. By comparing the local images corresponding to multiple local areas representing the rubber strip in image A, the local image that may have defects - image B1 is found. Through comparative analysis, images that may have defects are preliminarily screened out, and images without defects are quickly excluded. Without this step, all images will be input into the defect recognition model, increasing unnecessary calculations.

[0055] Exemplarily, the five partial images B in the image A are then compared with each other, and the image B with obvious difference is defined as image B1.

[0056] S3: If it exists, input the image B1 into the defect recognition model for recognition to obtain the defect type. The training samples of the defect recognition model include one or more of the following annotated images: knife marks, knife lines, ripples, and gaps.

[0057] Specifically, in the previous step S2, images B1 with large differences have been determined through comparative analysis. These images B1 are preliminarily determined to be images that may have defects. In order to conduct a detailed analysis of the defective images preliminarily screened out and determine the specific defect type, the selected images B1 are input into a pre-trained defect recognition model for defect type recognition. Commonly used models include convolutional neural networks (CNN), support vector machines (SVM), etc. The training samples of the model should include various types of defect images, such as knife marks, knife marks, ripples, and gaps. There should be a sufficient number of samples for each defect type to ensure the recognition accuracy of the model. The model will classify the input image B1 and output the defect type and its confidence. The reliability of the recognition result is judged based on the output results and confidence of the model. Usually a confidence threshold (such as 80%) is set, and the recognition result is considered valid only when the confidence exceeds the threshold. If the confidence of the model is lower than the set threshold, manual review or further processing may be required.

[0058] Exemplarily, five defective images B1 are input into a pre-trained CNN model.

[0059] Defect classification: The model output is:

[0060] Image B1-1: Knife marks, confidence level 85%

[0061] Image B1-2: Ripple, 90% confidence

[0062] Image B1-3: Gap, 80% confidence

[0063] Image B1-4: Knife marks, 95% confidence

[0064] Image B1-5: Knife marks, confidence level 75%

[0065] Result output: According to the output results of the model, 5 rubber strips were identified to have knife marks, ripples, gaps, knife marks and knife marks respectively.

[0066] The beneficial effects of this embodiment include:

[0067] ① The use of automated image acquisition and processing technology reduces the need for manual visual inspection, greatly shortens the inspection time, and improves inspection efficiency. The use of defect recognition models can more accurately identify multiple types of defects and improve the accuracy of inspection.

[0068] ② By comparing and analyzing multiple images first to determine whether there are defective images, only images with defects will be input into the defect recognition model. This method avoids the need for comprehensive and in-depth analysis of each image as in traditional methods, especially for a large number of defect-free images, without the need for complex defect recognition operations, greatly reducing the amount of calculation and analysis time, and only processing the defective parts, significantly improving the detection efficiency and meeting the needs of high-speed production.

[0069] ③ From obtaining partial images of the edges of multiple rubber strips to comparative analysis to determine whether there are defects, and then to identifying defective images, the entire process is more efficient and orderly. Compared with the traditional manual inspection method, this automated hierarchical inspection process reduces redundant steps and improves the overall fluency and speed of inspection. In an automated production line, this efficient inspection process can better match the production rhythm and will not affect the operating efficiency of the entire production line due to delays in the inspection link.

[0070] ④First, determine the images that may have defects through comparative analysis, so that the images input into the defect recognition model are all potential problems. This helps the defect recognition model focus more on the features that may have defects, reduce interference factors, and improve the accuracy of defect recognition. Clearly distinguish between defective and non-defective products. Non-defective products can directly enter the next process, and for defective products, accurately identify the defect type to facilitate the adoption of corresponding treatment measures. This targeted treatment method not only improves the detection efficiency, but also can more specifically solve quality problems and improve the level of product quality control.

[0071] ⑤ By collecting images of multiple rubber strips at one time for defect analysis, the image collection efficiency is improved compared to collecting each rubber strip independently, which can improve the overall defect recognition efficiency.

[0072] Specifically, step S2 includes the following steps:

[0073] S21: Segment image A to obtain N images B;

[0074] Specifically, an image B corresponding to each rubber strip is segmented from image A to obtain N images B. Each image B corresponds to a local area of ​​a rubber strip. The specific segmentation method can be implemented in the following ways: Coordinate-based segmentation: Determine the position of each rubber strip in image A and segment it based on the coordinate relationship between the arrangement position of the rubber strip and the camera's field of view. Image processing-based segmentation: Use image processing technology (such as edge detection and threshold segmentation) to automatically identify the boundary of each rubber strip in image A and segment it. After the segmentation is completed, each image B is preprocessed, including denoising, grayscale, and image normalization. The purpose of this step is to reduce the impact of image noise and illumination changes on the comparison results, and ensure that subsequent comparative analysis is more accurate. Exemplarily, image A is segmented into 5 images B through coordinate calculation or image processing technology.

[0075] S22: Compare and analyze any two images B to obtain a difference K;

[0076] Specifically, any two images B are selected and their difference K is calculated. The difference K can be calculated by pixel-level difference (such as absolute difference, mean square error) or feature-level difference (such as Euclidean distance, cosine similarity) to obtain the difference K.

[0077] Further, step S22 can be optimized as follows: any two images B taken from the same adhesive strip are compared and analyzed to obtain a difference degree K, where the regions corresponding to the two images B are located at two edge surfaces of the same adhesive strip.

[0078] Specifically, as long as there are no defects on the edge of the rubber strip, the images of its two edge surfaces are highly similar, so the local images of its two edge surfaces can be compared with each other, and the purpose of preliminarily screening out products that may have defects can also be achieved. In this step, the two images B are taken from two images A respectively, and the two images A are obtained by shooting the edge surfaces on both sides of the same set of rubber strips. The two images A can be taken from two corresponding areas on the two edge surfaces of the same set of rubber strips, or from two non-corresponding areas on the two edge surfaces of the same set of rubber strips, as long as the two images B are taken from the two edge surfaces of the same rubber strip, and the size and shape are the same. After obtaining the two images B, calculate their difference K. The difference K can be calculated by pixel-level difference (such as absolute difference, mean square error) or feature-level difference (such as Euclidean distance, cosine similarity) to obtain the difference K.

[0079] S23: Compare the difference K with the difference threshold K1;

[0080] Specifically, a difference threshold K1 is set. If the difference between any two images B is greater than K1, it means that there is a defect in the area of ​​the rubber strip corresponding to one or both images. If the difference is equal to K1, it means that there is no defect in the area of ​​the rubber strip corresponding to the two images. K1 is set based on experience. If the difference K is greater than the threshold K1, the next step S24 is entered; otherwise, it is considered that there is no significant difference between the two images B. It means that both images have no defects.

[0081] For further optimization, the following steps are added after step S23:

[0082] S231: if K≤K1, then loop through steps S1 to S23 to detect the undetected area of ​​the corresponding rubber strip until the undetected area is cleared;

[0083] Specifically, if the difference K is less than or equal to the threshold K1, it is considered that there is no significant difference between the two images B currently being compared, and further detection is required for the undetected area of ​​the adhesive strip. Steps S1 to S23 are re-executed to obtain images, segment, and calculate the difference for the undetected area of ​​the adhesive strip until all areas have been detected.

[0084] S24: If K>K1, then compare and analyze the corresponding image B with the standard image to obtain the difference degree K2;

[0085] Specifically, a standard image is used as a reference, and the standard image is a known defect-free adhesive strip image. The two images B corresponding to K greater than K1 are compared with the standard image of image B respectively, and the difference K2 is calculated. The difference K2 can be calculated by pixel-level difference (such as absolute difference, mean square error) or feature-level difference (such as Euclidean distance, cosine similarity).

[0086] S25: Compare the difference K2 and the difference threshold K3, K3<K1;

[0087] Specifically, a difference threshold K3 is set, because K3 represents the maximum difference allowed between the image and the standard image. If it is greater than K3, it means that there must be a defect, and if it is less than or equal to K3, it means that there is no defect. Therefore, the standard required for K3 is stricter than K1, so its value is smaller than K1. The two K2 obtained in step S24 are compared with K3 respectively. If the difference K2 is greater than the threshold K3, the image B is determined to be a defective image B1.

[0088] By comparing image B in image A with each other, images that may have defects are picked out according to the comparison results, and then the images that may have defects are compared with the standard images to determine the images with defects. Compared with comparing all images B with the standard images at the beginning, this method reduces the number of calculations, thereby reducing the amount of calculations, and can further improve detection efficiency.

[0089] For further optimization, the following steps are added after step S25:

[0090] S251: If K2≤K3, then loop through steps S1 to S25 to detect the undetected area of ​​the corresponding rubber strip until the undetected area is cleared.

[0091] Specifically, if the difference K2 is less than or equal to the threshold K3, it is considered that there is no significant difference between the image B and the standard image, and further detection is required for the undetected area of ​​the adhesive strip. Steps S1 to S25 are re-executed to obtain images, segment, calculate the difference, and compare with the standard image for the undetected area of ​​the adhesive strip until all areas have been detected.

[0092] S26: If K2>K3, the corresponding image B is determined to be a defective image B1.

[0093] Specifically, if the difference K2 exceeds the threshold K3, it is considered that the image B has defects and is determined to be a defective image B1, that is, the edge surface of the rubber strip corresponding to the image B1 has defects.

[0094] Furthermore, based on the same principle as S22, before comparing the images B, the following steps may be added before step S21:

[0095] S201: Compare and analyze any two images A to obtain a difference k;

[0096] S202: Subtract k from the difference threshold k1. If k>k1, execute S21.

[0097] Specifically, any two images A taken from the same set of rubber strips are highly similar in the absence of defects. Any two images A taken from the same set of rubber strips are compared and analyzed to obtain a difference k, and k is compared with the difference k1. When k>k1, it means that the similarity of the two images A is low and there may be defects. Then, step S21 is performed for further defect analysis. Before that, the image A that may have defects can be compared and analyzed with the standard image of image A (similar to step S24); when k≤k1, it means that the similarity of the two images A is very high, and there is no defect in the area of ​​the rubber strip edge surface corresponding to the two images A. The two images A can be two images taken successively from one edge surface of the same set of rubber strips, or two images taken from two edge surfaces of the same set of rubber strips.

[0098] Specifically, in order to design a model that can identify the edge defects of rubber strips and classify and grade them according to the defect type and degree, the following is a detailed model design scheme. This scheme combines the convolutional neural network (CNN) architecture and target detection algorithm in deep learning.

[0099] 1. Input layer: The input layer is used to receive the image to be inspected and preprocess it to obtain the preprocessed image. The preprocessing includes resizing, normalization and data enhancement. More specifically, the processing includes: resizing, adjusting the image to be inspected to the uniform size required by the model, and then normalizing, normalizing the pixel values ​​of the resized image to the range of [0,1] to accelerate training and improve model stability, and then performing data enhancement, applying random flipping, rotation, scaling and other data enhancement techniques to improve the generalization ability of the model.

[0100] 2. Feature extraction layer: extract multi-scale features from the preprocessed image and generate feature maps of different scales. Specifically, the YOLOv5s network architecture is used, including the focus layer, convolution block 1, convolution block 2, convolution block 3, CSPNet block 1 and CSPNet block 2. YOLOv5s is a lightweight target detection model, and its network architecture includes the following parts: Backbone: feature extraction network, responsible for extracting image features, using the CSPDarknet53 architecture; Neck: feature fusion network, responsible for fusing feature maps of different scales, using FPN and PAN architecture; Head: target detection and classification network, responsible for outputting detection results, including bounding boxes, confidence levels and category probabilities. The present invention uses YOLOv5s as a defect detection model, and its feature extraction layer corresponds to the Backbone part, as follows:

[0101] Focus Layer

[0102] The Focus layer is a unique layer in YOLOv5s, which is used to increase the number of channels in the initial stage of the network while retaining spatial information. The core idea of ​​the Focus layer is to increase the number of channels of the input image from 3 (RGB) to a higher dimension (such as 32) through slicing operations, thereby providing more feature information for subsequent convolution operations.

[0103] Focus layer workflow:

[0104] 1. Slicing operation: The input image is sliced ​​into multiple small blocks, each of which contains multiple adjacent pixels.

[0105] 2. Channel expansion: The sliced ​​image blocks are rearranged to form new channels. For example, a 3x3 slicing operation can expand a 3-channel RGB image to 32 channels.

[0106] 3. Convolution operation: The expanded feature map is processed through a convolution layer to extract higher-level features.

[0107] Conv Block 1

[0108] It contains multiple convolutional layers (Conv), batch normalization layers (Batch Normalization, BN) and activation functions (such as Leaky ReLU) to extract preliminary features of the image.

[0109] CSPNet Block 1

[0110] The CSPNet structure is used to divide the feature map into two parts, one part undergoes convolution operation, and the other part is directly concatenated with the convolved feature map, which enhances the feature representation capability and reduces the amount of calculation.

[0111] CSPNet workflow:

[0112] 1. Feature map segmentation: The input feature map is divided into two parts.

[0113] 2. Convolution operation: A portion of the feature map is processed through a series of convolutional layers.

[0114] 3. Feature fusion: The processed feature map is concatenated with the unprocessed feature map to form a new feature map.

[0115] 4. Repeated operation: The above steps are repeated in multiple stages to form the complete structure of CSPNet.

[0116] Conv Block 2 and CSPNet Block 2 repeat the above processes of Conv Block 1 and CSPNet Block 1 respectively to further extract more advanced features.

[0117] Conv Block 3

[0118] Through deeper convolution operations, deeper features of the image are extracted and multi-scale feature maps are generated, such as P3, P4, P5, P6, etc. Among them:

[0119] -P3: The feature map size is 1 / 8 of the input image, with less semantic information but rich spatial information;

[0120] -P4: The feature map size is 1 / 16 of the input image, and the semantic information is richer;

[0121] -P5: The feature map size is 1 / 32 of the input image, and the semantic information is rich;

[0122] -P6: The feature map size is 1 / 64 of the input image, with the richest semantic information but less spatial information.

[0123] These multi-scale feature maps are used in the subsequent feature fusion and enhancement layers. Under the action of FPN and PAN, feature maps of different scales are fused to generate fused feature maps for target detection and classification.

[0124] 3. Feature fusion and enhancement layer: Feature pyramid network (FPN) and path aggregation network (PAN) are used to fuse feature maps of different scales to obtain fused feature maps, where: FPN fuses high-level semantic information with low-level spatial information through a top-down path; PAN introduces a bottom-up path based on FPN to form a bidirectional connection with FPN, further enhancing the expressiveness of feature maps and the efficiency of information transmission. Its main function is to enhance the expressiveness of feature maps and improve the model's ability to detect targets of different sizes. The following is a detailed analysis of the specific functions, workflows, and relationships between FPN and PAN:

[0125] FPN (Feature Pyramid Network)

[0126] The main function of FPN is to establish lateral connections between feature maps of different scales, thereby fusing feature information at different levels. FPN combines high-level semantic information with low-level spatial information through a top-down path, allowing the model to simultaneously use features of different scales for target detection.

[0127] FPN workflow:

[0128] 1. Multi-scale feature extraction: In the feature extraction layer (Backbone), after the image passes through multiple convolutional layers, feature maps of different scales will be generated. For example, YOLOv5s will generate feature maps of multiple scales such as P3, P4, P5, and P6.

[0129] 2. Top-down path: Starting from the feature map of the highest layer, its size is enlarged to the same size as the feature map of the next layer through upsampling operations (such as bilinear interpolation).

[0130] 3. Horizontal connection: perform element-wise addition on the upsampled feature map and the feature map of the next layer to fuse feature information of different scales.

[0131] 4. Repeated operation: The above steps are repeated at multiple levels to form a pyramid-shaped feature graph structure.

[0132] Advantages of FPN:

[0133] Multi-scale feature fusion: By fusing feature maps of different scales, the model can better detect objects of different sizes. Enhance semantic information: High-level feature maps contain richer semantic information. Through a top-down path, this information is passed to low-level feature maps, enhancing the expressiveness of feature maps. Improve detection accuracy: FPN can effectively improve the accuracy of target detection, especially when dealing with multi-scale targets.

[0134] PAN (Path Aggregation Network)

[0135] PAN further enhances the transmission path of feature maps based on FPN. PAN introduces a bottom-up path to form a bidirectional connection with the top-down path of FPN, thereby further enhancing the expressiveness of feature maps and the efficiency of information transmission.

[0136] PAN workflow:

[0137] 1. Bottom-up path: After the top-down path of FPN is completed, PAN introduces a bottom-up path to pass the low-level feature map to the high-level feature map through convolution and upsampling operations.

[0138] 2. Feature fusion: The bottom-up feature map is fused with the top-down feature map to form a richer feature representation.

[0139] 3. Repeated operation: The above steps are repeated at multiple levels to form a bidirectional feature transfer path.

[0140] Advantages of PAN:

[0141] Bidirectional feature transfer: Through the bidirectional path from top to bottom and bottom to top, PAN can transfer and fuse feature information more effectively. Enhanced feature expression: The bidirectional path enables the feature map to retain more detailed information and semantic information during the transfer process, improving the expressiveness of the feature map. Improved detection performance: PAN can further improve the performance of target detection, especially when dealing with complex scenes and multi-scale targets.

[0142] Relationship between FPN and PAN

[0143] In the YOLOv5s model, FPN and PAN together constitute the feature fusion and enhancement layer (Neck), and their workflow is as follows:

[0144] 1.FPN processing: After the image passes through the feature extraction layer (Backbone), feature maps of multiple scales are generated. FPN transfers high-level semantic information to low-level feature maps through a top-down path for feature fusion.

[0145] 2. PAN processing: After the top-down path of FPN is completed, PAN introduces a bottom-up path to transfer the low-level feature maps to the high-level feature maps, further enhancing the expressiveness of the feature maps and the efficiency of information transfer.

[0146] 4. Object detection and classification layer: Detect and classify objects on the fused feature map, and output the bounding box, confidence, and category probability of the defect. The object detection and classification layer (Head) is the core part of the entire defect detection model, responsible for the final output of the detection results, that is, identifying and classifying defects in the image. The main functions of this layer include bounding box regression, confidence prediction, and category probability prediction. In addition, in order to optimize the model training process, the Head layer also designs a multi-task loss function for comprehensively evaluating the performance of the model on various tasks. The following is a detailed analysis of the Head layer:

[0147] 1. Detection Head

[0148] The detection head is responsible for extracting and outputting the final detection results from the feature map. Specifically, it contains three main tasks:

[0149] ① Bounding Box Regression:

[0150] Function: Predict the bounding box position of each defect, that is, the specific location of the defect in the image.

[0151] Output: The position parameters of the bounding box, including the center coordinates (x, y) and the width (w) and height (h).

[0152] Working process:

[0153] Feature extraction: Extract features related to target detection from the feature maps output by the feature fusion and enhancement layers (FPN and PAN).

[0154] Regression calculation: Through the convolution layer and the fully connected layer, the position parameters (x, y, w, h) of each bounding box are calculated.

[0155] Output: Generate bounding box coordinates of each defect for subsequent localization and visualization.

[0156] ②Confidence Prediction:

[0157] Role: Predict whether there is a defect in each bounding box and the category probability of the defect.

[0158] Output: A confidence score for each bounding box, indicating the probability of a defect existing within the box, and the specific category of the defect.

[0159] Working process:

[0160] Feature Map: Extract category-related features from the feature map.

[0161] Confidence calculation: Through the convolutional layer and the fully connected layer, the confidence score of each bounding box is calculated.

[0162] Category probability calculation: For each bounding box, calculate the probability that it belongs to each defect category (such as knife marks, knife patterns, ripples, and gaps, etc.).

[0163] Output: Generates confidence scores and class probabilities for each bounding box.

[0164] ③Class Probability:

[0165] Function: Predict the specific type of defects (such as knife marks, knife marks, ripples, and notches, etc.).

[0166] Output: The defect category corresponding to each bounding box.

[0167] Working process:

[0168] Category mapping: Map the feature map to the category space.

[0169] Probability calculation: Calculate the probability corresponding to each category through the softmax function.

[0170] Output: Generates the final class label for each bounding box.

[0171] 2. Multi-task Loss Function

[0172] In order to optimize the performance of the model in the three tasks of bounding box regression, confidence prediction, and category classification, the Head layer uses a multi-task loss function, that is, calculating the losses of multiple tasks at the same time and combining them for model optimization. Specifically, the multi-task loss function consists of the following three parts:

[0173] ① Bounding Box Loss:

[0174] Loss function: Use CIoU (Complete IoU) loss function.

[0175] Function: measure the difference between the predicted bounding box and the true bounding box to improve the accuracy of bounding box positioning.

[0176] CIoU advantage: CIoU not only considers the overlapping area of ​​the bounding boxes, but also the center point distance and aspect ratio of the bounding boxes, which can more accurately reflect the similarity of the bounding boxes.

[0177] ②Confidence Loss:

[0178] Loss function: Use Binary Cross Entropy Loss.

[0179] Function: Measures the model's accuracy in predicting the presence or absence of defects.

[0180] Working process: Calculate the cross entropy loss between the predicted confidence and the true label.

[0181] ③Classification Loss:

[0182] Loss function: Use the cross entropy loss function.

[0183] Function: Measures the model's prediction accuracy for defect categories.

[0184] Working process: Calculate the cross entropy loss between the predicted category probability and the true category label.

[0185] Combination of multi-task loss functions:

[0186] The bounding box loss, confidence loss and category loss are combined by weighted summation to form the final total loss function. Total loss function = α*bounding box loss + β*confidence loss + γ*category loss, where α, β, γ are weight coefficients used to balance the loss contributions of different tasks.

[0187] 5. Defect classification layer: classify and classify the detected defects according to the bounding box, confidence and category probability, and output the detection results. The defect classification layer is the final processing stage of the defect detection model, which is responsible for classifying and classifying the detected defects. The main functions of this layer include post-processing steps and classification strategies. The post-processing steps are used to filter and optimize the detection results, while the classification strategy further classifies and classifies the same type of defects according to the characteristics of the defects (such as confidence, size, shape, etc.). The following is a detailed analysis of this layer:

[0188] Post-processing steps:

[0189] The purpose of the post-processing step is to optimize the detection results and improve the accuracy and robustness of the model. It mainly includes the following steps:

[0190] 1. Confidence Thresholding:

[0191] Function: Filter out the detection results with lower confidence and only keep those defect detections with higher confidence.

[0192] Working process:

[0193] Set a threshold: Based on experience or experimental results, set a confidence threshold (for example, 0.5).

[0194] Filtering operation: Compare the confidence score of each detected defect with the threshold. If the confidence score is higher than the threshold, the detection result is retained. If the confidence score is lower than the threshold, the detection result is discarded. By setting the confidence threshold, false detections and missed detections can be effectively reduced, and the detection accuracy of the model can be improved.

[0195] 2. Non-Maximum Suppression (NMS):

[0196] Function: Eliminate overlapping bounding boxes, retain the most likely detection results, and avoid duplicate detection.

[0197] Working process:

[0198] Sorting: Sort all detected defects according to their confidence scores.

[0199] Select Highest Confidence: Select the defect with the highest confidence as the candidate result.

[0200] Suppress overlap: Calculate the overlap between other detection results and candidate results (usually using IoU, Intersection over Union). If the overlap is higher than the set threshold (for example, 0.5), suppress those overlapping detection results.

[0201] Repeat: Repeat the above steps for the remaining test results until all test results have been processed.

[0202] NMS can effectively reduce redundant detection results, avoid multiple detections of the same defect, and improve the detection efficiency of the model.

[0203] 3. Grading strategy:

[0204] ①Confidence-based Grading:

[0205] Purpose: Classify defects of the same type into different levels (e.g., high, medium, low) based on the confidence scores of the detection results.

[0206] Working process:

[0207] Set classification thresholds: Based on experience or experimental results, set multiple confidence thresholds (e.g., high level: >0.8, medium level: 0.5-0.8, low level: <0.5).

[0208] Classification operation: Compare the confidence score of each detected defect with the classification threshold and classify it into the corresponding level.

[0209] Through confidence-based grading, the severity of defects can be assessed more intuitively, providing a basis for subsequent processing.

[0210] ②Feature-based Grading based on size or shape:

[0211] Function: If more detailed classification is required, the size, shape and other features of the defects can be extracted and classified using a classifier.

[0212] Working process:

[0213] Feature extraction: Extract features such as size and shape from the bounding box of the detected defect. For example, calculate the area, perimeter, aspect ratio, etc. of the defect.

[0214] Feature Normalization: The extracted features are normalized to have the same scale.

[0215] Classifier training: Train the classifier using supervised learning algorithms (such as support vector machines, random forests, neural networks, etc.) to classify defects into different levels.

[0216] Classification operation: Input the extracted features into the trained classifier and output the level of the defect.

[0217] Feature-based classification can provide more refined grading results and is suitable for application scenarios that require accurate assessment of defect severity.

[0218] 4. Output of classification results:

[0219] Grading label: Each defect is assigned a grading label indicating its level (e.g., high, medium, low).

[0220] Visualization: The grading results can be combined with the original images for visual display, which is convenient for users to view and analyze.

[0221] Embodiment 2

[0222] An embodiment of the present invention provides a device for detecting edge defects of a rubber strip, comprising:

[0223] The positioning component is used to fix the rubber strip at the image acquisition station by precise positioning; the image acquisition component is used to acquire images on the edge surface of the rubber strip; the image analysis component is used to analyze and process the image, more specifically, to perform comparative analysis of the image and run the defect detection model.

[0224] Specifically, the positioning component is usually composed of a positioning table, which can accurately fix the rubber strip to ensure that its position is stable and highly repeatable. Before installation and use, the positioning component needs to be calibrated to ensure that it can accurately fix the rubber strip in a predetermined position. The calibration process may include the use of laser alignment, mechanical adjustment and other methods. Sensors (such as photoelectric sensors, position sensors) can be used to assist in positioning to ensure that the position of the rubber strip is accurate. In this embodiment, the positioning component is 5 support plates stacked at intervals in the vertical direction, and the support plates are movably connected so that when placing the rubber strip, each support plate can expose its upper surface to place the rubber strip, and the positioning component reciprocates on the production line to continuously transport the rubber strip to be inspected through the image acquisition component for inspection.

[0225] The image acquisition component uses a high-resolution industrial camera or other image acquisition device to ensure that the image quality is high enough to capture subtle defects. In order to obtain a clear image, a suitable light source is required. LED lights or other high-brightness light sources can be used, and diffuse reflection panels or ring light sources can be used to evenly illuminate the edge surface.

[0226] The image analysis component performs the steps in any of the methods for detecting defects on the edge of the rubber strip in the first embodiment, that is, comparing and analyzing the image B to obtain the difference K, and at the same time running the defect detection model to perform defect type and defect classification, thereby realizing the detection of whether there are defects, and realizing the classification and classification of qualified products and defective products.

[0227] The rubber strip edge defect detection device provided in this embodiment can significantly improve the efficiency and accuracy of rubber strip edge defect detection, and is suitable for automated detection in actual production environments.

[0228] Embodiment 3

[0229] The embodiment of the present invention also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the method for detecting the edge defects of the rubber strip of the first embodiment is implemented. The computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center that contains one or more available media. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a high-density digital video disc (DVD)), or a semiconductor medium (for example, a solid-state hard disk), etc. The computer-readable storage medium includes instructions that instruct the computing device to execute any one of the methods for detecting the edge defects of the rubber strip provided in the first embodiment.

[0230] Embodiment 4

[0231] An embodiment of the present invention further provides an electronic device, comprising a memory, a processor, that is, a computer program stored in the memory;

[0232] The processor executes the computer program to implement the method for detecting edge defects of the rubber strip in the first embodiment.

[0233] Those skilled in the art can clearly understand that, for the convenience and brevity of description, other working processes of the method described above can refer to the corresponding processes in the aforementioned embodiments and will not be described in detail here.

[0234] Through the description of the above implementation mode, the technicians in the field can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a ROM, a RAM, a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a training device, or a network device, etc.) to execute the method of each embodiment of the present application.

[0235] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0236] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, a computer, a training device or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, training device or data center. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. that includes one or more available media integration. Available media can be magnetic media, (e.g., floppy disk, hard disk, tape), optical media (e.g., DVD), or semiconductor media (e.g., solid-state hard disk (SSD)), etc.

[0237] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting defects on the edge of a rubber strip, characterized in that: include: S1: Obtaining local images of the edge surfaces of N rubber strips to obtain image A, where image A includes image B of the same area of ​​the N rubber strips, where N is an integer greater than or equal to 2; S2: Compare and analyze the N images B to determine whether there is a defective image B1; S3: If it exists, input the image B1 into the defect recognition model for recognition to obtain the defect type. The training samples of the defect recognition model include one or more of the following annotated images: knife marks, knife lines, ripples, and gaps.

2. A method for detecting edge defects of a rubber strip according to claim 1, characterized in that S2 include: S21: Segment image A to obtain N images B; S22: Compare and analyze any two images B to obtain a difference K; S23: Compare the difference K with the difference threshold K1; S24: If K>K1, then compare and analyze the corresponding image B with the standard image to obtain the difference degree K2; S25: Compare the difference K2 with the difference threshold K3, K3<K1; S26: If K2>K3, the corresponding image B is determined to be a defective image B1.

3. A method for detecting edge defects of a rubber strip according to claim 2, characterized in that: S23 and later also include: S231: if K≤K1, then loop through steps S1 to S23 to detect the undetected area of ​​the corresponding rubber strip until the undetected area is cleared; S25 and later also include: S251: If K2≤K3, then loop through steps S1 to S25 to detect the undetected area of ​​the corresponding rubber strip until the undetected area is cleared.

4. The edge defect detection method according to claim 2, characterized in that: Replace S22 with: Compare and analyze any two images B taken from the same adhesive strip to obtain the difference K. The areas corresponding to the two images B are located on the two edge surfaces of the same adhesive strip.

5. The edge detection method according to claim 2, characterized in that: After S1 and before S21, it also includes: S201: Compare and analyze any two images A to obtain a difference k; S202: Subtract k from the difference threshold k1. If k>k1, execute S21.

6. The edge defect detection method according to claim 1, characterized in that: The defect recognition model also has a defect classification function.

7. The edge defect detection method according to claim 1, characterized in that: Before S1, it also included: S0: Arrange N rubber strips in such a way that their edges are coplanar and flush.

8. An electronic device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the edge defect detection method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the edge defect detection method according to any one of claims 1 to 7 are implemented.

10. A device for detecting edge defects of a rubber strip, used to perform the edge defect detection method according to any one of claims 1 to 7, characterized in that: include: An image acquisition component, used for acquiring images of film edges; Image analysis component, used to analyze and process images.

Images