Shield cutter image recognition system based on FPGA morphological operator

Abstract

The invention relates to the technical field of image analysis, and provides a shield cutter image recognition system based on an FPGA morphological operator, and the system comprises the steps: obtaining a shield cutter image through an image obtaining module, and segmenting the shield cutter image into shield image blocks to carry out the parallel preprocessing of the shield cutter image; the morphological processing module constructs a morphological operation array to process the shield cutter image so as to extract contour information, edge gradient information and topological structure information of the shield cutter, and combines the topological structure information to fuse shield image blocks, so that the shape and structure characteristics of the shield cutter can be deeply analyzed, and the detection accuracy is improved. Key data support is provided for subsequent identification; the image recognition module judges the cutter type, the abrasion degree and the abrasion type of the shield cutter according to the contour information, the edge gradient information and the topological structure information of the shield cutter, and the accuracy and the reliability of shield cutter state monitoring are improved; and the result output module displays the cutter type, the wear degree and the wear type of the shield cutter and generates early warning information.

Description

Technical Field

[0001] This application relates to the field of image analysis technology, and particularly to a shield cutter image recognition system based on FPGA morphological operators. Background Art

[0002] During the shield construction process, as a key component, the state of the shield cutter directly affects the construction efficiency, quality, and safety. However, there are many problems in the field of shield cutter image recognition and condition monitoring. Traditional shield cutter image recognition technologies face great challenges in terms of accuracy. Many existing methods are difficult to accurately extract the characteristic information of the shield cutter, resulting in misjudgments when determining the cutter type, wear degree, and wear type (eccentric wear, chipping, and uniform wear) of the shield cutter. For example, relying solely on a single image feature for judgment. Moreover, conventional image acquisition and preprocessing methods cannot make full use of hardware resources, with a slow processing speed, unable to achieve parallel computing, and an overly long processing time that cannot meet the real-time requirements of shield construction. For example, during the shield construction process, the state of the shield cutter needs to be monitored in real time. If the image recognition processing speed is too slow, problems such as wear of the shield cutter cannot be detected in a timely manner, thus affecting the construction progress and safety. Summary of the Invention

[0003] Aiming at the deficiencies of the prior art, this application provides a shield cutter image recognition system based on FPGA morphological operators.

[0004] This application provides a shield cutter image recognition system based on FPGA morphological operators, which includes: an image acquisition module, a morphological processing module, an image recognition module, and a result output module;

[0005] The image acquisition module is used to acquire shield cutter images and divide the shield cutter images into shield image blocks according to the computing resources of the FPGA for parallel preprocessing of the shield cutter images;

[0006] The morphological processing module is used to adjust the structural element according to the edge direction and texture features of the shield image block, construct a morphological operation array to process the shield cutter image through the structural element, so as to extract the contour information and edge gradient information of the shield cutter, judge whether the pixel points of the shield image block are edge pixels based on the contour information and edge gradient information of the shield cutter, obtain the unrefined edge pixels, detect the neighborhood connection situation of the edge pixels in the shield image block to refine and extract topological structure information, and fuse the shield image blocks in combination with the topological structure information;

[0007] The image recognition module is used to judge the cutter type and wear degree of the shield cutter according to the contour information, edge gradient information, and topological structure information of the shield cutter;

[0008] The result output module is used to display the tool type and wear degree of the shield cutter, and generate a warning message.

[0009] As an optional implementation manner, the processing strategy of the shield cutter image includes:

[0010] Adapting to adjust the shape and size of the structural element according to the edge direction and texture feature of the shield image block;

[0011] Constructing a morphological operation array to perform morphological operations through the structural element, obtaining the contour information of the shield cutter, and extracting the edge gradient information of the shield cutter;

[0012] Refining and extracting the topological structure information of the shield cutter according to the contour information and edge gradient information of the shield cutter;

[0013] Fusing the shield image blocks according to the block index of the shield image block and combining with the topological structure information of the shield cutter.

[0014] As an optional implementation manner, the logic of adapting to adjust the shape and size of the structural element includes:

[0015] Determining the texture feature of each shield image block based on the local contrast of the shield image block, and determining the edge direction of each shield image block based on the gradient direction clustering of the shield image block;

[0016] Adjusting the size of the structural element according to the texture feature of the shield image block;

[0017] Adjusting the shape of the structural element according to the edge direction of the shield image block.

[0018] As an optional implementation manner, the execution logic of the morphological operation includes:

[0019] For each shield image block, taking the upper left pixel point of each shield image block as the sliding starting point in turn according to the shape and size of the structural element, and traversing one by one on the shield image block;

[0020] During the process of traversing the shield image block, sequentially executing the erosion operation array, dilation operation array, opening operation array and closing operation array to obtain the contour information of the shield cutter;

[0021] Extracting the edge gradient information of the shield cutter according to the contour information of the shield cutter.

[0022] As an optional implementation manner, the refinement extraction method of the topological structure information includes:

[0023] Judging whether the pixel point of each shield image block is an edge pixel based on the contour information and edge gradient information of the shield cutter;

[0024] Centering on the edge pixel, count the number of neighboring pixels of this edge pixel that are edge pixels to determine whether the edge pixel is thinned, and continuously iterate until the thinning termination condition is met;

[0025] Obtain the unthinned edge pixels, and detect the neighborhood connection situation of the edge pixels in each shield image block to determine the topological structure information of the shield cutter;

[0026] Sort out and label the topological structure information of the shield cutter.

[0027] As an alternative implementation, the detection logic of the neighborhood connection situation includes:

[0028] Traverse and count the unthinned edge pixels, and determine the information of the bifurcation points and intersection points through threshold judgment;

[0029] Search for edge pixels with a neighborhood connection number of 2 and a stable gradient direction as the starting points of the straight line segments. Use the gradient direction as the initial direction of the straight line segments, and sequentially check the unthinned edge pixels in the next neighborhood and the deviation angle between this edge pixel and the initial direction of the straight line segments along the gradient direction to judge the end points of the straight line segments and obtain the straight line segment information;

[0030] Search for edge pixels with a neighborhood connection number of 2 and an unstable gradient direction as the starting points of the curve segments. Use the gradient direction as the initial direction of the curve segments, and sequentially check the unthinned edge pixels in the next neighborhood and the curvature between adjacent unthinned edge pixels along the gradient direction to judge the end points of the curve segments and obtain the curve segment information.

[0031] As an alternative implementation, the judgment logic of the cutter type and wear degree of the shield cutter includes:

[0032] Construct a standard shield cutter model library;

[0033] Calculate and compare the similarity between the contour information of the shield cutter and the contour information in the standard shield cutter model library to obtain a shield cutter candidate list;

[0034] Compare the edge gradient information of the shield cutter with the edge gradient information in the standard shield cutter model library to narrow down the shield cutter candidate list;

[0035] Deeply match the topological structure information of the shield cutter with the topological structure information in the standard shield cutter model library to determine the cutter type of the shield cutter;

[0036] Judge the wear degree of the shield cutter according to the contour change and edge gradient change.

[0037] As an alternative implementation, obtain the shield cutter image through a camera, and the preprocessing logic of the shield cutter image includes:

[0038] Segment the shield cutter image according to the computing resources of the FPGA to obtain shield image blocks;

[0039] Establish a block index for each shield image block;

[0040] Execute the denoising operation and enhancement operation for each shield image block in parallel in the FPGA.

[0041] As an alternative implementation, the judgment sub-logic for the wear degree of the shield cutter includes:

[0042] Extract the contour area in the contour information, and calculate the contour deformation amount between the contour area of the shield cutter and the contour area in the standard shield cutter model library through the Euclidean distance;

[0043] Extract the gradient intensity in the edge gradient information, and calculate the gradient attenuation amount between the gradient intensity of the shield cutter and the gradient intensity in the standard shield cutter model library;

[0044] Configure the contour deformation threshold and the gradient attenuation threshold, compare the contour deformation amount with the contour deformation threshold to obtain the contour deformation degree, and compare the gradient attenuation amount with the gradient attenuation threshold to obtain the gradient attenuation degree;

[0045] Comprehensively judge the wear degree of the shield cutter according to the contour deformation degree and the gradient attenuation degree.

[0046] As an alternative implementation, the judgment sub-logic for the wear type of the shield cutter includes:

[0047] Mirror and flip along the contour central axis of the shield cutter, and calculate the contour symmetry before and after flipping;

[0048] Extract the gradient direction in the edge gradient information, count the distribution quantity of the gradient direction on the contour of the shield cutter, and calculate the gradient direction mean value of different shield image blocks to determine the gradient direction change;

[0049] Comprehensively judge the wear type of the shield cutter according to the contour symmetry and the gradient direction change. The wear types include eccentric wear, chipping and uniform wear.

[0050] Compared with the prior art, the beneficial effects of the present application are: The shield cutter image is obtained through the image acquisition module, and the shield cutter image is segmented into shield image blocks for parallel preprocessing of the shield cutter image, ensuring that a comprehensive and clear shield cutter image is obtained, and providing a high-quality image basis for subsequent processing, greatly improving the accuracy of image information and reducing the recognition error caused by image quality problems.

[0051] The morphological processing module constructs a morphological operation array to process the shield cutter images, so as to extract the contour information, edge gradient information and topological structure information of the shield cutter. Combining with the topological structure information, the shield image blocks are fused. For the complex shape of the shield cutter images, the contour information of the shield cutter can be efficiently extracted, and the edge gradient information and topological structure information can be accurately obtained, which helps to deeply analyze the shape and structure characteristics of the shield cutter and provides key data support for subsequent recognition.

[0052] The image recognition module judges the cutter type, wear degree and wear type of the shield cutter according to the contour information, edge gradient information and topological structure information of the shield cutter. Based on the various key information extracted by the morphological processing module, the cutter type, wear degree and wear type of the shield cutter can be accurately judged, improving the accuracy and reliability of the shield cutter state monitoring.

[0053] The result output module displays the cutter type, wear degree and wear type of the shield cutter, and generates a warning message. The identified cutter type, wear degree and wear type of the shield cutter are visually displayed, which is convenient for the staff to quickly understand the cutter state and can issue a warning for replacing the shield cutter in time. This helps to reasonably arrange the construction schedule, carry out equipment maintenance in advance, avoid construction accidents and delays caused by excessive cutter wear, and improve the safety and efficiency of shield construction.

[0054] Combining FPGA technology with morphological operators and applying them to the shield cutter image recognition scenario, using the reconfigurable logic resources and parallel processing capabilities of FPGA, the performance and efficiency of the image recognition system are improved, which is innovative and leading in the shield construction industry. At the same time, this image recognition system can meet the actual needs of shield cutter state monitoring and recognition during the shield construction process. By obtaining the cutter type, wear degree information and wear type information of the shield cutter in real time and accurately, and generating a warning in time, it ensures the equipment health and safety during the operation of the shield machine, provides strong support for the safe and efficient progress of shield construction, and has high practical application value and promotion significance. Description of the Drawings

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. Among them:

[0056] Figure 1 It is the system structure diagram of the shield cutter image recognition system based on FPGA morphological operators provided by the embodiments of the present application;

[0057] Figure 2 The processing strategy diagram of the shield cutter image for the shield cutter image recognition system based on FPGA morphological operators provided by the embodiments of this application;

[0058] Figure 3 The execution logic diagram of the morphological operation for the shield cutter image recognition system based on FPGA morphological operators provided by the embodiments of this application;

[0059] Figure 4 The refined extraction method diagram of the topological structure information for the shield cutter image recognition system based on FPGA morphological operators provided by the embodiments of this application. Detailed implementation manners

[0060] To make the objectives, technical solutions, and advantages of the embodiments of this application more obvious and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings of the specification. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all of the embodiments.

[0061] Embodiment 1

[0062] As Figure 1 shown, it is the system structure diagram of the shield cutter image recognition system based on FPGA morphological operators provided by the embodiments of this application. The system includes an image acquisition module, a morphological processing module, an image recognition module, and a result output module.

[0063] The image acquisition module is used to acquire and preprocess the shield cutter image.

[0064] Specifically, the shield cutter image is acquired through a camera, and the preprocessing logic of the shield cutter image includes:

[0065] The shield cutter image is segmented according to the computing resources of the FPGA to obtain shield image blocks;

[0066] A block index of each shield image block is established;

[0067] The denoising operation and enhancement operation of each shield image block are executed in parallel in the FPGA.

[0068] Obtain shield cutter images through multiple cameras, where the cameras include industrial cameras and 3D cameras. The industrial cameras are used to identify the cutter type, wear degree, and wear type of the shield cutter, and the 3D cameras are used to obtain the dimensional information of the shield cutter, taking pictures of the shield cutter from different angles. For example, cameras are set around the cutter head at a certain angular interval (such as 45 degrees) to ensure that all sides and key parts of the shield cutter can be covered, and ensure that all cameras take pictures at the same moment, guaranteeing that the obtained shield cutter images reflect the state of the shield cutter at the same instant. For each obtained shield cutter image, use the Scale-Invariant Feature Transform (SIFT) or Speeded-Up Robust Features (SURF) algorithm to extract the feature points in the shield cutter image. These feature points have scale invariance and rotation invariance, and can be accurately matched in shield cutter images with different perspectives. Use a feature point-based matching algorithm to find matching feature point pairs between different shield cutter images, and determine the spatial position relationship between different shield cutter images through the matching feature points. According to the spatial position relationship obtained by feature point matching, adopt a weighted average fusion algorithm to splice multiple shield cutter images into a complete shield cutter image. During the fusion process, reasonably weight the pixel values in the overlapping area to ensure that the spliced image has a natural and seamless transition.

[0069] Divide the spliced complete shield cutter image evenly according to a fixed size to obtain shield image blocks of the same size. For example, the size of each shield image block is n×n pixels, in order to adapt to the processing characteristics of the FPGA arithmetic unit, so as to ensure that each shield image block can be independently and efficiently processed simultaneously in different arithmetic units of the FPGA, thereby improving the overall processing efficiency. Build a unique block index identifier for each shield image block to facilitate the effective management and data transmission of the shield image blocks in the subsequent parallel processing process. The block index information covers the position coordinates of the shield image block in the original shield cutter image, etc., to ensure that the processed shield image blocks can be accurately restored to the complete shield cutter image. Through this way of shield image block division and index establishment, the parallel processing infrastructure of the shield cutter image is built, enabling the FPGA to efficiently process each shield image block independently, improving the overall processing speed, and at the same time ensuring that the processed shield image blocks can be accurately restored to the complete shield cutter image, providing a good foundation for subsequent morphological processing.

[0070] The type and intensity of noise in each shield image block are determined by calculating the variance of the pixel grayscale value in each shield image block. This is based on the characteristic differences of different noise types in the distribution of pixel grayscale values. According to the type and intensity of noise in the shield image block, a suitable denoising algorithm is selected and the denoising operation is performed. For example, for shield image blocks dominated by Gaussian noise, a parallel Gaussian filtering algorithm is used to perform filtering calculations in each shield image block at the same time. For shield image blocks with more salt and pepper noise, a parallel median filtering algorithm is used for denoising. The denoising operations of different shield image blocks are performed simultaneously in different computing units of the FPGA, which can greatly improve the denoising efficiency. After denoising, the originally blurred shield tool outline becomes clear, and the detail information in the shield tool image is retained, which improves the quality of the shield tool image and provides more accurate data for subsequent morphological processing.

[0071] After performing the denoising operation, in order to simplify the complexity of shield image block processing, grayscale conversion is performed in parallel on each shield image block, and the color shield image block is converted into a grayscale shield image block. The grayscale conversion process of each shield image block is independent of each other, making full use of the parallel computing resources of FPGA, improving the conversion speed, and saving time for the entire preprocessing process; for the grayscale shield image block, the contrast of each shield image block is enhanced in parallel through histogram equalization in FPGA. Through this parallel processing method, the detailed information of the shield image block can be better retained, making the edge and texture of the shield tool more prominent, and the processing speed can be improved while ensuring the processing effect, meeting the real-time requirements of the image recognition system.

[0072] The morphological processing module is used to construct a morphological operation array to process the shield tool image and extract the contour information, edge gradient information and topological structure information of the shield tool.

[0073] Specifically, Figure 2 As shown in the figure, the processing strategies of shield tool images include:

[0074] Adaptively adjust the shape and size of the structural element according to the edge direction and texture characteristics of the shield image block;

[0075] Construct a morphological operation array to perform morphological operations through structural elements to obtain the contour information of the shield tool and extract the edge gradient information of the shield tool;

[0076] Refine and extract the topological structure information of the shield tool according to the contour information and edge gradient information of the shield tool;

[0077] The shield image blocks are fused according to the block index of the shield image blocks and combined with the topological structure information of the shield cutter.

[0078] Further, the logic for adaptively adjusting the shape and size of the structural element includes:

[0079] Determine the texture feature of each shield image block based on the local contrast of the shield image block, and determine the edge direction of each shield image block based on the gradient direction clustering of the shield image block;

[0080] Adjust the size of the structural element according to the texture feature of the shield image block;

[0081] Adjust the shape of the structural element according to the edge direction of the shield image block.

[0082] Based on the characteristic that the standard deviation can reflect the degree of data dispersion, the texture feature of each shield image block is obtained by calculating the local standard deviation of the shield image block. If the local standard deviation of the shield image block is greater than or equal to the preset standard deviation threshold, it indicates that the texture of the shield image block is rough and there are many details. At this time, it is necessary to reduce the size of the structural element to better capture the details; on the contrary, when the local standard deviation of the shield image block is less than the preset standard deviation threshold, it indicates that the texture of the shield image block is smooth. At this time, it is necessary to increase the size of the structural element to improve the processing efficiency without losing important information. In this way, the accuracy and efficiency of the morphological operation for processing the texture feature of the shield image block are improved as a whole.

[0083] The gradient direction can reflect the direction of pixel intensity change in the shield image block. Therefore, the edge direction of each shield image block is obtained by calculating the gradient direction of the shield image block. The gradient directions of the shield image block are clustered, and the main direction after clustering is the edge direction of the shield image block. Select a structural element with a suitable shape according to the edge direction, so that the morphological operation can better fit the edge feature of the shield image block. When the edge direction of the shield image block is a single direction (horizontal, vertical or a certain slope direction), for example, when the edge direction of the shield image block is vertical, a structural element with a vertical shape, such as a vertical line shape, is selected. When using this vertical line-shaped structural element for morphological operation, it can better highlight the edge along the vertical direction and make the edge in the vertical direction clearer and more continuous; when the edge direction of the shield image block is multi-directional, such as twill, structural elements with shapes such as ellipses and rhombuses are used to adapt to the multi-directional edge direction, ensuring that the edges in all directions can be effectively processed and avoiding the situation of missing edge information due to the mismatch of the structural element shape; this provides a more reliable basis for accurately extracting the contour information, edge gradient information and topological structure information of the shield cutter in the subsequent process, and improves the accuracy of the entire image recognition system for extracting the image features of the shield cutter.

[0084] Further, the execution logic of the morphological operation is as Figure 3 shown, specifically including:

[0085] For each shield image block, taking the upper-left pixel point of each shield image block as the sliding starting point in sequence according to the shape and size of the structural element, traverse it one by one on the shield image block;

[0086] During the process of traversing the shield image block, perform the erosion operation array, dilation operation array, opening operation array, and closing operation array in sequence to obtain the contour information of the shield cutter;

[0087] Extract the edge gradient information of the shield cutter according to the contour information of the shield cutter.

[0088] For each shield image block, determine the starting position of the structural element generated by the morphological operator in the shield image block. Taking the upper-left pixel point of each shield image block as the starting point, in order to ensure that the structural element can act on each pixel area of the shield image block, so as to comprehensively perform morphological operations on the shield cutter image, slide and traverse row by row and column by column on each shield image block according to the shape and size of the structural element. When the structural element covers a certain pixel area of the shield image block, compare the pixel gray value at the corresponding position in the structural element with all the pixel gray values in this pixel area. For example, when the structural element is a 3×3 rectangle, compare the gray value of the central pixel point in this rectangle area with the 9 pixel gray values in this rectangle area, and then update the pixels of the shield image block. Through this traversal method, the morphological operation array can be applied to the shield image block, so as to realize the overall processing of the shield cutter image.

[0089] The pixel updates of different morphological operation arrays for shield image blocks are inconsistent. For example, in the erosion operation array, the minimum pixel gray value in the rectangular area is assigned to the pixel point of the shield image block corresponding to the central pixel point within the rectangular area, thus completing one erosion operation. When the structural operation slides through all positions on the shield image block, the erosion operation of this shield image block ends, and so on to complete the erosion operation of all shield image blocks. Its essence is to search for the shield cutter (cutter head) in the shield cutter image to preliminarily outline the main contour of the shield cutter; while in the dilation operation array, the maximum pixel gray value in the rectangular area is assigned to the pixel point of the shield image block corresponding to the central pixel point within the rectangular area, thus completing one dilation operation. When the structural operation slides through all positions on the shield image block, the dilation operation of this shield image block ends, and so on to complete the dilation operation of all shield image blocks. In the shield cutter image, some holes will appear after the erosion operation. The dilation operation can fill the holes that appear after erosion and enhance the contour information of the shield cutter. Its essence is to expand the pixel values around the shield cutter (cutter head) towards the center, thereby filling the holes and strengthening the contour; while the opening operation array first performs an erosion operation on the shield image block to remove small noise points and protruding parts in the shield image block and preliminarily outline the main contour of the shield cutter. On the basis of the shield image block after the erosion operation is completed, a dilation operation is performed to restore the parts eroded during the erosion process, and at the same time remove isolated noise and burrs, making the contour of the shield cutter smoother. The steps are the same as those of the above erosion operation array and dilation operation array; the closing operation array first performs a dilation operation on the shield image block. According to the steps of the dilation operation array, the edge part in the shield image block is expanded to fill small holes. Then, an erosion operation is performed on the shield image block after the dilation operation to contract the overly expanded part during the dilation process, connect the broken edges, and further improve the contour details of the shield cutter, thereby obtaining continuous and complete shield cutter contour information.

[0090] First, through the erosion operation, small noise points and protruding parts in the shield image block are removed, and the main contour of the shield cutter is preliminarily outlined. Immediately afterwards, a dilation operation is performed to fill the holes generated by the erosion, further strengthening the contour information of the shield cutter and making the contour more continuous and complete. These two basic morphological operations lay the foundation for obtaining contour information; on the basis of completing the erosion operation and dilation operation, opening operation and closing operation are performed. The opening operation removes isolated noise and burrs in the shield image block, making the contour smoother. The closing operation helps to connect the broken edges and further improve the contour details of the shield cutter. Through the combination of this series of morphological operation arrays, relatively accurate shield cutter contour information is gradually extracted. The synergistic effect of these morphological operation arrays makes the contour of the shield cutter clearer, more accurate and more complete in the image, providing a reliable basis for subsequent image recognition.

[0091] After completing the execution of the above morphological operation array, using the computing resources of the FPGA, an edge detection algorithm such as the Sobel operator is adopted to calculate the gradient intensity and gradient direction of each pixel point of the shield cutter profile. For example, for a pixel point on the profile of the shield cutter, calculate the gradient components of this pixel point in the horizontal and vertical directions. Assume that the horizontal direction gradient component G x = 20, and the vertical direction gradient component of this pixel point is G y = 30. Then the gradient intensity is . After calculation, the gradient intensity of this pixel point is approximately 36.06, and the gradient direction is , where represents the two-parameter arctangent function. By considering the sign signals of the two parameters in the parentheses, the quadrant where the angle is located can be determined more accurately, so that the gradient direction of this pixel point can be obtained to be approximately 56.31 degrees. By calculating the gradient intensity and gradient direction of each pixel point in the entire shield cutter image, the edge gradient information of the shield cutter is obtained. Since the shield image block has been preliminarily processed by morphological operations to remove noise and interference, the edge gradient information calculated at this time is more accurate and prominent, which can better reflect the edge characteristics of the shield cutter, contribute to subsequent judgment of the cutter type and wear degree analysis of the shield cutter, etc., and improve the accuracy and reliability of the entire image recognition system.

[0092] Furthermore, as Figure 4 shown, the refined extraction method of the topological structure information includes:

[0093] Based on the profile information and edge gradient information of the shield cutter, judge whether each pixel point of the shield image block is an edge pixel;

[0094] Taking the edge pixel as the center, count the number of edge pixels in the neighborhood pixels of this edge pixel to judge whether this edge pixel is refined, and continuously iterate until the refinement termination condition is met;

[0095] Obtain the unrefined edge pixels, and detect the neighborhood connection situation of the edge pixels in each shield image block to determine the topological structure information of the shield cutter;

[0096] Sort out and label the topological structure information of the shield cutter.

[0097] The contour information of the shield cutter directly reflects the outer boundary of the shield cutter, while the edge gradient information of the shield cutter reflects the areas where the pixel intensity changes significantly. These areas usually correspond to the contour edges of the shield cutter. When the contour information and edge gradient information of the shield cutter are extracted, each pixel point on the contour of the shield cutter is an edge pixel. Starting from the upper left corner of the overall shield cutter image, each pixel point in the shield image block is scanned one by one to determine whether the pixel point is an edge pixel. If it is not an edge pixel, the pixel point is skipped and the next pixel point is scanned. For edge pixels, the situation of their neighboring pixel points is further analyzed.

[0098] Taking this edge pixel as the center, consider the situation of the 8 neighboring pixel points of this edge pixel to determine whether this edge pixel can be thinned, that is, deleted. This is based on the basic characteristics of the topological structure. Non-critical edge pixels often have fewer and specifically distributed neighboring edge pixels. For example, only two adjacent pixel points among the 8 neighboring pixel points of this edge pixel are edge pixels, and these two adjacent pixel points are not diagonally adjacent. At the same time, if this edge pixel is not an endpoint pixel of the shield image block where it is located, then this edge pixel can be thinned. The endpoint pixel can be determined by judging that there is only one edge pixel in this shield image block. During the judgment process, attention should be paid to avoiding misdeleting intersection points. It is necessary to check whether this edge pixel conforms to the characteristics of intersection points. These intersection points are crucial for maintaining the integrity of the topological structure. If it conforms, then this edge pixel is retained.

[0099] According to the above process of thinning edge pixels, after all pixel points in the shield cutter image are scanned and processed once, one iteration is completed. Check whether the thinning termination condition is satisfied. If not, start again from the upper left corner of the overall shield cutter image, repeat the process of scanning each pixel point in the shield image block one by one and making thinning judgments, and perform the next iteration until the thinning termination condition is satisfied. In this way, unimportant edge pixels can be gradually removed. The thinning termination condition is to stop the iteration when the maximum number of iterations (such as 100 times) or the number of changes in the edge pixels in the shield cutter image is less than a certain threshold (such as 5 pixel points) after several consecutive iterations (such as 3 times). After multiple iterations, the finally remaining unthinned edge pixels are the key pixels that can represent the topological structure of the shield cutter.

[0100] Based on the neighborhood connection of each shield image block detected by the last remaining unrefined edge pixels, the topological structure information of the shield cutter is determined. The topological structure information includes intersection points, bifurcation points, and topological structure features, and the topological structure features include straight line segments and curve segments. For example, the number of pixel neighborhood connections at intersection points and bifurcation points is relatively large and shows a specific distribution pattern, while straight line segments and curve segments can be judged by the direction and continuity of adjacent edge pixels. These topological structure information fully reflects the shape and structural characteristics of the shield cutter, providing key data support for subsequent image recognition.

[0101] Sort out the extracted topological structure information, classify and store information such as the positions and attributes of branch points, intersection points, and topological structure features. Mark each extracted topological structure feature. For example, assign a unique number to each branch point and intersection point, and also perform corresponding encoding for different shape features. For example, the straight line segment is L and the curve segment is C, and store the topological structure information in an array. When it is convenient to perform image recognition later to compare and judge the cutter type and wear degree of the shield cutter, it can be quickly and accurately called and compared, improving the accuracy and efficiency of judging the cutter type and wear degree of the shield cutter.

[0102] Furthermore, the detection logic of the neighborhood connection includes:

[0103] Traverse and count the unrefined edge pixels, and determine the information of bifurcation points and intersection points through threshold judgment;

[0104] Search for edge pixels with a neighborhood connection number of 2 and a stable gradient direction as the starting point of the straight line segment. Use the gradient direction as the initial direction of the straight line segment, and sequentially check the unrefined edge pixels in the next neighborhood along the gradient direction and the deviation angle between the edge pixel and the initial direction of the straight line segment to judge the end point of the straight line segment and obtain the straight line segment information;

[0105] Search for edge pixels with a neighborhood connection number of 2 and an unstable gradient direction as the starting point of the curve segment. Use the gradient direction as the initial direction of the curve segment, and sequentially check the unrefined edge pixels in the next neighborhood along the gradient direction and the curvature between adjacent unrefined edge pixels to judge the end point of the curve segment and obtain the curve segment information.

[0106] For each unthinned edge pixel in the shield image block, starting from the upper left corner of the shield image block, traverse each unthinned edge pixel encountered. For each such pixel, check each pixel point in its 8-neighborhood, determine whether each pixel point in the 8-neighborhood is also an unthinned edge pixel, and count the number of unthinned edge pixels among the 8-neighborhood pixel points. If the number of unthinned edge pixels among the 8-neighborhood pixel points is greater than or equal to 3, then this unthinned edge pixel is determined to be a fork point or a crossing point. Continuously traverse to obtain all the crossing points and fork points, and store them in the fork point list. The fork point list records the pixel coordinates of each crossing point or fork point. And based on the angular relationship and positional relationship between these edge pixels, it can be distinguished whether the edge pixel is a fork point or a crossing point. For example, if three or more edge pixels extend from this edge pixel in different directions, it is a fork point; if a pattern of two edges intersecting each other appears, it is a crossing point. This can effectively identify the key nodes in the topological structure of the shield cutter.

[0107] Traverse all unthinned edge pixels in the shield image block again. Since unthinned edge pixels will be at one end of a straight line segment, search for edge pixels with a neighborhood connection number of 2 and a stable gradient direction as the starting points of the straight line segments. Here, the stability of the gradient direction can be judged by calculating the included angle between the lines connecting two unthinned edge pixels in the neighborhood and the central pixel point. If the included angle is within a certain threshold range, such as less than 15 degrees, then the gradient direction is considered stable, and these edge pixels are used as the starting points of the straight line segments.

[0108] For each starting point of the straight line segment found, use the gradient direction as the initial direction of the straight line segment. Along the initial direction of the straight line segment, with a step size of moving one pixel each time, based on the linear feature of the straight line segment, determine the range of the straight line segment by tracking the direction consistency of the edge pixels. Sequentially check the unthinned edge pixels in the next neighborhood, calculate the deviation angle between the newly checked unthinned edge pixel and the initial direction of the current straight line segment. When the deviation angle is greater than a certain threshold, such as 15 degrees, then it is considered that a straight line segment is found, and the newly checked unthinned edge pixel is the end point of the straight line segment. Thus, the straight line segment information in the topological structure of the shield cutter can be accurately obtained, and the straight line segment information is stored in the straight line segment list.

[0109] Traverse all unthinned edge pixels in the shield image block again. Search for edge pixels with a neighborhood connection number of 2 and an unstable gradient direction as the starting points of the curve segments. Here, the instability of the gradient direction can be judged by calculating the included angle between the lines connecting two unthinned edge pixels in the neighborhood and the central pixel point. If the included angle is greater than 30 degrees, then the gradient direction is considered unstable, and these edge pixels are used as the starting points of the curve segments.

[0110] For the starting point of each found curve segment, using the gradient direction as the initial direction of the curve segment, along the initial direction of the curve segment, based on the bending characteristics of the curve segment, determine the range of the curve segment by tracking the curvature change of the edge pixels. Successively check the curvature between the unthinned edge pixels in the next neighborhood and between adjacent unthinned edge pixels. The curvature can be estimated by the angular change between adjacent unthinned edge pixels. For example, if the angular change between adjacent unthinned edge pixels is , and the straight-line distance between adjacent unthinned edge pixels is , then the curvature between adjacent unthinned edge pixels is . Thus, continuously calculate the curvature between adjacent unthinned edge pixels. When the reciprocal of the curvature is greater than the set minimum curvature radius, where the reciprocal of the curvature is the curvature radius (at this time ), it means that the bending degree of the curve in this section is relatively small, meeting the set end condition of the curve segment, then it is considered that a curve segment is found. Then the last unthinned edge pixel is the end point of the curve segment, so as to accurately obtain the curve segment information in the shield tool topology structure and store the curve segment information in the curve segment list.

[0111] Through the above detection logic of neighborhood connection conditions, key features such as bifurcation points, intersection points, straight line segments, and curve segments in the shield tool topology structure can be accurately extracted. These key features completely describe the shape and structural characteristics of the shield tool, providing accurate data support for subsequent image recognition. Store the extracted bifurcation point, intersection point, straight line segment information, and curve segment information in the corresponding lists respectively, making this topological structure information more orderly and easy to manage. During the subsequent image recognition process, these topological structure information can be quickly called and compared, greatly improving the accuracy and efficiency of judging the tool type and wear degree of the shield tool.

[0112] After successively completing the adaptive adjustment of the structural element in the morphological operation, the execution of the morphological operation, and the extraction of the shield tool topology structure information, according to the block index of each shield image block, preliminarily arrange the block indexes of all shield image blocks according to their positions in the original shield tool image to form a preliminary image stitching framework. On the basis of the preliminary stitching, use the topological structure information for topological structure alignment.

[0113] For example, if a straight line segment is found on the right edge of the shield image block A, and a straight line segment is also found on the left edge of the shield image block B, and these two straight line segments should be continuous in the topological structure, by finely adjusting the position of the shield image block B, the two straight line segments are accurately docked; for bifurcation points and intersection points, similar matching and alignment operations are also carried out. If there is a bifurcation point at the lower right corner of the shield image block A, and there is a corresponding bifurcation point at the upper left corner of the shield image block C, ensure that the positions of the two bifurcation points are consistent after splicing, and the surrounding topological structures can be naturally connected.

[0114] After completing the position matching and topological structure alignment, the overlapping area of adjacent shield image blocks is fused. For example, for the overlapping area of the shield image blocks A and B, according to the topological structure information, judge how the pixels in the overlapping area should be fused. If the topological structure in the overlapping area shows that there should be a continuous curve segment, then during fusion, smooth transition processing is performed on the pixels in the overlapping area, so that the curve segment looks natural and continuous after splicing, without obvious splicing marks, and the image quality is improved, which helps to more clearly observe and analyze the characteristics of the shield cutter. Methods such as weighted average can be used to calculate the pixel values of the overlapping area to obtain the fused pixel values.

[0115] By combining the block index and topological structure information for fusion, the integrity of the shield cutter image is ensured, so that all parts of the shield cutter are accurately spliced together, and topological features such as bifurcation points, intersection points, straight line segments, and curve segments can be naturally connected and presented, without obvious breaks or misalignments caused by block processing, providing complete image data for subsequent image recognition to judge the cutter type and wear degree of the shield cutter.

[0116] The image recognition module is used to judge the cutter type, wear degree, and wear type of the shield cutter according to the contour information, edge gradient information, and topological structure information of the shield cutter.

[0117] Specifically, the judgment logic for the cutter type, wear degree, and wear type of the shield cutter includes:

[0118] Construct a standard shield cutter model library;

[0119] Calculate and compare the similarity between the contour information of the shield cutter and the contour information in the standard shield cutter model library to obtain a list of shield cutter candidates;

[0120] Compare the edge gradient information of the shield cutter with the edge gradient information in the standard shield cutter model library to narrow down the list of shield cutter candidates;

[0121] Deeply match the topological structure information of the shield cutter with the topological structure information in the standard shield cutter model library to determine the cutter type of the shield cutter;

[0122] Comprehensively judge the wear degree and wear type of the shield cutter according to the profile information and edge gradient information.

[0123] The standard shield cutter model library is the basis for subsequent comparison and judgment. It contains the profile information, edge gradient information, and topological structure information of various types of shield cutters in the normal state. By comparing with the standard shield cutter model library, the type, wear degree, and wear type of the cutter to be detected can be accurately judged.

[0124] Collect the profile information, edge gradient information, and topological structure information of shield cutters of various different cutter types to construct a standard shield cutter model library. These information represent the typical characteristics of shield cutters of different cutter types and are the basis for subsequent comparison and judgment. By establishing a standard shield cutter model library, the image features of the actually collected shield cutters can be compared with it, so as to identify the cutter type, wear degree, and wear type of the shield cutter. The cutter types include hob, scraper, and cutter, etc. For example, through multiple images under different angles and lighting conditions, after processing, the profile information, edge gradient information, and topological structure information of shield cutters of different cutter types are extracted. The profile of the hob is relatively round, and the edge gradient is relatively strong in some parts. The cutter has a unique forked structure, and its topological structure has obvious branch point characteristics. The edge of the scraper is relatively straight, and the straight line segment accounts for a large proportion in the topological structure. After organizing these information, it is stored in the standard shield cutter model library.

[0125] The profile information is the basic shape feature of the shield cutter. Shield cutters of different cutter types usually have different profile shapes. By calculating the similarity between the profile information of the shield cutter and the profile information in the standard shield cutter model library, the cutter types similar to the current shield cutter profile can be initially screened out to form a shield cutter candidate list. Commonly used similarity calculation methods such as Euclidean distance and cosine similarity can quantify the difference degree between two profiles. For example, the calculation results show that the similarity between the profile information of this shield cutter and the profile information of the scraper and hob is greater than the preset threshold, then the scraper and hob are included in the shield cutter candidate list.

[0126] Shield cutters of different tool types also have differences in edge features. Even shield cutters with similar contours may have different edge gradient information due to different designs and manufacturing processes. The edge gradient information of the shield cutter is compared with the edge gradient information in the standard shield cutter model library to further narrow the candidate list of shield cutters and exclude some tool types that have only similar contours but mismatched edge gradient information. For example, the edge gradient information of the shield cutter is compared with the edge gradient information of the scraper and hob in the standard shield cutter model library. It is found that the distribution and intensity of the edge gradient of the shield cutter in some key parts are more consistent with the hob, but there are significant differences with the scraper. Therefore, the scraper is excluded from the candidate list, leaving only the hob.

[0127] Topological structure information is a representation of the internal structure and shape of the shield cutter. Shield cutters of different tool types often have unique patterns in topological structure, which have high recognition. By deeply matching the topological structure information of the shield cutter with the topological structure information in the standard shield cutter model library, the tool type of the shield cutter can be accurately determined; for example, the topological structure information of the shield cutter is further deeply matched with the topological structure information of the roller cutter, and the branch points, intersection points, straight line segment information and curve segment information are carefully checked. It is found that the topological structures of the two are completely consistent, thereby determining that the tool type of the shield cutter is a roller cutter; by constructing a standard shield cutter model library and comparing the contour information, edge gradient information and topological structure information in turn, the tool type of the shield cutter can be accurately determined. This step-by-step screening and precise matching method greatly improves the accuracy of tool type judgment and reduces the possibility of misjudgment.

[0128] Contour information and edge gradient information can reflect the wear of the tool. Comprehensive consideration of these two types of information can more accurately determine the degree and type of wear. In the actual use of shield tools, the contour and edge gradient of the shield tool will change. Wear will cause contour deformation and edge blunting, and the shield tool needs to be replaced. These changes will be reflected in the contour information and edge gradient information. By comparing the contour information and edge gradient information of the current shield tool with the contour information and edge gradient information under normal conditions in the standard shield tool model library, the degree of wear of the shield tool can be determined. For example, the degree of deformation of the contour and the weakening of the edge gradient strength can be used as the basis for judging the degree of wear.

[0129] Furthermore, the judgment sub-logic of the wear degree of the shield tool includes:

[0130] Extract the contour area from the contour information, and calculate the contour deformation variable of the contour area of the shield tool and the contour area in the standard shield tool model library through the Euclidean distance;

[0131] Extract the gradient intensity from the edge gradient information, and calculate the gradient attenuation amount between the gradient intensity of the shield cutter and the gradient intensity in the standard shield cutter model library;

[0132] Configure the contour deformation threshold and the gradient attenuation threshold. Compare the contour deformation amount with the contour deformation threshold to obtain the contour deformation degree, and compare the gradient attenuation amount with the gradient attenuation threshold to obtain the gradient attenuation degree;

[0133] Comprehensively judge the wear degree of the shield cutter according to the contour deformation degree and the gradient attenuation degree.

[0134] The change in the contour area is an important manifestation of the wear of the shield cutter. By calculating the contour deformation amount, the wear degree of the shield cutter can be quantified; for the extracted contour, use the pixel counting method to calculate the number of pixels it contains, multiply the number of pixels by the actual area corresponding to a single pixel, and the actual area corresponding to a single pixel can be obtained by calibrating the resolution and actual size of the image, so as to obtain the actual area of the contour. Find the contour area of the standard shield cutter with the same type as the cutter to be detected in the standard shield cutter model library. Let the contour area of the cutter to be detected be A1 and the contour area of the standard shield cutter be A0, then the contour deformation amount is |A1 - A0|; thus, the change in the contour area of the shield cutter is quantified, making the wear degrees between different shield cutters comparable, facilitating the subsequent setting of thresholds for judgment, providing specific values for subsequent comparison with the contour deformation threshold, and serving as the basis for judging the contour deformation degree.

[0135] The attenuation of the gradient intensity is also an important manifestation of the wear of the shield cutter. By calculating the gradient attenuation amount, the wear degree of the shield cutter can be further quantified; obtain the gradient intensity G1 of the shield cutter from the above steps, obtain the gradient intensity G0 of the corresponding standard shield cutter from the standard shield cutter model library, and calculate the gradient attenuation amount as |G1 - G0|; it quantifies the wear condition on the surface of the shield cutter from a microscopic perspective. Combined with the change in the contour area, it can more comprehensively reflect the wear degree of the shield cutter, provides specific values for subsequent comparison with the gradient attenuation threshold, and is used to judge the gradient attenuation degree.

[0136] By setting the contour deformation threshold and the gradient attenuation threshold, the contour deformation amount and the gradient attenuation amount can be converted into specific deformation degrees and attenuation degrees, which is convenient for comprehensively judging the wear degree. A large number of shield cutter samples with different wear degrees are collected, and the contour deformation amount and the gradient attenuation amount of the shield cutter samples are calculated respectively. The ranges of the contour deformation amount and the gradient attenuation amount corresponding to different wear degrees (mild wear, moderate wear, and severe wear) are determined through the standard deviation, so as to determine the contour deformation threshold and the gradient attenuation threshold. The contour deformation threshold includes the contour deformation base value (mild deformation threshold) and the contour deformation extreme value (moderate deformation threshold), and the gradient attenuation threshold includes the gradient attenuation base value (mild attenuation threshold) and the gradient attenuation extreme value (moderate attenuation threshold). The contour deformation amount is compared with the contour deformation threshold, and the gradient attenuation amount is compared with the gradient attenuation threshold to obtain the quantified contour deformation degree (mild, moderate, and severe) and gradient attenuation degree (mild, moderate, and severe) respectively. The quantified wear index is converted into an intuitive degree level, which is convenient for the operator to quickly understand the wear condition of the shield cutter and provides clear degree level information for comprehensively judging the wear degree.

[0137] Comprehensively considering the contour deformation degree and the gradient attenuation degree can more comprehensively and accurately judge the wear degree of the worn cutter. According to practical experience and experimental data, the following comprehensive judgment rules are formulated: If both the contour deformation degree and the gradient attenuation degree are mild, the wear degree is mild; if one is moderate and the other is mild, the wear degree is moderate; if one is severe, or both are moderate, the wear degree is severe. Thus, the accuracy and reliability of the wear degree judgment are improved, the limitation of single-index judgment is avoided, and accurate wear degree information is provided for the result output module for display and generating warning information.

[0138] Furthermore, the judgment sub-logic of the wear type of the shield cutter includes:

[0139] Mirror flip along the contour central axis of the shield cutter and calculate the contour symmetry before and after flipping.

[0140] Extract the gradient direction in the edge gradient information, count the distribution quantity of the gradient direction on the contour of the shield cutter, and calculate the gradient direction mean value of different shield image blocks to determine the gradient direction change.

[0141] Comprehensively judge the wear type of the shield cutter according to the contour symmetry and the gradient direction change. The wear types include eccentric wear, chipping, and uniform wear.

[0142] Different wear types will cause varying degrees of changes in the symmetry of the tool profile. Calculating the profile symmetry can provide an important basis for judging the wear type. Find the central axis of the profile of the shield tool, mirror the profile along the central axis, and then calculate the similarity of the profile before and after flipping, such as the proportion of the overlapping area, as a measure of the profile symmetry. Thus, the symmetry of the shield tool profile is quantified, providing a specific numerical basis for the subsequent judgment of the wear type and specific numerical information for judging the wear type based on the profile symmetry.

[0143] Different wear types will result in different distributions and change situations of the edge gradient direction of the tool. By counting the distribution quantity of the gradient direction and calculating the mean value, this change can be reflected, providing a basis for judging the wear type. Extract the edge gradient direction information of the tool to be detected, divide the gradient direction into several intervals, for example, divide 0 - 180 degrees into 10 intervals, count the number of pixels of the gradient direction in each interval, and at the same time calculate the mean value of the gradient direction in each shield image block. Compare the difference in the mean values of the gradient directions of adjacent shield image blocks, count the number of shield image blocks with a difference greater than the difference threshold (such as 15°), and determine the change situation of the gradient direction. Thus, the distribution and change situation of the gradient direction are quantified, providing a specific numerical basis for the subsequent judgment of the wear type and specific numerical information for judging the wear type based on the change of the gradient direction.

[0144] Considering both the profile symmetry and the change of the gradient direction can more accurately judge the wear type of the shield tool. According to the combination of the profile symmetry and the change of the gradient direction, formulate judgment rules. If the profile symmetry is poor (obtained by comparing the profile symmetry with the symmetry threshold), it is partial wear. If the profile symmetry is not poor, then judge the change of the gradient direction. If the change of the gradient direction is large (obtained by comparing the number of shield image blocks with a difference greater than the difference threshold with the quantity threshold), it is chipping. If the change of the gradient direction is uniform, it is uniform wear. Thus, the accuracy of the wear type judgment is improved, providing accurate wear type information for the result output module for display and generating warning information.

[0145] Meanwhile, in practical applications, if it is recognized that the contour direction of the hob does not change for a long time, it can also be judged as tool wear. For example, when the hob is stuck and only one side wears while the other side remains unchanged. Then, we need to first judge according to the above-mentioned sub-logic for judging the wear type of the shield tool. If the preliminary judgment result cannot directly obtain the judgment results of wear types such as uneven wear, chipping, and uniform wear, then the change time of the contour direction needs to be introduced for secondary judgment. For example, a detection time period (such as 5 minutes) is set. When it is detected that the hob rotates continuously within this detection time period but the contour direction does not change, it is judged that the final wear type of the hob is uneven wear, thus improving the accuracy and reliability of the wear type judgment and ensuring the safe progress of shield construction.

[0146] The result output module is used to display the tool type, wear degree, and wear type of the shield tool and generate a warning message.

[0147] After determining the tool type, wear degree, and wear type of the shield tool in the image recognition module, corresponding warning messages need to be generated according to the wear degree. Different warning levels are preset for different wear degrees. For example, mild wear corresponds to a yellow warning, moderate wear corresponds to an orange warning, and severe wear corresponds to a red warning. Such a setting is to enable the operator to intuitively understand the wear condition of the shield tool so as to take corresponding measures and judge whether it is necessary to replace the shield tool. For example, if the above judgment shows that the shield tool is mildly worn, then the warning message is a yellow warning.

[0148] To enable the operator to comprehensively understand the situation of the shield tool, it is necessary to integrate and display the tool type, wear degree, wear type, and warning message of the shield tool, so that the operator can obtain key information at a glance and make reasonable decisions. For example, display on the display screen "Tool type: hob, wear degree: mild wear, wear type: uniform wear, warning message: yellow warning", and at the same time output these information in the form of a data report for easy recording and archiving.

[0149] To ensure that relevant operators can be informed of the wear condition of the shield tool in a timely manner, it is necessary to send warning messages through various methods, which helps to take maintenance measures in a timely manner and avoid affecting the construction progress and safety due to excessive wear of the shield tool. For example, through the built-in notification function of the system, send a push notification to the mobile APP of relevant operators, with the content "The hob shows mild wear and the wear type is uniform wear. Please pay attention to its usage", and at the same time issue a sound prompt in the monitoring center at the construction site to remind the duty personnel to pay attention.

[0150] By integrating and displaying the tool type, wear degree, wear type and warning information of the shield cutter and notifying through multiple channels, relevant operators can obtain the status information of the shield cutter in a timely and accurate manner. Whether on the construction site or under remote monitoring, they can conveniently understand the situation of the shield cutter, improving the efficiency and accuracy of information transmission; the clear warning information, detailed wear degree and wear type display provide clear decision-making basis for operators. In the mild wear stage, operators can arrange the maintenance plan in advance and prepare the relevant matters for replacing the cutter, avoiding the aggravation of shield cutter wear, which helps to ensure the safety and efficiency of shield construction and reduce the construction delay and accident risks caused by cutter problems.

Claims

1. A shield cutter image recognition system based on FPGA morphological operators, characterized in that Including: An image acquisition module, a morphological processing module, an image recognition module, and a result output module; The image acquisition module is used to acquire a shield cutter image and divide the shield cutter image into shield image blocks according to the computing resources of the FPGA for parallel preprocessing of the shield cutter image; The morphological processing module is used to adjust the structural element according to the edge direction and texture features of the shield image block, construct a morphological operation array to process the shield cutter image through the structural element, extract the contour information and edge gradient information of the shield cutter, judge whether the pixel points of the shield image block are edge pixels based on the contour information and edge gradient information of the shield cutter, obtain the unrefined edge pixels, detect the neighborhood connection situation of the edge pixels in the shield image block to refine and extract the topological structure information, and fuse the shield image blocks in combination with the topological structure information; The image recognition module is used to judge the cutter type, wear degree, and wear type of the shield cutter according to the contour information, edge gradient information, and topological structure information of the shield cutter; The result output module is used to display the cutter type, wear degree, and wear type of the shield cutter and generate a warning message.

2. The shield cutter image recognition system based on the FPGA morphological operator according to claim 1, characterized in that The processing strategy of the shield cutter image includes: Adaptive adjustment of the shape and size of the structural element according to the edge direction and texture features of the shield image block; Constructing a morphological operation array to perform morphological operations through the structural element to obtain the contour information of the shield cutter and extract the edge gradient information of the shield cutter; Refining and extracting the topological structure information of the shield cutter according to the contour information and edge gradient information of the shield cutter; Fusing the shield image blocks according to the block index of the shield image block and in combination with the topological structure information of the shield cutter.

3. The shield cutter image recognition system based on the FPGA morphological operator according to claim 2, characterized in that, The logic for adaptively adjusting the shape and size of the structural element includes: Determining the texture feature of each shield image block based on the local contrast of the shield image block, and determining the edge direction of each shield image block based on the gradient direction clustering of the shield image block; Adjusting the size of the structural element according to the texture feature of the shield image block; Adjusting the shape of the structural element according to the edge direction of the shield image block.

4. The shield cutter image recognition system based on the FPGA morphological operator according to claim 3, characterized in that The execution logic of the morphological operation includes: For each shield image block, sequentially use the upper left corner pixel point of each shield image block as the sliding starting point according to the shape and size of the structural element, and traverse the shield image block one by one; During the process of traversing the shield image block, sequentially execute the erosion operation array, dilation operation array, opening operation array, and closing operation array to obtain the contour information of the shield cutter; Extracting the edge gradient information of the shield cutter according to the contour information of the shield cutter.

5. The shield cutter image recognition system based on the FPGA morphological operator according to claim 4, characterized in that, The method for refining and extracting the topological structure information includes: Judging whether the pixel points of each shield image block are edge pixels based on the contour information and edge gradient information of the shield cutter; Taking the edge pixel as the center, counting the number of neighborhood pixel points of the edge pixel that are edge pixels to judge whether the edge pixel is refined, and continuously iterating until the refinement termination condition is met; Obtaining the unrefined edge pixels, detecting the neighborhood connection situation of the edge pixels in each shield image block to determine the topological structure information of the shield cutter; Sorting out and marking the topological structure information of the shield cutter.

6. The shield cutter image recognition system based on the FPGA morphological operator according to claim 5, characterized in that, The detection logic of the neighborhood connection situation includes: Traverse and count the unthinned edge pixels, and determine the information of the fork points and intersection points through threshold judgment; Search for edge pixels with a neighborhood connection number of 2 and a stable gradient direction as the starting points of the straight line segments. Use the gradient direction as the initial direction of the straight line segments, and sequentially check the unthinned edge pixels in the next neighborhood along the gradient direction and the deviation angle between the edge pixels and the initial direction of the straight line segments to judge the end points of the straight line segments and obtain the straight line segment information; Search for edge pixels with a neighborhood connection number of 2 and an unstable gradient direction as the starting points of the curve segments. Use the gradient direction as the initial direction of the curve segments, and sequentially check the unthinned edge pixels in the next neighborhood along the gradient direction and the curvature between adjacent unthinned edge pixels to judge the end points of the curve segments and obtain the curve segment information.

7. The shield cutter image recognition system based on the FPGA morphological operator according to claim 6, characterized in that, The judgment logic of the tool type, wear degree, and wear type of the shield cutter includes: Build a standard shield cutter model library; Perform similarity calculation and comparison between the contour information of the shield cutter and the contour information in the standard shield cutter model library to obtain a shield cutter candidate list; Compare the edge gradient information of the shield cutter with the edge gradient information in the standard shield cutter model library to narrow down the shield cutter candidate list; Deeply match the topological structure information of the shield cutter with the topological structure information in the standard shield cutter model library to determine the tool type of the shield cutter; Comprehensively judge the wear degree and wear type of the shield cutter according to the contour information and edge gradient information.

8. The shield cutter image recognition system based on the FPGA morphological operator according to claim 7, characterized in that, Obtain the shield cutter image through a camera. The preprocessing logic of the shield cutter image includes: Segment the shield cutter image according to the computing resources of the FPGA to obtain shield image blocks; Establish a block index for each shield image block; Parallelly execute the denoising operation and enhancement operation for each shield image block in the FPGA.

9. The shield cutter image recognition system based on the FPGA morphological operator according to claim 8, wherein, The judgment sub-logic of the wear degree of the shield cutter includes: Extract the contour area in the contour information, and calculate the contour deformation amount between the contour area of the shield cutter and the contour area in the standard shield cutter model library through the Euclidean distance; Extract the gradient intensity in the edge gradient information, and calculate the gradient attenuation amount between the gradient intensity of the shield cutter and the gradient intensity in the standard shield cutter model library; Configure the contour deformation threshold and gradient attenuation threshold, compare the contour deformation amount with the contour deformation threshold to obtain the contour deformation degree, and compare the gradient attenuation amount with the gradient attenuation threshold to obtain the gradient attenuation degree; Comprehensively judge the wear degree of the shield cutter according to the contour deformation degree and gradient attenuation degree.

10. The shield cutter image recognition system based on the FPGA morphological operator according to claim 9, characterized in that, The judgment sub-logic of the wear type of the shield cutter includes: Mirror and flip along the contour central axis of the shield cutter, and calculate the contour symmetry before and after flipping; Extract the gradient direction in the edge gradient information, count the distribution quantity of the gradient direction on the contour of the shield cutter, and calculate the gradient direction mean value of different shield image blocks to determine the gradient direction change; Comprehensively judge the wear type of the shield cutter according to the contour symmetry and gradient direction change. The wear types include uneven wear, chipping, and uniform wear.

Images