Chain identification method and device, equipment and storage medium
Through artificial intelligence technology and deep learning methods, chain defects are automatically identified, which solves the problem of time-consuming, labor-intensive and inaccurate traditional detection methods, and realizes efficient and accurate chain defect detection and solution strategy formulation, ensuring the normal operation of industrial equipment. and production safety.
Patent Information
- Application Number
- CN202411272291.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-11
- Publication Date
- 2024-12-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional chain defect detection methods rely on manual inspection and simple mechanical equipment, which is time-consuming and labor-intensive and difficult to ensure the accuracy and consistency of detection. Especially in complex working environments, there are noise and interference, which increases image preprocessing and defect recognition. The challenge.
Using artificial intelligence technology, image registration and three-dimensional point cloud model construction are carried out by taking multi-angle images on the surface of the chain, combined with deep learning and morphological operations, image segmentation and defect recognition are carried out, defect types of chains are identified and resolution strategies are formulated.
It realizes automated chain defect identification, improves the accuracy and reliability of detection, reduces the influence of human factors, ensures the consistency and objectivity of the detection results, and improves the efficiency and safety of chain detection.
Smart Images

Figure CN119068178A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of chains, and in particular to a chain identification method, device, equipment and storage medium. Background Art
[0002] As an important transmission component in industrial equipment, the operation status of the chain is directly related to the overall performance and safety of the equipment. However, the chain is prone to various defects during long-term use, such as wear, breakage, deformation, etc. If these defects are not discovered and handled in time, they will lead to equipment failure or even safety accidents. Therefore, early identification and accurate judgment of chain defects are of great significance. At present, traditional chain defect detection methods mainly rely on manual inspection and simple mechanical equipment detection. This method is not only time-consuming and labor-intensive, but also difficult to ensure the accuracy and consistency of detection.
[0003] With the rapid development of artificial intelligence technology, chain defect detection methods based on image recognition have gradually attracted attention. Artificial intelligence, especially deep learning technology, has demonstrated strong capabilities in image processing and pattern recognition, and can automatically, quickly and accurately identify various defects on the surface of the chain. However, there are still some problems in existing research and applications. First, chain defects are of various types and complex in form, and different defects have different manifestations in the image, which makes defect identification more difficult. Secondly, the working environment of the chain is complex and changeable, and there may be noise and interference in the captured image, which increases the challenges of image preprocessing and defect recognition.
[0004] Therefore, how to make full use of artificial intelligence technology to improve the robustness and accuracy of chain defect recognition methods is an important direction of current research. Designing effective image preprocessing and segmentation strategies for different types of chain defects, developing efficient defect recognition algorithms, and formulating appropriate solution strategies based on defect types will significantly improve the efficiency and reliability of chain defect detection and ensure the normal operation and safe production of industrial equipment. Summary of the invention
[0005] The main purpose of the present invention is to provide a chain identification method, device, equipment and storage medium, which solves the technical problem that the traditional chain defect detection method mainly relies on manual inspection and simple mechanical equipment detection, which is not only time-consuming and labor-intensive, but also difficult to ensure the accuracy and consistency of detection.
[0006] To achieve the above object, the present invention provides a chain identification method, comprising the following steps: Shooting the surface of the target chain to obtain a shot image, and preprocessing the shot image to obtain a preprocessed image; Dividing the preprocessed image into a plurality of regions of interest to obtain a chain sub-image to be detected corresponding to each region of interest; Performing defect recognition on the chain sub-image to be detected to obtain a defect recognition result; If the defect recognition result is that the chain sub-image to be detected has a chain defect, then a defect type analysis is performed on the chain sub-image to be detected with the chain defect to obtain the chain defect type; Develop corresponding solution strategies based on the chain defect types.
[0007] Further, photographing the surface of the chain to obtain a photographed image, and preprocessing the photographed image to obtain a preprocessed image, including: The surface of the chain is photographed at multiple angles to obtain images taken at multiple angles, and the images taken at multiple angles are combined into a multi-angle photographed image set; Performing image registration on the multi-angle shot image set to obtain a registered image set; Constructing a three-dimensional point cloud model according to the registered image set to obtain three-dimensional point cloud data of the chain surface; Performing noise reduction processing on the three-dimensional point cloud data to obtain noise-reduced three-dimensional point cloud data; Performing surface reconstruction based on the denoised three-dimensional point cloud data to obtain a three-dimensional mesh model of the chain surface; Perform a two-dimensional projection transformation on the three-dimensional grid model to obtain a preprocessed image.
[0008] Furthermore, the pre-processed image is divided into a plurality of regions of interest to obtain chain sub-images to be detected, including: Performing edge detection on the preprocessed image to obtain a chain edge contour map; Performing image segmentation on the preprocessed image according to the chain edge contour map to obtain the chain main body area; Applying morphological operations to the main region of the chain to obtain a chain skeleton graph; According to the chain skeleton diagram, the chain main body area is sampled at equal intervals to obtain a sampling point set; Taking each sampling point in the sampling point set as the center, construct a rectangular area of a fixed size to obtain a set of candidate regions of interest; Performing overlap analysis on the candidate region of interest set, merging regions with overlaps higher than a preset threshold, and obtaining multiple regions of interest; According to the multiple regions of interest, the pre-processed image is cropped and extracted to obtain multiple chain sub-images to be detected.
[0009] Further, defect recognition is performed on the chain sub-image to be detected to obtain a defect recognition result, including: Performing multi-scale wavelet transform on the chain sub-image to be detected to obtain wavelet decomposition coefficients of multiple scales; According to the wavelet decomposition coefficients, adaptive threshold segmentation is performed on the chain sub-image to be detected to obtain a binary image; Performing morphological operations and connected domain analysis on the binary image to obtain candidate defect regions; Convert the candidate defect region into a vector format to obtain a converted format vector; The conversion format vector is input into a preset defect recognition algorithm to perform defect recognition and obtain a defect recognition result.
[0010] Furthermore, the defect type analysis is performed on the chain sub-image to be detected corresponding to the chain defect to obtain the chain defect type, including: Performing a convolution transformation on the sub-image of the chain to be detected corresponding to the chain defect to obtain a convolution feature map; wherein the convolution feature map is equivalent to highlighting the tiny defects and material changes on the surface of the chain; Based on the convolution feature map, the chain sub-image to be detected is subjected to self-organized critical segmentation to obtain a multi-level segmentation result; wherein the multi-level segmentation result includes the chain surface, the chain link edge and the potential defect area; The multi-level segmentation results are filtered and constructed into a complex sequence to obtain continuous homology features; wherein the continuous homology features include topological changes of the chain surface, connectivity of defective areas, and relative positional relationships between chain links; Performing Betti number sequence calculation on the continuous homology feature to obtain a topological feature descriptor; wherein the topological feature descriptor includes area ratio, perimeter ratio, roundness, and rectangularity; Performing continuous landscape transformation on the topological feature descriptor to obtain a high-dimensional feature vector; wherein the high-dimensional feature vector is used to reflect the multi-scale characteristics and topological continuity of the chain defect; Through a preset defect type algorithm, deep learning and defect type analysis are performed on the chain sub-image to be detected based on the high-dimensional feature vector to obtain a defect recognition result.
[0011] Furthermore, the Betti number sequence is calculated for the continuous homology feature to obtain a topological feature descriptor, including: Performing multi-scale filtering on the continuous coherent features to obtain a filtered continuous coherent feature sequence; Calculate the Betti number based on the filtered continuous coherent characteristic sequence to obtain a multi-scale Betti number sequence; constructing a continuous barcode based on the multi-scale Betti number sequence; Performing continuous entropy calculation based on the continuous barcode to obtain a topological complexity index of the chain defect; Adaptively threshold segmenting the continuous barcode based on the topological complexity index to extract significant topological features of the chain defect; Performing stable manifold learning on the significant topological features of the chain defect to obtain a low-dimensional topological representation of the chain defect; A feature vector of a chain defect is constructed based on the low-dimensional topological representation, and the feature vector of the chain defect is used as a topological feature descriptor.
[0012] Furthermore, a corresponding solution strategy is formulated based on the chain defect type, including: Querying a preset defect-solution mapping table based on the chain defect type to obtain an initial solution set; Performing a feasibility assessment on each solution in the initial solution set to obtain a feasibility score; Sorting the initial solution set according to the feasibility score to obtain a sorted solution set; Selecting the top N solutions from the ranked solution set to construct a candidate solution set; Performing a cost-benefit analysis on each solution in the candidate solution set to obtain a cost-benefit ratio; An optimal solution is selected according to the cost-benefit ratio, and the optimal solution is used as a chain defect resolution strategy.
[0013] The present invention also provides a chain identification device, comprising: A shooting module is used to shoot the surface of the target chain to obtain a shot image, and pre-process the shot image to obtain a pre-processed image; A division module, used for dividing the pre-processed image into a plurality of regions of interest to obtain a chain sub-image to be detected corresponding to each region of interest; An identification module is used to perform defect identification on the chain sub-image to be detected to obtain a defect identification result; An analysis module, configured to, if the defect recognition result is that the chain sub-image to be detected has a chain defect, perform defect type analysis on the chain sub-image to be detected having the chain defect, to obtain the chain defect type; A solution module is used to formulate a corresponding solution strategy based on the chain defect type.
[0014] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.
[0015] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.
[0016] The chain identification method provided by the present invention comprises the following steps: photographing the surface of the target chain to obtain a photographed image, and preprocessing the photographed image to obtain a preprocessed image; dividing the preprocessed image into a plurality of regions of interest to obtain a sub-image of the chain to be detected corresponding to each region of interest; performing defect identification on the sub-image of the chain to be detected to obtain a defect identification result; if the defect identification result is that there is a chain defect in the sub-image of the chain to be detected, then performing defect type analysis on the sub-image of the chain to be detected with the chain defect to obtain the chain defect type; formulating a corresponding solution strategy based on the chain defect type, through the above-mentioned technical means, solves the technical problem that the traditional chain defect detection method mainly relies on manual inspection and simple mechanical equipment detection, which is not only time-consuming and labor-intensive, but also difficult to ensure the accuracy and consistency of detection, and realizes an automated defect recognition method to reduce the influence of human factors, reduce the probability of human misjudgment or missed judgment, and ensure the consistency and objectivity of the detection result. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 is a schematic diagram of the steps of a chain identification method in one embodiment of the present invention; Figure 2 is a structural block diagram of a chain identification device in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.
[0018] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0020] like Figure 1 As shown, Figure 1 This is a schematic diagram of the steps of a chain identification method in one embodiment of the present invention; In one embodiment of the present invention, a chain identification method is provided, comprising the following steps: Step S1, photographing the surface of the target chain to obtain a photographed image, and preprocessing the photographed image to obtain a preprocessed image.
[0021] Specifically, for the implementation of the step of "shooting the surface of the target chain to obtain a shot image, and preprocessing the shot image to obtain a preprocessed image", it is necessary to combine the content in the claims and elaborate its specific implementation process in detail from both hardware and software aspects. First, it is necessary to install a high-resolution industrial camera or camera equipment in an industrial environment to continuously shoot the surface of the target chain to ensure that high-definition images of the chain surface details are captured. These shot images should cover the entire surface of the chain as much as possible to ensure that no possible defects are missed. In order to improve the clarity and stability of the image, a fixed bracket and anti-vibration equipment can be used to stabilize the camera position to prevent image blurring due to vibration or movement during shooting. In addition, a light source with uniform illumination can be installed in the shooting area to eliminate the interference of ambient light so that the shot image has consistent brightness and contrast. After shooting, the shot images obtained may contain background noise, uneven lighting and other interference factors, so these images need to be preprocessed. The preprocessing step includes operations such as image denoising, grayscale transformation, contrast enhancement and edge detection. Specifically, firstly, a denoising algorithm (such as median filtering, mean filtering or Gaussian filtering) is applied to remove noise in the image, thereby improving the clarity of the image. Then, a grayscale transformation is performed to convert the color image into a grayscale image to simplify the subsequent processing process. In order to enhance the contrast of the image, a histogram equalization technique can be used to make the details of the chain surface more prominent, which is convenient for subsequent defect identification. After the image enhancement process, edge detection can be performed, and the edge information of the chain surface can be extracted using methods such as the Canny edge detection algorithm or the Sobel operator to highlight the chain outline and possible defect areas. After these preprocessing steps, the original captured image is converted into a clearer and higher contrast preprocessed image, which will be used as the input for subsequent defect identification to improve the accuracy and reliability of identification. For example, suppose that in a chain production workshop, an automated inspection system is installed, which consists of a high-resolution camera and a light source device. The camera continuously shoots the surface of the chain at a fixed interval, and the resulting image is transmitted to the central processing unit through the network. The central processing unit first denoises these images to eliminate the noise caused by the production environment; then applies grayscale transformation and contrast enhancement technology to make the subtle defects on the chain surface more obvious; finally, uses edge detection algorithm to extract the outline and defect area of the chain. After these preprocessing, the preprocessed images obtained will be used for further defect identification to ensure the efficiency and accuracy of the detection process.
[0022] Step S2, dividing the preprocessed image into a plurality of regions of interest to obtain a chain sub-image to be detected corresponding to each region of interest.
[0023] Specifically, for the implementation of the step of "dividing the preprocessed image into multiple regions of interest to obtain a sub-image of the chain to be detected corresponding to each region of interest", it is necessary to combine the content in the claims and elaborate its specific implementation process in detail from the perspective of image processing technology. First, the division of the preprocessed image is based on the understanding of the surface structure and characteristics of the chain. The preprocessed image has usually been processed by de-noising, grayscale transformation and contrast enhancement, so the surface details and contours of the chain have become clear. Next, it is necessary to automatically identify and divide the regions of interest in these clear images through algorithms. These regions of interest are usually key parts of the chain or locations where defects may exist, such as chain links, pins and rollers. In order to achieve this goal, image segmentation technology can be used first. Common methods include threshold segmentation, edge-based segmentation and region growing algorithms. Specifically, the threshold segmentation method can be used to set one or more grayscale thresholds to divide the image into foreground and background, thereby preliminarily determining the main structural area of the chain. Then, an edge detection algorithm (such as Canny edge detection) is used to further accurately locate the edge of the chain to ensure that the division of the region of interest is more accurate. The region growing algorithm can aggregate connected pixel groups into regions of interest based on the similarity of pixels on the chain surface. After obtaining the regions of interest, these regions can be marked using rectangular or polygonal frames. These marked regions are regions of interest, and the corresponding image parts are called sub-images of the chain to be detected. In order to improve the accuracy and efficiency of detection, further feature extraction and analysis can be performed on each region of interest. The extracted features can include shape features, texture features, and color features, etc., which are convenient for subsequent defect detection algorithms to process. For example, suppose that in an automated inspection system, the preprocessing operation of the preprocessed image of the chain is performed as described above. The system uses the threshold segmentation method to identify the region in the image with a grayscale value higher than a certain set value as the main structural region of the chain. Then, the Canny edge detection algorithm is applied to further obtain the edge information of the chain and accurately locate each link, pin, roller, and other parts of the chain. Next, the region growing algorithm is used to aggregate the connected pixel groups into multiple regions of interest based on the similarity of pixels on the chain surface, and these regions are marked with rectangular frames. Each marked area is a sub-image of the chain to be inspected, and these sub-images will be used in the subsequent defect detection process. Through the above steps, the system can divide the pre-processed image into multiple regions of interest, ensuring that the sub-image of the chain to be inspected corresponding to each region of interest has high detection accuracy and reliability, thereby improving the overall detection efficiency and accuracy.
[0024] Step S3, performing defect recognition on the chain sub-image to be detected to obtain a defect recognition result.
[0025] Specifically, for the implementation of the step of "performing defect recognition on the sub-image of the chain to be detected to obtain the defect recognition result", it is necessary to combine the content in the claims and elaborate its implementation process from the perspective of specific image processing and machine learning technology. First, the sub-image of the chain to be detected has been divided from the pre-processed image in the previous step, and each sub-image contains an important part of the chain or a potential defect area. In order to perform defect recognition, advanced image processing and pattern recognition technology must be used. In the specific implementation process, a combination of traditional image processing algorithms and modern deep learning methods can be used. Before starting defect recognition, it is necessary to first extract features from each sub-image of the chain to be detected. Feature extraction can be achieved through traditional methods such as SIFT (scale-invariant feature transform) and SURF (speeded up robust features), or through deep learning methods such as convolutional neural networks (CNN). Taking CNN as an example, the first few layers of the network can automatically extract low-level features of the chain image such as edges, textures, and colors, while the deep network can extract higher-level features such as the shape, structure, and defect pattern of the chain. After feature extraction, the extracted features need to be input into the classifier for defect recognition. The classifier can be a traditional support vector machine (SVM), random forest, etc., or a fully connected neural network based on deep learning. Through the trained classifier, the system can automatically identify the types of defects in the chain sub-image, such as cracks, wear, rust, etc. In order to further improve the accuracy and robustness of recognition, data enhancement technology can be used to expand the training data, such as rotation, scaling, translation and other operations to increase the generalization ability of the model. In addition, the cross-validation method can effectively prevent overfitting and ensure the stability and accuracy of the model in practical applications. For example, assuming that the system has acquired a chain sub-image to be detected, the pre-trained convolutional neural network model is used to extract its features, and the extracted feature vector is input into the support vector machine classifier. The classifier determines whether there is a crack defect in the sub-image based on the pattern of the feature vector and outputs the defect recognition result. If a crack is identified, the system will record the defect location and severity and generate a corresponding defect recognition report. Through the above steps, the system can efficiently and accurately identify defects in the chain sub-image to be detected and obtain defect recognition results. The entire process combines traditional image processing technology and modern deep learning methods to ensure high accuracy and reliability of defect identification, thereby greatly improving the automation level and detection efficiency of chain detection.
[0026] Step S4: if the defect recognition result is that the chain sub-image to be detected has a chain defect, a defect type analysis is performed on the chain sub-image to be detected having the chain defect to obtain the chain defect type.
[0027] Specifically, for the implementation of the step of "if the defect recognition result is that the chain sub-image to be detected has a chain defect, then the defect type analysis is performed on the chain sub-image to be detected with the chain defect to obtain the chain defect type", it is necessary to combine the content in the claims and elaborate on the specific process from the defect recognition result to the defect type analysis. After defect recognition, if the system determines that the chain sub-image to be detected has a chain defect, then a more in-depth analysis of the sub-image is required to determine the specific defect type. This process requires a detailed identification of the chain defect, and advanced image classification and pattern recognition techniques can usually be used. First, through the defect recognition results of the previous step, the system has marked the sub-image with defects. At this time, more complex deep learning models can be used, such as deeper convolutional neural networks (CNN) or other types of neural networks such as ResNet, Inception, etc. These models can provide higher accuracy and stronger feature extraction capabilities. Through these models, the system can perform a detailed analysis of the chain sub-image with defects and extract advanced features related to the defect type. In the specific implementation, a multi-classification model can be designed, which can distinguish different types of chain defects, such as cracks, wear, rust, deformation, etc. Training this model requires a large number of labeled chain defect image data sets, and each defect type requires a rich sample to ensure that the model can learn the characteristics of each defect type. By introducing data enhancement techniques such as rotation, scaling, translation, etc. during the training process, the generalization ability of the model can be improved and overfitting can be reduced. In practical applications, when the system detects that a sub-image has a chain defect, the sub-image will be input into the pre-trained multi-classification model. The model will analyze the features in the image and output the probability distribution of the defect type. By analyzing the probability distribution, the system can determine the most likely defect type. For example, if the crack feature in a sub-image is significant, the model will output a higher probability of crack defect, thereby judging that the defect type is crack. For example, assuming that a chain sub-image is detected to have a defect, the system inputs the sub-image into the multi-classification model for defect type analysis. The model determines that the defect type is wear by analyzing the features in the image and outputs the probability of wear defect. Based on the output of the model, the system records and reports that the defect type of the chain sub-image is wear, and generates a detailed defect type analysis report, including information such as the location, shape, and severity of the defect. Through the above steps, the system can accurately analyze the defect type of the chain sub-image to be detected with chain defects and obtain the chain defect type. This process not only improves the accuracy of defect identification, but also provides an important reference for subsequent chain maintenance and repair, ensuring the comprehensiveness and reliability of the chain detection system.
[0028] Step S5: formulate a corresponding solution strategy based on the chain defect type.
[0029] Specifically, after analyzing the chain defect type, the next step is to "formulate a corresponding solution strategy based on the chain defect type". The implementation of this process needs to be combined with the content in the claims, and through the analysis of different defect types, a corresponding solution is formulated to ensure that the maintenance and repair of the chain can be carried out in a targeted manner. First, after the system identifies the chain defect and determines the specific chain defect type, it will enter the solution strategy formulation stage. At this stage, the system will refer to the preset solution strategy library according to different chain defect types. The solution strategy library contains repair and maintenance solutions for various chain defects. These solutions are summarized based on experience and practice and can effectively solve the chain defect problem. Specifically, the system will retrieve the corresponding solution strategy from the solution strategy library according to the chain defect type. For example, if the chain defect type is a crack, the system will look for a solution strategy related to the crack. The solution strategy for the crack may include using a specific welding technique for repair, or replacing the chain segment when the crack is severe. In addition, the system will also provide detailed operating steps and precautions to ensure the safety and effectiveness of the repair process. For chain defects of the wear type, the solution strategy may include chain lubrication, adjusting chain tension or replacing severely worn links. The system will recommend the appropriate lubricant type and lubrication frequency based on the degree of wear, and provide specific operation steps and tool requirements. These strategies can not only extend the service life of the chain, but also improve the operating efficiency of the chain and reduce the downtime of the equipment. If the chain defect type is rust, the system will recommend the use of specific rust removers for cleaning, or replace the damaged links when the rust is severe. The solution strategy for rust removal may include chemical rust removal, electrochemical rust removal or mechanical rust removal, each of which has its applicable conditions and steps. The system will recommend the most appropriate rust removal method based on the severity of the rust, and provide detailed operation guidelines to ensure that the operator can successfully complete the rust removal work. In the treatment of chain defects of the deformation type, the system will provide methods to correct the deformation, or recommend replacing the deformed links when correction is not possible. The solution strategy for deformation may involve the use of specific tools for correction, such as chain tensioners or other professional equipment. The system will provide detailed correction steps, precautions for using tools and safe operation guidelines to ensure the safety and effectiveness of the repair process. For example, suppose the system detects a wear defect in a chain sub-image and determines that the defect type is wear. The system retrieves a solution strategy for wear from the solution strategy library, recommends lubrication with a specific type of lubricant, and suggests the frequency and method of lubrication. The operator performs lubrication operations according to the guidance provided by the system to ensure that the chain wear problem is effectively solved, thereby extending the service life of the chain. Through the above steps, the system can formulate a corresponding solution strategy based on the chain defect type.This process not only improves the efficiency and accuracy of chain maintenance and repair, but also provides detailed guidance for operators to ensure the effective implementation of solution strategies, ultimately ensuring the normal operation of the chain and the long-term stability of the equipment.
[0030] In a specific embodiment, photographing the surface of the chain to obtain a photographed image, and preprocessing the photographed image to obtain a preprocessed image includes: The surface of the chain is photographed at multiple angles to obtain images taken at multiple angles, and the images taken at multiple angles are combined into a multi-angle photographed image set; Performing image registration on the multi-angle shot image set to obtain a registered image set; Constructing a three-dimensional point cloud model according to the registered image set to obtain three-dimensional point cloud data of the chain surface; Performing noise reduction processing on the three-dimensional point cloud data to obtain noise-reduced three-dimensional point cloud data; Performing surface reconstruction based on the denoised three-dimensional point cloud data to obtain a three-dimensional mesh model of the chain surface; Perform a two-dimensional projection transformation on the three-dimensional grid model to obtain a preprocessed image.
[0031] Specifically, in the process of shooting the chain surface and obtaining the shooting image, the surface of the chain must first be shot at multiple angles to ensure that all the details of the chain surface can be fully captured. The purpose of this step is to construct the complete image information of the chain surface through images shot at different angles, thereby laying the foundation for subsequent image processing. Multi-angle shooting can be done by using multiple cameras at the same time, or by using a single camera to shoot from different angles in sequence, and finally obtain multi-angle shooting images. When the multi-angle shooting images are acquired, these images will be combined into a multi-angle shooting image set. This image set contains chain surface images shot from different perspectives, which can provide rich visual information. However, since images shot at different angles may have differences in position, angle and scale, these images need to be registered. Image registration is the process of aligning images shot at different angles to ensure that they have consistent positions and directions in the same coordinate system. Through image registration, the deviation caused by different shooting angles can be eliminated to obtain a registered image set. Next, based on the registered image set, a three-dimensional point cloud model of the chain surface can be constructed. The 3D point cloud model is composed of a large number of points, which represent the shape and structure of the chain surface in 3D space. The process of constructing a 3D point cloud model usually involves converting the registered 2D image information into 3D space coordinates, which can be achieved by calculating the depth information of each point in the image. The obtained 3D point cloud data can accurately describe the geometric shape of the chain surface and provide a basis for the next step of processing. The 3D point cloud data may contain noise, which comes from the illumination changes, image resolution, camera errors, etc. during the shooting process. Therefore, it is necessary to perform denoising on the 3D point cloud data to remove the noise and improve the accuracy of the data. Various algorithms can be used for denoising, such as mean filtering, statistical filtering, etc., through which the noise points in the 3D point cloud data can be eliminated to obtain smoother and more accurate denoised 3D point cloud data. After obtaining the denoised 3D point cloud data, surface reconstruction can be performed based on these data. The purpose of surface reconstruction is to convert the point cloud data into a continuous 3D surface model, that is, a 3D mesh model. The 3D mesh model is composed of a series of interconnected triangular patches, through which the shape and details of the chain surface can be accurately described. Surface reconstruction usually uses algorithms such as Delaunay triangulation and Poisson surface reconstruction, which can convert point cloud data into a high-quality three-dimensional mesh model. Finally, the three-dimensional mesh model is subjected to a two-dimensional projection transformation to obtain a preprocessed image. Two-dimensional projection transformation is the process of converting a three-dimensional mesh model into a two-dimensional image. By projecting the three-dimensional model onto a plane, a two-dimensional image containing rich details can be obtained.The preprocessed image is the final output. It retains all the detailed information of the chain surface and has been processed through multiple steps such as multi-angle shooting, image registration, 3D point cloud construction, noise reduction and surface reconstruction. It has high quality and accuracy. For example, suppose we need to detect the surface defects of a conveyor chain. First, we use multiple cameras to shoot the chain surface from different angles to obtain multiple images shot at different angles. Then, these images are combined into a multi-angle shooting image set and image registration is performed to ensure that all images are aligned in the same coordinate system. Next, the registered image set is used to build a 3D point cloud model of the chain surface. By calculating the depth information in the image, the 2D image is converted into 3D space coordinates. The obtained 3D point cloud data may contain noise, so it is necessary to perform noise reduction to remove noise points and obtain more accurate denoised 3D point cloud data. Subsequently, surface reconstruction is performed based on the denoised point cloud data to obtain a 3D mesh model of the chain surface. Finally, the 3D mesh model is converted into a 2D preprocessed image through a 2D projection transformation. This preprocessed image retains all the detailed information of the chain surface and can be used for subsequent defect detection and analysis.
[0032] Through the above steps, the process of photographing the chain surface and obtaining pre-processed images can be realized. This process can not only provide high-quality image data, but also provide a reliable basis for the detection and analysis of chain surface defects, thereby ensuring the normal operation of the chain and the long-term stability of the equipment.
[0033] In a specific embodiment, the pre-processed image is divided into a plurality of regions of interest to obtain a chain sub-image to be detected, including: Performing edge detection on the preprocessed image to obtain a chain edge contour map; Performing image segmentation on the preprocessed image according to the chain edge contour map to obtain the chain main body area; Applying morphological operations to the main region of the chain to obtain a chain skeleton graph; According to the chain skeleton diagram, the chain main body area is sampled at equal intervals to obtain a sampling point set; Taking each sampling point in the sampling point set as the center, construct a rectangular area of a fixed size to obtain a set of candidate regions of interest; Performing overlap analysis on the candidate region of interest set, merging regions with overlaps higher than a preset threshold, and obtaining multiple regions of interest; According to the multiple regions of interest, the pre-processed image is cropped and extracted to obtain multiple chain sub-images to be detected.
[0034] Specifically, in order to realize the process of dividing the preprocessed image into multiple regions of interest to obtain the chain sub-image to be detected, we first need to start with the preprocessed image and perform a series of image processing operations on it. These operations include edge detection, image segmentation, morphological operations, skeleton extraction, construction of sampling point sets, generation of rectangular regions and overlap analysis. The following is a detailed explanation of these steps. First, edge detection is performed on the preprocessed image. The purpose of this step is to identify the edge contour of the chain in the image. Edge detection is an image processing technique that identifies edges by detecting areas in the image where the grayscale value changes significantly. Commonly used edge detection algorithms include Canny edge detection, Sobel operator and Laplacian operator. Through edge detection, we can obtain a chain edge contour map, which can clearly show the position and shape of the chain in the preprocessed image. Next, the preprocessed image is segmented according to the chain edge contour map to obtain the main chain area. The purpose of image segmentation is to divide the image into different regions, each of which corresponds to our object of interest. Here, we need to separate the chain from the background and obtain an area containing only the chain. Edge-based segmentation methods, such as the active contour model (also known as Snake), can be used to iteratively adjust the segmentation boundary to fit the edge contour of the chain to achieve segmentation. Then, morphological operations are applied to the main area of the chain to obtain a chain skeleton map. Morphological operations are a processing method based on image shapes. Common operations include dilation, erosion, opening operations, and closing operations. By performing morphological operations on the main area of the chain, noise can be removed, holes can be filled, and the skeleton structure of the chain can be extracted. The chain skeleton map is a refined representation of the main body of the chain, which retains the basic shape and topological structure of the chain. Subsequently, the main area of the chain is sampled at equal intervals according to the chain skeleton map to obtain a set of sampling points. The purpose of equal-interval sampling is to select a series of evenly distributed points on the chain skeleton, which will be used as the center of the subsequent region of interest. Sampling can be achieved by selecting points at equal distances along the chain skeleton, and the obtained sampling point set can represent the position and distribution of the chain in the image. A fixed-size rectangular area is constructed with each sampling point in the sampling point set as the center to obtain a set of candidate regions of interest. The purpose of this step is to define a region of interest around each sampling point of the chain for subsequent detailed analysis. According to the actual size of the chain and the image resolution, the appropriate rectangular area size can be selected to ensure that each rectangular area can cover a part of the chain. Next, the candidate region of interest set is analyzed for overlap, and the regions with overlaps above a preset threshold are merged to obtain multiple regions of interest. In the generated candidate region of interest set, there may be some regions that overlap with each other, resulting in information redundancy.Through the overlap analysis, the overlapping area of each pair of candidate regions can be calculated, and the regions with an overlap higher than a preset threshold can be merged, thereby reducing redundancy and obtaining more independent regions of interest. Finally, according to the multiple regions of interest, the preprocessed image is cropped and extracted to obtain multiple sub-images of the chain to be detected. For each region of interest, the corresponding position in the preprocessed image can be cropped to extract the sub-image containing the region. These sub-images retain the detailed information of the chain and can be used for subsequent defect detection and analysis. For example, suppose we have a preprocessed image showing the surface of a section of chain. First, we use the Canny edge detection algorithm to perform edge detection on the image to obtain the edge contour map of the chain. Then, based on this contour map, we use the active contour model to segment the image to obtain the main area containing only the chain. Next, we apply corrosion and expansion operations to the main area of the chain to extract the skeleton map of the chain. According to the skeleton map, we select sampling points at equal intervals along the chain skeleton to construct a set of sampling points. At each sampling point position, we define a rectangular area of fixed size to obtain a set of candidate regions of interest. In order to eliminate redundancy, we analyze the overlap of these candidate regions and merge the regions with high overlap to obtain multiple independent regions of interest. Finally, we crop these regions of interest in the preprocessed image to obtain multiple sub-images of the chain to be inspected. These sub-images can be used for further analysis, such as detecting defects and wear on the chain surface. Through this processing flow, the detailed areas of the chain can be effectively segmented and extracted, providing high-quality image data for subsequent inspection and analysis.
[0035] In a specific embodiment, defect recognition is performed on the chain sub-image to be detected to obtain a defect recognition result, including: Performing multi-scale wavelet transform on the chain sub-image to be detected to obtain wavelet decomposition coefficients of multiple scales; According to the wavelet decomposition coefficients, adaptive threshold segmentation is performed on the chain sub-image to be detected to obtain a binary image; Performing morphological operations and connected domain analysis on the binary image to obtain candidate defect regions; Convert the candidate defect region into a vector format to obtain a converted format vector; The conversion format vector is input into a preset defect recognition algorithm to perform defect recognition and obtain a defect recognition result.
[0036] Specifically, in order to realize the process of defect recognition of the sub-image of the chain to be detected, we need to perform image processing and analysis according to a series of detailed steps. These steps include multi-scale wavelet transform, adaptive threshold segmentation, morphological operation and connected domain analysis, vector format conversion and defect recognition algorithm. The following is a detailed explanation of these steps. First, we perform multi-scale wavelet transform on the sub-image of the chain to be detected, with the aim of extracting image features at different scales. Wavelet transform is a powerful image processing tool that can decompose an image into different scales and frequency components to capture details and edges in the image. In multi-scale wavelet transform, we select appropriate wavelet basis functions, such as Haar wavelet or Daubechies wavelet, and decompose the sub-image of the chain to be detected to obtain wavelet decomposition coefficients of multiple scales. These coefficients represent the characteristics of the image at different levels of detail and can be used to identify anomalies or defects in the image. Next, adaptive threshold segmentation is performed on the sub-image of the chain to be detected according to the wavelet decomposition coefficients to obtain a binary image. Adaptive threshold segmentation is a dynamic image segmentation method that can automatically adjust the segmentation threshold according to the local characteristics of the image. Specifically, we use wavelet decomposition coefficients to determine the eigenvalues of each pixel in the image, and set thresholds based on these eigenvalues to segment the image into foreground (possible defect area) and background (normal area). In this way, we can get a binary image in which white pixels represent possible defect areas and black pixels represent normal areas. Then, morphological operations and connected domain analysis are performed on the binary image to further extract candidate defect areas. Morphological operation is a processing method based on image shape. Common operations include corrosion, expansion, opening operation and closing operation. Here, we can use closing operation to fill small holes in the binary image and use opening operation to remove noise points. After morphological operation processing, we get a smoother and more coherent binary image. Next, we perform connected domain analysis, that is, identify and mark all connected white pixel areas in the image, and each connected domain represents a candidate defect area. Through this analysis, we can obtain a series of information such as the outline and area of candidate defect areas. Subsequently, the candidate defect area is converted into a vector format to obtain a converted format vector. The purpose of vector format conversion is to convert the candidate defect area in the image into a data format that can be input into the defect recognition algorithm. Specifically, we can extract the features of each candidate defect area, such as its area, perimeter, shape, position, and wavelet decomposition coefficients, and organize these features into a vector. Each vector represents the feature information of a candidate defect area, and these vectors will be used as input data for subsequent defect recognition algorithms. Finally, the converted format vector is input into the preset defect recognition algorithm for defect recognition to obtain the defect recognition result.Defect recognition algorithms can be based on machine learning methods, such as support vector machines (SVM), convolutional neural networks (CNN), or deep learning models. These models are trained to identify various defects on the chain surface, such as cracks, wear, corrosion, etc. Specifically, we input the transformed format vector into the trained model, and the model will classify the input feature vector to determine whether there is a defect in the area, and output information such as the defect category and confidence. In this way, we can get the defect recognition results for each candidate defect area, including detailed information such as its defect type and location. For example, suppose we have a sub-image of a chain to be detected, which shows the surface of a section of the chain. First, we perform a multi-scale wavelet transform on this sub-image, using the Haar wavelet basis function to decompose the image into wavelet decomposition coefficients of different scales. Then, based on these decomposition coefficients, we perform adaptive threshold segmentation to obtain a binary image, in which the white area represents the possible defect area. Then, we apply morphological operations to the binary image, using the opening operation to remove noise points and the closing operation to fill small holes to obtain a smooth binary image. Next, we performed a connected domain analysis to identify all connected white areas and obtain a series of candidate defect areas. Subsequently, we extracted the features of each candidate defect area, including its area, perimeter, shape, position, and wavelet decomposition coefficients, and constructed them into a vector format. Finally, we input these vectors into a pre-trained convolutional neural network model for defect recognition. The model outputs the recognition results of each candidate defect area, such as cracks, wear, or normal areas, and gives the confidence level. Through this method, we successfully realized the defect recognition of the sub-image of the chain to be inspected and obtained detailed defect recognition results, which provided an important basis for subsequent chain maintenance and quality control.
[0037] In a specific embodiment, defect type analysis is performed on the chain sub-image to be detected corresponding to the chain defect to obtain the chain defect type, including: Performing a convolution transformation on the sub-image of the chain to be detected corresponding to the chain defect to obtain a convolution feature map; wherein the convolution feature map is equivalent to highlighting the tiny defects and material changes on the surface of the chain; Based on the convolution feature map, the chain sub-image to be detected is subjected to self-organized critical segmentation to obtain a multi-level segmentation result; wherein the multi-level segmentation result includes the chain surface, the chain link edge and the potential defect area; The multi-level segmentation results are filtered and constructed into a complex sequence to obtain continuous homology features; wherein the continuous homology features include topological changes of the chain surface, connectivity of defective areas, and relative positional relationships between chain links; Performing Betti number sequence calculation on the continuous homology feature to obtain a topological feature descriptor; wherein the topological feature descriptor includes area ratio, perimeter ratio, roundness, and rectangularity; Performing continuous landscape transformation on the topological feature descriptor to obtain a high-dimensional feature vector; wherein the high-dimensional feature vector is used to reflect the multi-scale characteristics and topological continuity of the chain defect; Through a preset defect type algorithm, deep learning and defect type analysis are performed on the chain sub-image to be detected based on the high-dimensional feature vector to obtain a defect recognition result.
[0038] Specifically, when processing the sub-image of the chain to be detected corresponding to the chain defect, it is first necessary to analyze the defect type of the sub-image to accurately identify and classify the chain defect type. This process involves multiple steps and complex image processing and analysis techniques, mainly including convolution transformation, self-organizing critical segmentation, continuous homology feature construction, Betti number sequence calculation, continuous landscape transformation, and finally deep learning and defect type analysis. First, a convolution transformation is performed on the sub-image of the chain to be detected. Convolution transformation is a commonly used technique in deep learning. It can extract local features in the image by operating on the image through a series of convolution kernels. The convolution feature map is the result of the convolution transformation, which can highlight the tiny defects and material changes on the surface of the chain. For example, if there are small cracks or wear on the surface of the chain, these details will appear in the convolution feature map through the convolution transformation, which becomes the basis for subsequent analysis. Then, the sub-image of the chain to be detected is self-organized critical segmented based on the convolution feature map. Self-organized critical segmentation is an image segmentation method that can automatically adjust the segmentation threshold according to the characteristics of the image to achieve multi-level segmentation. In this step, the chain sub-image to be detected will be segmented into different levels, including the chain surface, the edge of the chain link, and the potential defect area. For example, if there are rust areas on the chain surface, the self-organizing critical segmentation can distinguish these rust areas from the normal chain surface, forming a clear segmentation result. Subsequently, the multi-level segmentation results are subjected to the filtered complex sequence construction to obtain the continuous homology features. The filtered complex sequence construction is a topological data analysis method. By analyzing the topological structure of the segmentation results, the topological structure changes of the chain surface, the connectivity of the defect area, and the relative position relationship between the chain links can be extracted. For example, in a certain part of the chain, if there are multiple connected defect areas, the filtered complex sequence construction can capture the connectivity characteristics of these areas and provide data support for subsequent topological analysis. Next, the Betti number sequence is calculated for the continuous homology features to obtain the topological feature descriptor. The Betti number is a basic concept in topology, which is used to describe the connectivity and number of holes in the topological space. In this step, by calculating the Betti numbers in different dimensions, the topological feature descriptors of the chain surface can be obtained, including area ratio, perimeter ratio, roundness, rectangularity, etc. For example, if there are multiple circular wear areas on the chain surface, the geometric characteristics and distribution of these areas can be accurately described by calculating the Betti number sequence. Then, the topological feature descriptors are subjected to continuous landscape transformation to obtain a high-dimensional feature vector. Continuous landscape transformation is a method of mapping topological features to high-dimensional space, which can reflect the multi-scale characteristics and topological persistence of chain defects. In this step, by transforming the topological feature descriptors, a high-dimensional feature vector can be obtained, which contains various detailed features of the chain defects.For example, if the chain defect has multi-scale features (such as wear areas of varying sizes), the continuous landscape transformation can comprehensively represent these features in a high-dimensional space, providing input data for subsequent deep learning. Finally, through the preset defect type algorithm, the chain sub-image to be inspected is subjected to deep learning and defect type analysis based on the high-dimensional feature vector to obtain the defect recognition result. Deep learning is a machine learning method based on neural networks that can automatically extract and classify large-scale data. By inputting high-dimensional feature vectors into the deep learning model, the model can identify and classify chain defects. For example, if there are cracks and wear on the surface of the chain, the deep learning model can accurately distinguish these defect types and give specific recognition results. In general, through the comprehensive application of the above steps, efficient recognition and classification of chain defects can be achieved. This process not only utilizes image processing technologies such as convolution transformation and self-organizing critical segmentation, but also combines topological data analysis methods such as filtered complex sequence construction, Betti number sequence calculation, and continuous landscape transformation, as well as the powerful classification ability of deep learning, so as to achieve comprehensive analysis and accurate recognition of chain defects. For example, in actual applications, if a part of a chain has small cracks and wear, the automation of the above steps can quickly identify these defects and provide corresponding repair suggestions based on the defect type, greatly improving the efficiency and accuracy of chain maintenance.
[0039] In a specific embodiment, the Betti number sequence is calculated for the continuous homology feature to obtain a topological feature descriptor, including: Performing multi-scale filtering on the continuous coherent features to obtain a filtered continuous coherent feature sequence; Calculate the Betti number based on the filtered continuous coherent characteristic sequence to obtain a multi-scale Betti number sequence; constructing a continuous barcode based on the multi-scale Betti number sequence; Performing continuous entropy calculation based on the continuous barcode to obtain a topological complexity index of the chain defect; Adaptively threshold segmenting the continuous barcode based on the topological complexity index to extract significant topological features of the chain defect; Performing stable manifold learning on the significant topological features of the chain defect to obtain a low-dimensional topological representation of the chain defect; A feature vector of a chain defect is constructed based on the low-dimensional topological representation, and the feature vector of the chain defect is used as a topological feature descriptor.
[0040] Specifically, in the above claims, the specific steps of calculating the Betti number sequence of the continuous homology feature to obtain the topological feature descriptor are described in detail. This process gradually extracts and analyzes the topological features of the chain defects through a series of complex topological data analysis and machine learning techniques, thereby realizing accurate identification and classification of chain defects. First, the system needs to perform multi-scale filtering on the continuous homology feature to obtain a filtered continuous homology feature sequence. Multi-scale filtering is a technology for smoothing and denoising data at different scales. By performing multi-scale filtering on the continuous homology feature, the system can eliminate noise and unnecessary details while retaining important topological information at different scales. The importance of this step is that it can ensure that the Betti number sequence calculated subsequently is more accurate and reliable. Next, the Betti number is calculated based on the filtered continuous homology feature sequence to obtain a multi-scale Betti number sequence. The Betti number is an important topological invariant used to describe connected components and holes in the topological space. In this step, the system calculates the Betti number at different scales to obtain a multi-scale Betti number sequence. These sequences reflect the topological structure changes of the chain defect area at different scales and provide rich topological information. Then, the system constructs a persistent barcode based on the multi-scale Betti number sequence. A persistent barcode is a graphical representation of topological data that displays the duration of topological features at different scales in the form of a bar graph. The persistent barcode can intuitively display the topological features of the chain defect area and its changes at different scales, providing a clear reference for subsequent analysis. After obtaining the persistent barcode, the system calculates the persistent entropy based on the barcode to obtain the topological complexity index of the chain defect. Persistent entropy is a metric used to quantify the complexity of topological data, which can measure the diversity and distribution of topological features. By calculating the persistent entropy, the system can obtain the topological complexity index of the chain defect area, which reflects the complexity and structural characteristics of the defect area. Then, based on the topological complexity index, the system performs adaptive threshold segmentation on the persistent barcode to extract the significant topological features of the chain defect. Adaptive threshold segmentation is a method of dynamically adjusting the threshold, which can automatically select the appropriate threshold according to the topological complexity index, so as to extract the most significant topological features in the persistent barcode. In this way, the system can accurately identify the most critical topological features in the chain defect area, providing an important basis for subsequent feature representation and classification. Subsequently, the system performs stable manifold learning on the significant topological features of the chain defects to obtain a low-dimensional topological representation of the chain defects. Stable manifold learning is a nonlinear dimensionality reduction technology that can map high-dimensional data to a low-dimensional space while maintaining the topological structure of the data. Here, the system performs stable manifold learning on the significant topological features to obtain a low-dimensional topological representation of the chain defects. This representation retains the key topological structure of the chain defect area while reducing the data dimension, facilitating subsequent feature extraction and classification.Finally, based on the low-dimensional topological representation, the system constructs the feature vector of the chain defect and uses it as a topological feature descriptor. Feature vector is a vectorized form used to represent data features, which can convert complex topological information into an easy-to-process numerical form. Here, the system constructs the feature vectors of the chain defect based on the low-dimensional topological representation. These vectors contain the key topological features and structural information of the chain defect area. Finally, the system uses these feature vectors as topological feature descriptors for subsequent defect identification and classification. To give a specific example, suppose we have a chain with a complex crack in one of the links. First, the system performs multi-scale filtering on the persistent homology features of the link to eliminate noise and retain topological information at different scales. Next, the system calculates the Betti numbers at different scales to obtain multi-scale Betti number sequences, which reflect the topological structure changes of the crack area at different scales. Then, a persistent barcode is constructed based on the multi-scale Betti number sequence, which shows the topological features of the crack area and its changes at different scales. Next, the system performs persistent entropy calculation based on the persistent barcode to obtain the topological complexity index of the crack area. Based on this indicator, the system will perform adaptive threshold segmentation on the continuous barcode to extract the significant topological features of the crack area. Subsequently, the system performs stable manifold learning on the significant topological features to obtain a low-dimensional topological representation of the crack area. Finally, based on the low-dimensional topological representation, the system constructs a feature vector of the crack area and uses it as a topological feature descriptor. Through this series of steps, the system can accurately identify and classify cracks in the chain links, thereby outputting defect identification results. Through the above process, the system can achieve accurate detection and analysis of chain defects to ensure the quality and performance of the chain. This process not only utilizes advanced topological data analysis technology, but also combines machine learning methods such as deep learning to provide strong technical support for chain defect detection.
[0041] In a specific embodiment, a corresponding solution strategy is formulated based on the chain defect type, including: Querying a preset defect-solution mapping table based on the chain defect type to obtain an initial solution set; Performing a feasibility assessment on each solution in the initial solution set to obtain a feasibility score; Sorting the initial solution set according to the feasibility score to obtain a sorted solution set; Selecting the top N solutions from the ranked solution set to construct a candidate solution set; Performing a cost-benefit analysis on each solution in the candidate solution set to obtain a cost-benefit ratio; An optimal solution is selected according to the cost-benefit ratio, and the optimal solution is used as a chain defect resolution strategy.
[0042] Specifically, in the process of solving chain defects, we must first formulate corresponding solution strategies based on the chain defect type. This process can be divided into multiple steps, each of which is critical to ensure that the final solution selected is not only effective but also economical. First, based on the chain defect type, query the preset defect-solution mapping table. In this mapping table, various possible chain defects and their corresponding preliminary solutions are covered. For example, if chain breakage is a common defect type, then the mapping table may list a variety of preliminary solutions for repair or replacement, such as using reinforcement materials, re-welding, or directly replacing new chain segments. By querying the mapping table, we can quickly get an initial solution set that contains all potential solutions related to the current defect type. Next, each solution in the initial solution set is evaluated for feasibility and a feasibility score is obtained. The feasibility assessment involves multiple aspects, including technical feasibility, resource availability, implementation difficulty, and time requirements. For example, in the case of chain breakage, re-welding may be technically feasible, but if the chain is in an inaccessible position, the implementation difficulty and time cost may be high. By evaluating these factors, each solution will receive a feasibility score, which reflects the possibility and difficulty of each solution being successfully implemented in practice. After obtaining the feasibility scores, the initial solution set is sorted according to these scores to obtain the sorted solution set. The purpose of this step is to rank those solutions that are more technically feasible, more easily accessible, less difficult to implement, and less time-consuming. For example, if re-welding has a low feasibility score and the score of using reinforcement materials is high, then in the sorted solution set, the solution of using reinforcement materials will be ranked before re-welding. Then, the top N solutions are selected from the sorted solution set to construct the candidate solution set. This step ensures that we only consider those solutions with the best feasibility and avoid wasting time and resources on those solutions with a lower possibility of successful implementation. Assuming that there are five solutions in the sorted solution set, based on the scores, we may only select the top two or three solutions as the candidate solution set. Perform a cost-benefit analysis on each solution in the candidate solution set to obtain a cost-benefit ratio. Cost-benefit analysis mainly examines the relationship between the implementation cost of each solution and its expected benefit. The implementation costs include material costs, labor costs, time costs, etc., while the expected benefits include repair effects, extended service life, improved production efficiency, etc. For example, the solution using reinforcement materials may be low-cost and quick to implement, but its effectiveness and durability may not be as good as replacing the chain segments with a new one. Therefore, by calculating the cost-benefit ratio of each solution, we can objectively compare these solutions and select the best solution in terms of comprehensive benefits. Finally, the optimal solution is selected based on the cost-benefit ratio, and the optimal solution is used as the chain defect solution strategy.This step ensures that the selected solution is not only feasible, but also achieves the best balance between cost and benefit. Assuming that in the cost-benefit analysis, the solution of using reinforcement materials has the lowest cost but the worst benefit, while the solution of completely replacing the chain segments has the highest cost but the greatest benefit, then we may choose a compromise solution, such as partial replacement with high-quality materials, to find the best balance between cost and benefit. Through the above steps, we can systematically analyze and select the optimal chain defect solution strategy to ensure that the selected solution is technically feasible, economically reasonable, and can effectively solve the problem. This process not only improves the scientificity and accuracy of decision-making, but also maximizes resource and time savings and improves overall efficiency.
[0043] The above describes the chain identification method in the embodiment of the present invention. The following describes the chain identification device in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a chain identification device includes: The shooting module 21 is used to shoot the surface of the target chain to obtain a shot image, and pre-process the shot image to obtain a pre-processed image; A division module 22, used for dividing the pre-processed image into a plurality of regions of interest to obtain a chain sub-image to be detected corresponding to each region of interest; The recognition module 23 is used to perform defect recognition on the chain sub-image to be detected to obtain a defect recognition result; The analysis module 24 is used for, if the defect recognition result is that the chain sub-image to be detected has a chain defect, then performing defect type analysis on the chain sub-image to be detected having the chain defect to obtain the chain defect type; The solution module 25 is used to formulate a corresponding solution strategy based on the chain defect type.
[0044] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.
[0045] Reference Figure 3 The present invention also provides a computer device in an embodiment, wherein the internal structure of the computer device can be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.
[0046] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.
[0047] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.
[0048] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.
[0049] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.
[0050] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.
Claims
1. A chain identification method, characterized in that: The following steps are involved: Shooting the surface of the target chain to obtain a shot image, and preprocessing the shot image to obtain a preprocessed image; Dividing the preprocessed image into a plurality of regions of interest to obtain a chain sub-image to be detected corresponding to each region of interest; Performing defect recognition on the chain sub-image to be detected to obtain a defect recognition result; If the defect recognition result is that the chain sub-image to be detected has a chain defect, then a defect type analysis is performed on the chain sub-image to be detected with the chain defect to obtain the chain defect type; Develop corresponding solution strategies based on the chain defect types.
2. The chain identification method according to claim 1, characterized in that: The surface of the chain is photographed to obtain a photographed image, and the photographed image is preprocessed to obtain a preprocessed image, including: The surface of the chain is photographed at multiple angles to obtain images taken at multiple angles, and the images taken at multiple angles are combined into a multi-angle photographed image set; Performing image registration on the multi-angle shot image set to obtain a registered image set; Constructing a three-dimensional point cloud model according to the registered image set to obtain three-dimensional point cloud data of the chain surface; Performing noise reduction processing on the three-dimensional point cloud data to obtain noise-reduced three-dimensional point cloud data; Performing surface reconstruction based on the denoised three-dimensional point cloud data to obtain a three-dimensional mesh model of the chain surface; Perform a two-dimensional projection transformation on the three-dimensional grid model to obtain a preprocessed image.
3. The chain identification method according to claim 1, characterized in that: The pre-processed image is divided into a plurality of regions of interest to obtain a chain sub-image to be detected, including: Performing edge detection on the preprocessed image to obtain a chain edge contour map; Performing image segmentation on the preprocessed image according to the chain edge contour map to obtain the chain main body area; Applying morphological operations to the main region of the chain to obtain a chain skeleton graph; According to the chain skeleton diagram, the chain main body area is sampled at equal intervals to obtain a sampling point set; Taking each sampling point in the sampling point set as the center, construct a rectangular area of a fixed size to obtain a set of candidate regions of interest; Performing overlap analysis on the candidate region of interest set, merging regions with overlaps higher than a preset threshold, and obtaining multiple regions of interest; According to the multiple regions of interest, the pre-processed image is cropped and extracted to obtain multiple chain sub-images to be detected.
4. The chain identification method according to claim 1, characterized in that: Defect recognition is performed on the chain sub-image to be detected to obtain a defect recognition result, including: Performing multi-scale wavelet transform on the chain sub-image to be detected to obtain wavelet decomposition coefficients of multiple scales; According to the wavelet decomposition coefficients, adaptive threshold segmentation is performed on the chain sub-image to be detected to obtain a binary image; Performing morphological operations and connected domain analysis on the binary image to obtain candidate defect regions; Convert the candidate defect region into a vector format to obtain a converted format vector; The conversion format vector is input into a preset defect recognition algorithm to perform defect recognition and obtain a defect recognition result.
5. The chain identification method according to claim 1, characterized in that: The defect type analysis is performed on the sub-image of the chain to be detected corresponding to the chain defect to obtain the chain defect type, including: Performing a convolution transformation on the sub-image of the chain to be detected corresponding to the chain defect to obtain a convolution feature map; wherein the convolution feature map is equivalent to highlighting the tiny defects and material changes on the surface of the chain; Based on the convolution feature map, the chain sub-image to be detected is subjected to self-organized critical segmentation to obtain a multi-level segmentation result; wherein the multi-level segmentation result includes the chain surface, the chain link edge and the potential defect area; The multi-level segmentation results are filtered and constructed into a complex sequence to obtain continuous homology features; wherein the continuous homology features include topological changes of the chain surface, connectivity of defective areas, and relative positional relationships between chain links; Performing Betti number sequence calculation on the continuous homology feature to obtain a topological feature descriptor; wherein the topological feature descriptor includes area ratio, perimeter ratio, roundness, and rectangularity; Performing continuous landscape transformation on the topological feature descriptor to obtain a high-dimensional feature vector; wherein the high-dimensional feature vector is used to reflect the multi-scale characteristics and topological continuity of the chain defect; Through a preset defect type algorithm, deep learning and defect type analysis are performed on the chain sub-image to be detected based on the high-dimensional feature vector to obtain a defect recognition result.
6. The chain identification method according to claim 5, characterized in that: The Betti number sequence is calculated for the continuous homology feature to obtain a topological feature descriptor, including: Performing multi-scale filtering on the continuous coherent features to obtain a filtered continuous coherent feature sequence; Calculate the Betti number based on the filtered continuous coherent characteristic sequence to obtain a multi-scale Betti number sequence; constructing a continuous barcode based on the multi-scale Betti number sequence; Performing continuous entropy calculation based on the continuous barcode to obtain a topological complexity index of the chain defect; Adaptively threshold segmenting the continuous barcode based on the topological complexity index to extract significant topological features of the chain defect; Performing stable manifold learning on the significant topological features of the chain defect to obtain a low-dimensional topological representation of the chain defect; A feature vector of a chain defect is constructed based on the low-dimensional topological representation, and the feature vector of the chain defect is used as a topological feature descriptor.
7. The chain identification method according to claim 1, characterized in that: Develop corresponding solution strategies based on the chain defect types, including: Querying a preset defect-solution mapping table based on the chain defect type to obtain an initial solution set; Performing a feasibility assessment on each solution in the initial solution set to obtain a feasibility score; Sorting the initial solution set according to the feasibility score to obtain a sorted solution set; Selecting the top N solutions from the ranked solution set to construct a candidate solution set; Performing a cost-benefit analysis on each solution in the candidate solution set to obtain a cost-benefit ratio; An optimal solution is selected according to the cost-benefit ratio, and the optimal solution is used as a chain defect resolution strategy.
8. A chain identification device, characterized in that: include: A shooting module is used to shoot the surface of the target chain to obtain a shot image, and pre-process the shot image to obtain a pre-processed image; A division module, used for dividing the pre-processed image into a plurality of regions of interest to obtain a chain sub-image to be detected corresponding to each region of interest; An identification module is used to perform defect identification on the chain sub-image to be detected to obtain a defect identification result; An analysis module, configured to, if the defect recognition result is that the chain sub-image to be detected has a chain defect, perform defect type analysis on the chain sub-image to be detected having the chain defect, to obtain the chain defect type; A solution module is used to formulate a corresponding solution strategy based on the chain defect type.
9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.
10. 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 method according to any one of claims 1 to 7 are implemented.
Citation Information
Patent Citations
Device and method for automatically detecting quality of elevator balance compensation chain based on machine vision
CN104764750A
Machine vision-based elevator compensation chain place-R crack detection method
CN107610104A
Defect detection method of mobile phone camera bracket, computer equipment and storage medium
CN117274190A
Method, device and equipment for detecting surface defects of copper strip
CN118275449A
Cited By
Data processing method and related equipment based on edge AI computing and storage integrated chip
CN119762897A
Data Processing Method and Related Devices Based on Edge AI Computing and Storage Integrated Chip
CN119762897B
Intelligent nuclear power welding quality evaluation system based on ray image recognition
CN119784702A
A nuclear power welding quality intelligent evaluation system based on radiographic image recognition
CN119784702B
Product quality detection method, system and equipment applied to endoscope camera module
CN119958823A
