A method, system, medium and device for identifying a semi-loaded state vehicle-mounted scrap steel main body target
By combining deep learning and texture background segmentation methods, and using the DeepLabv3++ model and spectral clustering segmentation algorithm, the problem of misidentifying the inner wall of the carriage as scrap steel under half-load conditions was solved, and high-precision scrap steel target recognition was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ANQING NORMAL UNIV
- Filing Date
- 2024-06-21
- Publication Date
- 2026-07-28
AI Technical Summary
Existing methods for identifying scrap steel targets on vehicles have a problem of misidentifying the inner wall of the vehicle compartment as scrap steel when the vehicle is half-loaded, resulting in low recognition accuracy.
The DeepLabv3++ deep recognition network model based on dilated convolution and residual modules is used for initial recognition. Background texture filtering is performed by combining relative total differential algorithm. Edge purification is performed by Canny edge detection and breadth-first search. Edges are reconstructed by cubic spline interpolation. Finally, background segmentation is performed by weighted distance spectral clustering segmentation algorithm. The number of adaptive classifications is determined to improve recognition accuracy.
It effectively improves the identification accuracy of the main scrap steel target, and can accurately separate the inner wall of the car body from the main scrap steel body, thus improving the accuracy and efficiency of identification.
Smart Images
Figure CN118898835B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of scrap steel identification technology, and in particular to a method, system, medium, and device for identifying main targets of scrap steel on a vehicle in a half-loaded state. Background Technology
[0002] Scrap steel is the only green, environmentally friendly, and recyclable resource that can replace iron ore in steelmaking, and it is widely used in steel smelting. In industrial manufacturing, using 1 ton of scrap steel can save 0.4 tons of coke, reduce the consumption of 1.7 tons of iron ore, and reduce carbon emissions by 1.6 tons. In recent years, scrap steel recycling technology has made significant progress with the rapid development of electronic information technology. Many computer vision processing technologies have been applied to the scrap steel recycling process to ensure efficient recycling. Among them, visual intelligent assessment based on computer vision processing technology has been applied by many steel companies to the scrap steel quality assessment task. The application of this technology not only solves the problem of low assessment efficiency caused by manual analysis of large amounts of complex tasks for a long time, but also avoids personnel injury during the assessment process.
[0003] Typically, intelligent scrap steel assessment involves evaluating and analyzing the main scrap steel target in acquired images. The assessment process usually begins with acquiring images of the unloading site, followed by target segmentation to extract the assessment subject and complete the intelligent quality assessment of the target. Identifying the main scrap steel target is the primary task of intelligent assessment. Numerous studies have addressed this type of onboard scrap steel target identification problem, broadly categorized into two types. One type uses feature extraction algorithms to extract the four vertices of the truck bed and divide it into rectangles, using the resulting rectangular region as the assessment target. However, this type of algorithm may misidentify the inner wall of the truck bed under half-load conditions as the assessment target. The other type of algorithm is based on currently popular deep network technologies, such as YOLOV, Faster RcNN, and Mask RcNN segmentation models, which learn the features of onboard scrap steel targets through deep networks. In practical applications, such methods have limited effectiveness in extracting the main body of half-loaded scrap steel. This is mainly because the scrap steel main body target reference of these network segmentation algorithms is determined by human identification based on the features of the carriage. Due to the site limitations of image acquisition and camera perspective, a large number of exposed carriage inner walls are included in the main target. After carriage edge feature segmentation, the carriage inner walls will also be regarded as a certain type of scrap steel in the main target, which greatly reduces the accuracy of main target recognition. Summary of the Invention
[0004] To address the technical problems existing in the background art, this invention proposes a method, system, medium, and equipment for identifying the main target of scrap steel on a vehicle in a half-loaded state.
[0005] This invention proposes a method for identifying the main target of scrap steel on a vehicle in a half-loaded state, comprising:
[0006] S1. Obtain the image of the scrap steel to be identified on the vehicle;
[0007] S2. Recognize the image to be recognized to obtain the target region image;
[0008] S3. Perform background texture filtering on the target area image;
[0009] S4. Perform edge extraction on the target region image after background texture filtering to obtain the edge-extracted image;
[0010] S5. Perform contour detection on the edge extraction image to obtain multiple complete and closed contour blocks; extract feature values for each contour block to obtain the feature values of each contour block.
[0011] S6. The feature values of each contour block are fused using the feature value fusion method to obtain distinct feature values, and the number of distinct feature values is used as the number of adaptive classifications.
[0012] S7. Based on the adaptive classification number, perform background segmentation on the image to be identified to obtain an accurate image of the main target of scrap steel.
[0013] Preferably, in S2, the image to be identified is identified to obtain the target region image, specifically including:
[0014] The DeepLabv3++ deep recognition network model based on dilated convolution and residual modules is used to recognize the image to be recognized and obtain the target region image.
[0015] Preferably, in S3, background texture filtering processing is performed on the target region image, specifically including:
[0016] The relative total differential algorithm is used to perform background texture filtering on the target region image.
[0017] Preferably, in S4, edge extraction is performed on the target region image after background texture filtering to obtain an edge-extracted image, specifically including:
[0018] S41. Perform edge extraction on the target region image after background texture filtering to obtain an edge gradient map including edge nodes and gradient values.
[0019] S42. Perform edge purification on the edge gradient map;
[0020] S43. Reconstruct the edge lines based on the purified edge gradient map to obtain the edge extraction image.
[0021] Preferably, in S41, image edge extraction is performed on the target region image after background texture filtering to obtain an edge gradient map including edge nodes and gradient values, specifically including:
[0022] The Canny edge detection algorithm is used to extract the image edges of the target region image after background texture filtering, resulting in an edge gradient map including edge nodes and gradient values.
[0023] Preferably, in S43, edge line reconstruction is performed based on the purified edge gradient map to obtain an edge extraction image, specifically including:
[0024] The edge line was reconstructed based on the purified edge gradient map using cubic spline interpolation to obtain the edge extraction image.
[0025] Preferably, in S42, edge purification of the edge gradient map is performed, specifically including:
[0026] A breadth-first search method is used to purify the edge gradient map.
[0027] Preferably, in S42, a breadth-first search method is used to purify the edge gradient map, specifically as follows:
[0028] S421. Denote the edge nodes in the edge gradient graph as nodes v(i,j); where i and j represent the horizontal and vertical coordinate positions of the edge nodes in the graph, the value of non-empty nodes is v(i,j) = 1, and the value of empty nodes is v(i,j) = 0.
[0029] S422. Select any node v(i0, j0), and sequentially find its adjacent nodes v(i, j0). k j k ), and record all node paths D 0k : v(i0, j0) → v(i k j k The position adjacent to (i0, j0) in ) k j k );
[0030] S423. Select non-empty adjacent nodes v(i) in sequence. k j k Using this as the new starting point, re-search for adjacent nodes v(i). r j r ), and update path D 0r : v(i0, j0) → v(i k j k )→v(i r j r );
[0031] S424: Repeat S423, if the node v(i) found in the nth search is... n j n If the corresponding new neighboring node is empty or coincides with the already searched node, then stop the search and calculate path D. 0n : v(i0, j0) → v(i k j k )→v(i r j r )→…→v(i n j n The number of nodes N) 0n ;
[0032] S425: Determine based on the preset threshold γ, if... Then delete all nodes corresponding to that path (i.e., set the value of the node to 0);
[0033] S426: Repeat steps S422-S425 until all edge nodes have been traversed.
[0034] Preferably, in S7, background segmentation is performed on the image to be identified based on the adaptive classification number to obtain an accurate scrap steel image, specifically including:
[0035] S71. Construct a spectral clustering segmentation algorithm based on weighted distance;
[0036] S72. A weighted distance-based spectral clustering segmentation algorithm is used to segment the background of the image to be identified based on the adaptive number of classifications, thereby obtaining an accurate scrap steel image.
[0037] Preferably, in S71, a weighted distance-based spectral clustering segmentation algorithm is constructed, specifically including:
[0038] S711: Calculate the gradient features of the target region based on the edge gradient image, and use the gradient features to enhance the low-scale edges of the target region.
[0039] S712: A weighted distance metric d is constructed by combining the linear gradation of Euclidean distance for large-difference data and the sensitivity of Mahalanobis distance for low-difference data.
[0040] S713: Construct the degree matrix D and the adjacency matrix W;
[0041] S714: Construct the Laplacian matrix L = DW, and find all eigenvectors η of L. i (i = 1, 2, ..., N), construct the feature matrix F NN = (η1, η2, ..., η) N ).
[0042] Preferably, in S72, a weighted distance-based spectral clustering segmentation algorithm is used to segment the background of the image to be identified according to the adaptive number of classifications, thereby obtaining an accurate image of the main scrap steel target. Specifically, this includes:
[0043] Using adaptive classification number, feature matrix F NN The weighted distance metric d and the Kmeans algorithm are used to perform spectral clustering segmentation on the image to be identified, separating the target from the background to obtain an accurate image of the main target of scrap steel.
[0044] This invention also proposes a vehicle-mounted scrap steel main target identification system in a half-loaded state, comprising:
[0045] The acquisition module is used to acquire images of the scrap steel on the vehicle that need to be identified.
[0046] The preliminary recognition module is used to recognize the image to be recognized and obtain the target area image;
[0047] The filtering module is used to perform background texture filtering on the target area image;
[0048] The edge extraction module is used to extract edges from the target region image after background texture filtering to obtain an edge-extracted image.
[0049] The feature value extraction module is used to perform contour detection on the edge extraction image to obtain multiple complete and closed contour blocks; feature value extraction is performed on each contour block to obtain the feature value of each contour block;
[0050] The adaptive classification number determination module is used to fuse the feature values of each contour block using a feature value fusion method to obtain distinct feature values, and the number of distinct feature values is used as the adaptive classification number.
[0051] The segmentation module is used to perform background segmentation on the image to be identified based on the adaptive number of classifications, so as to obtain an accurate image of the main target of scrap steel.
[0052] The present invention also proposes a computer-readable storage medium storing a half-loaded vehicle-mounted scrap steel main target identification program, which, when executed by a processor, implements the half-loaded vehicle-mounted scrap steel main target identification method as described in any of the above.
[0053] The present invention also proposes a device, including a memory, a processor, and a half-loaded vehicle-mounted scrap steel main target identification program stored in the memory and executable on the processor. When the processor executes the half-loaded vehicle-mounted scrap steel main target identification program, it implements the half-loaded vehicle-mounted scrap steel main target identification method as described in any of the above.
[0054] The present invention proposes a method, system, medium, and device for identifying the main target of scrap steel in a half-loaded vehicle. First, the target area image is obtained by identifying the image to be identified. Then, the background texture of the target area image is filtered. Next, the adaptive number of classifications is determined based on edge detection and contour detection. Then, the background of the image to be identified is segmented according to the adaptive number of classifications, which can obtain an accurate image of the main target of scrap steel and effectively improve the identification accuracy of scrap steel targets. Attached Figure Description
[0055] Figure 1 This is a flowchart illustrating a method for identifying the main target of a vehicle-mounted scrap steel in a half-loaded state, as proposed in this invention. Detailed Implementation
[0056] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0057] Reference Figure 1 The present invention proposes a method for identifying the main target of scrap steel on a vehicle in a half-loaded state, comprising:
[0058] S1. Obtain the image of the scrap steel to be identified on the vehicle;
[0059] S2. Recognize the image to be recognized to obtain the target region image;
[0060] S3. Perform background texture filtering on the target area image;
[0061] S4. Perform edge extraction on the target region image after background texture filtering to obtain the edge-extracted image;
[0062] S5. Perform contour detection on the edge extraction image to obtain multiple complete and closed contour blocks; extract feature values for each contour block to obtain the feature values of each contour block.
[0063] S6. The feature values of each contour block are fused using the feature value fusion method to obtain distinct feature values, and the number of distinct feature values is used as the number of adaptive classifications.
[0064] S7. Based on the adaptive classification number, perform background segmentation on the image to be identified to obtain an accurate image of the main target of scrap steel.
[0065] This invention first identifies the image to be identified to obtain the target region image, then performs background texture filtering on the target region image, then determines the adaptive number of classifications based on edge detection and contour detection, and then performs background segmentation on the image to be identified according to the adaptive number of classifications to obtain an accurate scrap steel target image, which effectively improves the recognition accuracy of scrap steel targets.
[0066] In this embodiment, in S2, the image to be identified is identified to obtain the target region image, specifically including:
[0067] The DeepLabv3++ model, a deep recognition network based on dilated convolution and residual modules, is used to recognize the image to obtain the target region image, so as to effectively extract the overall region of the main target area of the vehicle-mounted scrap steel.
[0068] In this embodiment, in S3, background texture filtering processing is performed on the target region image, specifically including:
[0069] The relative total differential algorithm is used to perform background texture filtering on the target area image to enhance the overall feature uniformity of the background of the carriage interior and improve the accuracy of carriage background segmentation.
[0070] In this embodiment, in S4, edge extraction is performed on the target region image after background texture filtering to obtain an edge-extracted image, specifically including:
[0071] S41. Extract the edge of the target region image after background texture filtering to obtain an edge gradient map including edge nodes and gradient values, so as to roughly divide the target region image and provide a basis for determining the number of target region categories.
[0072] S42. Perform edge purification on the edge gradient map to eliminate interference from small divisions within the region, improve the integrity of the region background, and improve the accuracy of the number of target region classifications.
[0073] S43. Reconstruct the edge lines based on the purified edge gradient map to obtain the edge extraction image, so as to fill the edge loss caused by excessive threshold during the edge extraction process, improve the independence of each region of the image, and further improve the accuracy of the number of target regions classified.
[0074] It is important to understand that edge purification in this embodiment refers to removing interfering features similar to edges to eliminate interference from minor divisions within the region, thereby improving the integrity of the regional background and increasing the accuracy of the number of target region classifications.
[0075] To better perform coarse region segmentation of the target region image, in a further embodiment, in S41, image edge extraction is performed on the target region image after background texture filtering to obtain an edge gradient map including edge nodes and gradient values, specifically including:
[0076] The Canny edge detection algorithm is used to extract the image edges of the target region image after background texture filtering, resulting in an edge gradient map including edge nodes and gradient values.
[0077] To further eliminate interference from minor subdivisions within the region, improve the integrity of the regional background, and enhance the accuracy of target region classification, in a further embodiment, in S41, edge purification is performed on the edge gradient map, specifically including:
[0078] A breadth-first search method is used to purify the edge gradient map.
[0079] In S42, a breadth-first search method is used to purify the edge gradient map, specifically as follows:
[0080] S421. Denote the edge nodes in the edge gradient graph as nodes v(i,j); where i and j represent the horizontal and vertical coordinate positions of the edge nodes in the graph, the value of non-empty nodes is v(i,j) = 1, and the value of empty nodes is v(i,j) = 0.
[0081] S422. Select any node v(i0, j0), and sequentially find its adjacent nodes v(i, j0). k j k Record all node paths D0. k : v(i0, j0) → v(i k j k The position adjacent to (i0, j0) in ) k j k );
[0082] S423. Select non-empty adjacent nodes v(i) in sequence. k j k Using this as the new starting point, re-search for adjacent nodes v(i). r j r ), and update path D 0r : v(i0, j0) → v(i k j k )→v(i r j r );
[0083] S424: Repeat S423, if the node v(i) found in the nth search is... n j nIf the corresponding new neighboring node is empty or coincides with the already searched node, then stop the search and calculate path D. 0n : v(i0, j0) → v(i k j k )→v(i r j r )→…→v(i n j n The number of nodes N) 0n ;
[0084] S425: Determine based on the preset threshold γ, if... Then delete all nodes corresponding to that path (i.e., set the value of the node to 0);
[0085] S426: Repeat steps S422-S425 until all edge nodes have been traversed.
[0086] This embodiment determines the path length by using a preset threshold γ during the traversal process. Short paths are deleted to remove interference from small objects and the edges of stains and rust spots, thus achieving purification and obtaining a larger separable area with similar characteristics.
[0087] To compensate for edge loss caused by excessively high thresholds in the Canny operator, improve the independence of different regions in the image, and further enhance the accuracy of target region classification, in a further embodiment, in S43, edge line reconstruction is performed based on the purified edge gradient map to obtain an edge extraction image, specifically including:
[0088] The edge line was reconstructed based on the purified edge gradient map using cubic spline interpolation to obtain the edge extraction image.
[0089] To improve the classification efficiency of the target region, in this embodiment, in S7, background segmentation is performed on the image to be identified based on the adaptive classification number to obtain an accurate scrap steel image, specifically including:
[0090] S71. Construct a spectral clustering segmentation algorithm based on weighted distance;
[0091] S72. A weighted distance-based spectral clustering segmentation algorithm is used to segment the background of the image to be identified based on the adaptive number of classifications, thereby obtaining an accurate scrap steel image.
[0092] In a further embodiment, in S71, a weighted distance-based spectral clustering segmentation algorithm is constructed, specifically including:
[0093] S711: Calculate the gradient features of the target region based on the edge gradient image, and use the gradient features to enhance the low-scale edges of the target region.
[0094] S712: A weighted distance metric d is constructed by combining the linear gradation of Euclidean distance for large-difference data and the sensitivity of Mahalanobis distance for low-difference data.
[0095] S713: Construct the degree matrix D and the adjacency matrix W;
[0096] S714: Construct the Laplacian matrix L = DW, and find all eigenvectors η of L. i (i = 1, 2, ..., N), construct the feature matrix F NN = (η1, η2, ..., η) N ).
[0097] In a further embodiment, in S72, a weighted distance-based spectral clustering segmentation algorithm is used to segment the background of the image to be identified according to the adaptive number of classifications, thereby obtaining an accurate image of the scrap steel main target. Specifically, this includes:
[0098] Using adaptive classification number, feature matrix F NN The weighted distance metric d and the Kmeans algorithm are used to perform spectral clustering segmentation on the image to be identified, separating the target from the background to obtain an accurate image of the main target of scrap steel.
[0099] In a further embodiment, when separating the target from the background, the background area of the inner wall of the vehicle body is automatically selected using the principle of maximizing the area, which can adaptively achieve the removal of the inner wall of the vehicle body and obtain the accurate main target of scrap steel.
[0100] In summary, compared with existing scrap steel assessment target recognition algorithms, this invention adopts a scheme that combines deep networks and texture background segmentation, which effectively makes up for the deficiency of a single deep network model in excluding the background of the carriage, while solving the difficulty of non-learning algorithms in determining the target without experience reference. The combination of these two improvements greatly improves the recognition accuracy of the target subject.
[0101] This invention also proposes a vehicle-mounted scrap steel main target identification system in a half-loaded state, comprising:
[0102] The acquisition module is used to acquire images of the scrap steel on the vehicle that need to be identified.
[0103] The preliminary recognition module is used to recognize the image to be recognized and obtain the target area image;
[0104] The filtering module is used to perform background texture filtering on the target area image;
[0105] The edge extraction module is used to extract edges from the target region image after background texture filtering to obtain an edge-extracted image.
[0106] The feature value extraction module is used to perform contour detection on the edge extraction image to obtain multiple complete and closed contour blocks; feature value extraction is performed on each contour block to obtain the feature value of each contour block;
[0107] The adaptive classification number determination module is used to fuse the feature values of each contour block using a feature value fusion method to obtain distinct feature values, and the number of distinct feature values is used as the adaptive classification number.
[0108] The segmentation module is used to perform background segmentation on the image to be identified based on the adaptive number of classifications, so as to obtain an accurate image of the main target of scrap steel.
[0109] It should be noted that the working methods of each module in this embodiment are the same as those described in the above method embodiments, and will not be repeated here.
[0110] The present invention also proposes a computer-readable storage medium storing a half-loaded vehicle-mounted scrap steel main target identification program, which, when executed by a processor, implements the half-loaded vehicle-mounted scrap steel main target identification method as described in any of the above.
[0111] The present invention also proposes a device, including a memory, a processor, and a half-loaded vehicle-mounted scrap steel main target identification program stored in the memory and executable on the processor. When the processor executes the half-loaded vehicle-mounted scrap steel main target identification program, it implements the half-loaded vehicle-mounted scrap steel main target identification method as described in any of the above.
[0112] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for identifying the main target of scrap steel on a vehicle in a half-loaded state, characterized in that, include: S1. Obtain the image of the scrap steel to be identified on the vehicle; S2. Recognize the image to be recognized to obtain the target region image; S3. Perform background texture filtering on the target area image; S4. Extract edges from the target region image after background texture filtering to obtain an edge gradient map including edge nodes and gradient values; purify the edge gradient map; reconstruct the edge lines based on the purified edge gradient map to obtain the edge-extracted image. S5. Perform contour detection on the edge extraction image to obtain multiple complete and closed contour blocks, and extract feature values for each contour block to obtain the feature values of each contour block. S6. The feature values of each contour block are fused using the feature value fusion method to obtain distinct feature values, and the number of distinct feature values is used as the number of adaptive classifications. S7. Calculate the gradient features of the target region based on the edge gradient image, and use the gradient features to enhance the low-scale edges of the target region; construct a weighted distance metric by combining the linear gradation of Euclidean distance for large-difference data and the sensitivity of Mahalanobis distance for low-difference data. Construct the degree matrix and adjacency matrix ; Constructing the Laplace matrix Seeking All feature vectors Construct the feature matrix ; Using adaptive classification number and feature matrix Weighted distance metric and The algorithm performs spectral clustering segmentation on the image to be identified, separating the target from the background to obtain an accurate image of the main target of the scrap steel.
2. The method for identifying the main target of on-board scrap steel in a half-loaded state according to claim 1, characterized in that, In S2, the image to be recognized is identified to obtain the target region image, specifically including: A deep recognition network based on dilated convolution and residual modules is adopted. The model identifies the image to be recognized and obtains the target region image.
3. The method for identifying the main target of on-board scrap steel in a half-loaded state according to claim 1, characterized in that, In S3, background texture filtering is performed on the target region image, specifically including: The relative total differential algorithm is used to perform background texture filtering on the target region image.
4. The method for identifying the main target of on-board scrap steel in a half-loaded state according to claim 1, characterized in that, In S41, image edge extraction is performed on the target region image after background texture filtering to obtain an edge gradient map including edge nodes and gradient values, specifically including: The Canny edge detection algorithm is used to extract the image edges of the target region image after background texture filtering, and the edge gradient map including edge nodes and gradient values is obtained. In step S43, edge line reconstruction is performed based on the purified edge gradient map to obtain the edge extraction image, specifically including: The edge line was reconstructed based on the purified edge gradient map using cubic spline interpolation to obtain the edge extraction image.
5. The method for identifying the main target of on-board scrap steel in a half-loaded state according to claim 1, characterized in that, In S42, edge purification is performed on the edge gradient map, specifically including: S421. Denote the edge nodes in the edge gradient graph as nodes. ;in, This represents the horizontal and vertical coordinate positions of edge nodes in the graph, and the values of non-empty nodes. The value of an empty node is ; S422, Select any node Search for adjacent nodes in sequence. And record all node paths Zhongyu adjacent positions ; S423. Select non-empty adjacent nodes in sequence. As a new starting point, re-search for adjacent nodes. and update the path ; S424: Repeat S423, if the... The node searched in the second step If the corresponding new neighboring node is empty or overlaps with an already searched node, stop the search and calculate the path. Number of nodes ; S425: Based on a preset threshold Make a judgment, if If so, then delete all nodes corresponding to that path; S426: Repeat steps S422-S425 until all edge nodes have been traversed.
6. A vehicle-mounted scrap steel main target identification system in a half-loaded state, characterized in that, include: The acquisition module is used to acquire images of the scrap steel on the vehicle that need to be identified. The preliminary recognition module is used to recognize the image to be recognized and obtain the target area image; The filtering module is used to perform background texture filtering on the target area image; The edge extraction module is used to extract edges from the target region image after background texture filtering to obtain an edge gradient map including edge nodes and gradient values; to purify the edge gradient map; and to reconstruct the edge lines based on the purified edge gradient map to obtain the edge-extracted image. The feature value extraction module is used to perform contour detection on the edge extraction image to obtain multiple complete and closed contour blocks; feature value extraction is performed on each contour block to obtain the feature value of each contour block; The adaptive classification number determination module is used to fuse the feature values of each contour block using a feature value fusion method to obtain distinct feature values, and the number of distinct feature values is used as the adaptive classification number. The segmentation module calculates the gradient features of the target region based on the edge gradient image and uses these features to enhance the low-scale edges of the target region. A weighted distance metric is constructed by combining the linear gradation of Euclidean distance for large-difference data and the sensitivity of Mahalanobis distance for low-difference data. Construct the degree matrix and adjacency matrix ; Constructing the Laplace matrix Seeking All feature vectors Construct the feature matrix ;Utilizing adaptive classification number and feature matrix Weighted distance metric and The algorithm performs spectral clustering segmentation on the image to be identified, separating the target from the background to obtain an accurate image of the main target of the scrap steel.
7. A computer-readable storage medium, characterized in that, It stores a half-loaded vehicle-mounted scrap steel main target identification program, which, when executed by the processor, implements the half-loaded vehicle-mounted scrap steel main target identification method as described in any one of claims 1-5.
8. A device, characterized in that, The system includes a memory, a processor, and a half-loaded vehicle-mounted scrap steel main target identification program stored in the memory and executable on the processor. When the processor executes the half-loaded vehicle-mounted scrap steel main target identification program, it implements the half-loaded vehicle-mounted scrap steel main target identification method as described in any one of claims 1-5.