Single terminal line defect detection method and system based on image processing

Abstract

The invention provides a single-terminal line defect detection method and system based on image processing, and the method comprises the steps: extracting a standard sequence of each detection target from a template image of a single-terminal line with a detection frame, and extracting the feature information of each detection frame of the standard sequence and the image feature information of the detection target in each detection frame; according to the standard sequence, mapping each detection frame of the to-be-detected image detected by the deep learning model, marking the position not mapped by the standard sequence as a missing position, and according to the position information of the detection frame adjacent to the missing position and the position information of each detection frame in the standard sequence, determining the position area of the missing position, extracting image feature information of the position area, performing multi-dimensional feature matching on the image feature information and a standard sequence, and determining whether matching succeeds or not; the standard sequence and the feature information of each detection target of the single-terminal line are extracted, and detection and feature matching are performed on the missing position of the to-be-detected image based on the standard sequence, so that missing detection and error detection can be effectively prevented.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of harness detection, and in particular to a single-terminal wire defect detection method and system based on image processing. BACKGROUND

[0002] In the field of industrial visual detection, deep learning-based target detection models (such as the YOLO series) have been widely used in product defect detection, component identification, and other scenarios. However, in actual industrial environments, affected by factors such as light changes, product position shifts, and image quality fluctuations, detection models are difficult to maintain high-precision output at all times and are prone to missed detection and false detection, i.e., some detection targets are not identified or have too low confidence and are filtered out, or non-target regions are incorrectly identified as targets or target categories are incorrectly classified.

[0003] In particular, in detection scenarios with fixed sequence structures (such as rubber feet, exposed rubber, exposed wire cores, wire core feet, and head components arranged in a fixed order), when a target at a certain position in the sequence is not detected, it will lead to errors in subsequent quality judgment and affect the accuracy of product qualification determination. SUMMARY

[0004] The present application provides a single-terminal wire defect detection method and system based on image processing, a computer device, and a storage medium, which extracts a standard sequence of each detection target appearing in a single-terminal wire, detects and features matches the missing position of the image to be tested based on the standard sequence, which can effectively prevent missed detection and false detection.

[0005] In a first aspect, the present application provides a single-terminal wire defect detection method based on image processing, comprising: extracting a standard sequence of each detection target appearing in a single-terminal wire detection frame template image, and extracting feature information of each detection frame in the standard sequence and image feature information of the detection target in each detection frame; mapping each detection frame of the image to be tested after being detected by a deep learning model according to the standard sequence, and marking the position as a missing position if it is not mapped by the standard sequence; determining the position region of the missing position according to the position information of the detection frame adjacent to the missing position and the position information of each detection frame in the standard sequence; extracting image feature information of the position region, and performing multi-dimensional feature matching between the image feature information of the position region and the image feature information of the detection target in the corresponding detection frame of the standard sequence to determine whether the matching is successful.

[0006] In one of the embodiments, before the mapping of the bounding boxes of the to-be-tested image detected by the deep learning model according to the standard sequence of the appearance of each detection target, the position not mapped by the standard sequence is marked as a missing position, further comprising: performing mutual matching tests on the image feature information of each detection target of a plurality of qualified template images, and setting a matching threshold of each detection target of the to-be-tested image and the detection target in each bounding box of the standard sequence according to the matching result and the image complexity of each detection target.

[0007] In one of the embodiments, the mapping of each bounding box of the to-be-tested image detected by the deep learning model according to the standard sequence, and the position not mapped by the standard sequence is marked as a missing position, further comprising: filtering each bounding box of the to-be-tested image detected by the deep learning model according to the confidence, and retaining the bounding box with high confidence; sorting the retained bounding boxes in the horizontal direction to form a detection sequence; mapping each bounding box in the detection sequence to the corresponding bounding box of the standard sequence according to the greedy matching algorithm, and marking the position not mapped by the standard sequence as a missing position.

[0008] In one of the embodiments, the multi-dimensional feature matching includes multi-scale matching, multi-angle matching, and multi-feature matching. In one of the embodiments, the multi-feature matching is specifically multi-feature matching of the image feature information of the position region and the template image feature information under each scale and angle combination, and the matching score of each feature is weighted and summed based on the image complexity of the corresponding detection target to obtain a feature comprehensive score; the template image is the image of the detection target in the bounding box of the missing position corresponding to the standard sequence.

[0009] In one of the embodiments, for the detection target with an image complexity lower than a preset value, further comprising color segmentation auxiliary positioning and feature matching.

[0010] In one of the embodiments, after the extraction of the image feature information of the position region, the multi-dimensional feature matching of the image feature information of the position region and the image feature information of the detection target in the corresponding bounding box of the standard sequence, and the determination of whether the matching is successful, further comprising: performing multi-dimensional feature verification of the position region with the matching success and the template image, wherein the multi-dimensional verification includes color similarity verification, edge density verification, histogram similarity verification, and feature weighted comprehensive verification.

[0011] In one of the embodiments, after the image feature information of the position area is extracted, the image feature information of the position area is subjected to multi-dimensional feature matching with the image feature information of the detection target in the detection frame corresponding to the standard sequence, and it is determined whether the matching is successful or not, and further comprising: generating a supplementary detection frame in the position area where the matching is successful, outputting the class label, matching degree and position coordinates of the supplementary detection frame, and marking the supplementary detection frame.

[0012] In a second aspect, the present application provides a single-terminal wire defect detection system based on image processing, comprising: a standard sequence extraction unit configured to extract a standard sequence of each detection target from a template image of a single-terminal wire detection frame, and extract feature information of each detection frame of the standard sequence and image feature information of the detection target in each detection frame; a missing position detection unit configured to map each detection frame of a to-be-detected image after deep learning model detection according to the standard sequence, and mark a position not mapped by the standard sequence as a missing position; a position area determination unit configured to determine a position area of the missing position according to position information of the detection frame adjacent to the missing position and position information of each detection frame in the standard sequence; a feature matching unit configured to extract image feature information of the position area, and subject the image feature information of the position area to multi-dimensional feature matching with image feature information of the detection target in the detection frame corresponding to the standard sequence, and determine whether the matching is successful or not.

[0013] In a third aspect, the present application further provides a computer device, comprising a processor and a memory, the memory stores a computer program, and the computer program is loaded and executed by the processor to implement the method of any one of the above.

[0014] In a fourth aspect, the present application further provides a computer storage medium, the computer storage medium stores a computer program, and the computer program is executed to implement the method of any one of the above.

[0015] The application discloses a single-terminal wire defect detection method and system based on image processing, a computer device and a storage medium. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. The drawings in the following description are only corresponding to some embodiments of the present application, and other drawings of other embodiments can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0017] Figure 1 For one of the embodiments of the present application, a method flow chart of a single-terminal wire defect detection method based on image processing is provided. Figure 2 For another embodiment of the present application, a method flow chart of a single-terminal wire defect detection method based on image processing is provided. Figure 3 For one of the embodiments of the present application, a system block diagram of a single-terminal wire defect detection system based on image processing is provided. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort fall within the scope of the present application.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0020] Reference Figure 1 In one of the embodiments of the present application, a single-terminal wire defect detection method based on image processing is provided. S101, extracting a standard sequence of each detection target appearing from a template image of a single-terminal line detection frame, and extracting feature information of each detection frame of the standard sequence and image feature information of the detection target in each detection frame.

[0021] The template image of the detection frame is a template image set obtained after detection by a deep learning model or a manually labeled template image set.

[0022] Specifically, the appearance sequence of the detection target in all template images is counted, and the sequence with the highest frequency of appearance is selected as the standard sequence.

[0023] For each detection target in the standard sequence, the image region of the detection target in each detection frame is extracted from the template image conforming to the sequence pattern as the appearance feature template, and multiple sample templates are collected for each detection target to cover the appearance changes of each detection target.

[0024] The size distribution of the detection frame corresponding to each detection target is counted, the average width and average height are calculated, and the spacing distribution between adjacent detection frames is counted, and the average spacing is calculated as the feature information of each detection frame of the standard sequence, which is used for search area positioning.

[0025] The average RGB value, average edge density, image complexity, color histogram and other image feature information of each detection target in the template image are calculated as the template image feature information of each detection target in the standard sequence.

[0026] S102, mapping each detection frame of the to-be-detected image after detection by the deep learning model according to the standard sequence, and marking the positions not mapped by the standard sequence as missing positions.

[0027] Specifically, each detection frame of the to-be-detected image after detection by the deep learning model is filtered according to the confidence, and the detection frame with high confidence is retained; The retained detection frames are sorted in the horizontal direction to form a detection sequence; According to the greedy matching algorithm, each detection frame in the detection sequence is mapped to the detection frame corresponding to the standard sequence, and the positions not mapped by the standard sequence are marked as missing positions.

[0028] The index and category label of the detection frame corresponding to the missing position are recorded, and potential false detection is identified. When the category label of the detection frame does not belong to any category in the standard sequence, or the number of detection frames exceeds the length of the standard sequence, it is determined that there is redundant detection; when the category of the detection frame after being mapped to a position is inconsistent with the standard category of the position, it is determined that there is a category error. For the detection frame suspected of false detection, the standard template of the corresponding position is used for multi-dimensional feature matching verification, and the detection frame is removed or corrected if the matching score is lower than the threshold.

[0029] S103, determining a position region of the missing position according to position information of the detection frame adjacent to the missing position and position information of each detection frame in the standard sequence.

[0030] Specifically, when there are detected targets on both left and right sides of the missing position, the total interval between the right boundary of the left target and the left boundary of the right target is calculated, and the missing position is assigned a corresponding search interval according to the number of missing positions therebetween, and the search interval is appropriately expanded on the basis thereof to increase robustness.

[0031] When there is only a detected target on the left side of the missing position, the right boundary of the left target is taken as a starting point, and the approximate X coordinate range of the target is estimated according to the average interval between the position and the left position in the standard sequence and the average width of the detection frame at the position, and the search region is expanded to the right on the basis of the estimated range.

[0032] When there is only a detected target on the right side of the missing position, the left boundary of the right target is taken as an ending point, and the approximate X coordinate range of the target is estimated to the left according to the average interval between the position and the right position in the standard sequence and the average width of the detection frame at the position, and the search region is expanded to the right on the basis of the estimated range.

[0033] S104, extracting image feature information of the position region, and performing multi-dimensional feature matching between the image feature information of the position region and image feature information of the detected target in the corresponding detection frame of the standard sequence to determine whether the matching is successful.

[0034] After determining the position region of the missing target in the to-be-tested image, the missing target is searched in the position region, and multi-dimensional feature matching is performed between the image feature information of the missing target and the image feature information of the detected target in the corresponding detection frame of the standard sequence, to finally determine the matching region and matching result of the detected target. If the matching is successful, the detection is passed, otherwise the detection is failed.

[0035] The single-terminal line defect detection method based on image processing of the embodiment further detects the to-be-tested image after detection by the deep learning model. First, the standard sequence of each detection target of the single-terminal line is extracted from the template image, and information such as the order, category, size and position of the detection frame in the standard sequence, and image feature information of the detected target in the detection frame are extracted. Based on the standard sequence, each detection frame of the to-be-tested image after detection by the deep learning model is mapped, and positions not mapped by the standard sequence are marked as missing positions. Meanwhile, misdetected and redundant detection frames are screened out, the redundant detection frames are deleted, and the position region of the missing position or the misdetected detection frame is verified by multi-dimensional feature matching with the template image. This method can effectively compensate for the defects of the existing deep learning model, reduce the probability of misdetection and missed detection, and improve the accuracy of single-terminal line defect detection based on image processing.

[0036] Referring to Figure 2 In one embodiment, the single-terminal line defect detection method based on image processing includes: S201, extracting a standard sequence of each detection target appearing from a template image of a single-terminal line detection frame, and extracting feature information of each detection frame of the standard sequence and image feature information of the detection target in each detection frame.

[0037] S202, performing mutual matching test on the image feature information of each detection target of a plurality of qualified template images, and setting a matching threshold of each detection target of the to-be-tested image and the detection target in each detection frame of the standard sequence according to the matching result and the image complexity of each detection target.

[0038] Specifically, the image complexity of the detection target in each detection frame is calculated according to a plurality of template images conforming to the standard sequence mode. The image complexity is evaluated by the edge density and the gradient variance: the template image is subjected to Canny edge detection, and the edge pixel ratio is calculated as the edge density; the standard deviation of the Sobel gradient is calculated as the gradient variance.

[0039] Mutual matching test is performed on a plurality of sample templates corresponding to each detection target, the minimum matching score, the average matching score and the score standard deviation between the sample templates are calculated, and the adaptive formula is used to calculate the matching threshold of each detection target in combination with the image complexity: the basic threshold = the minimum mutual matching score x a; the complexity adjustment = (1.0-complexity) x b; the standard deviation adjustment = standard deviation x g; the adaptive threshold of each detection target = basic threshold + complexity adjustment + standard deviation adjustment, and is limited within a certain range. The adaptive threshold is used as the matching threshold of each detection target of the to-be-tested image and the detection target in each detection frame of the standard sequence.

[0040] S203, mapping each detection frame of the to-be-tested image after the detection by the deep learning model according to the standard sequence, and marking the position as a missing position if it is not mapped by the standard sequence.

[0041] S204, determining the position area of the missing position according to the position information of the detection frame adjacent to the missing position and the position information of each detection frame in the standard sequence.

[0042] S205, extracting the image feature information of the position area, performing multi-dimensional feature matching on the image feature information of the position area and the image feature information of the detection target in the corresponding detection frame of the standard sequence, and determining whether the matching is successful.

[0043] Specifically, the multi-dimensional feature matching includes multi-scale matching, multi-angle matching, and multi-feature matching; wherein the multi-feature matching specifically is that the image feature information of the position region is matched with the template image feature information under each scale and angle combination, a feature comprehensive score is obtained by weighting and summing the matching scores of each feature based on the image complexity of the corresponding detection target; the template image is an image of a detection target in a detection box corresponding to the missing position of the standard sequence.

[0044] In a specific implementation, the multi-scale matching includes: scaling the template of each detection target under a plurality of scale factors such as [0.6, 0.7, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3, 1.4] to generate template copies of different scales.

[0045] The multi-angle matching includes: rotating the template of each scale under a plurality of angles such as [-3°, 0°, 3°] to generate template copies of different angles.

[0046] The multi-feature matching includes: for the template of each scale-angle combination, normalized cross-correlation template matching is respectively performed in the following four feature spaces: 1. Gray feature matching: matching after converting the search region and the corresponding template into gray images; 2. RGB color feature matching: matching directly in the RGB three-channel color space; 3. HSV H channel matching: converting to the HSV color space and matching only in the H (hue) channel; 4. Color histogram matching: calculating the color histogram correlation coefficient of the search region and the corresponding template. The template is an image template of a detection target in each detection box of the standard sequence.

[0047] Feature comprehensive score: dynamically adjusting the weights of each feature according to the image complexity of the corresponding template of each detection target and normalizing. For high-texture templates, the weights of RGB and gray features are higher; for low-texture templates, the weight of the color histogram feature is higher. The comprehensive score formula is: comprehensive score = RGB matching score × W RGB + gray matching score × W Gray + HSV matching score × W HSV + histogram matching score × W Hist .

[0048] After determining the position region where the missing target of the to-be-tested image is located, the detection target is searched in the position region, the search region is matched based on the feature comprehensive score of the combined template of a plurality of scale-angle combinations, the template features are fully covered, the detection accuracy is improved, the calculated feature comprehensive score is compared with the matching threshold of the detection target in each detection box of the above standard sequence, if it is higher than the matching threshold, the matching is successful, indicating that the region detection passes, otherwise the detection does not pass.

[0049] Further, for the detection target with image complexity lower than the preset value, color segmentation assisted positioning and feature matching are further included.

[0050] In a specific implementation, for a low-texture template with a template image complexity lower than 0.35, a color segmentation method is additionally used for assisted positioning. The average color of the template sample in the HSV space is calculated, color threshold segmentation is performed in the search area, morphological denoising is performed, the contour is extracted, and the contour with an area and an aspect ratio meeting the expectation is selected as a candidate area. When the color segmentation positioning score is high or the template matching score is low, the result of the color segmentation positioning is preferentially used to position the search area and perform feature matching.

[0051] S206, the matching successful position area is subjected to multi-dimensional feature verification with the template image, and the multi-dimensional verification includes color similarity verification, edge density verification, histogram similarity verification, and feature weighted comprehensive verification.

[0052] The matching area with a feature comprehensive score exceeding a matching threshold is further subjected to multi-dimensional verification to ensure the accuracy of detection, and specifically: Color similarity verification: the Euclidean distance between the average RGB color value of the matching area and the average color of the corresponding template at the position is calculated, and the smaller the distance, the higher the similarity. When the color distance exceeds 100, it is determined that the verification fails.

[0053] Edge density verification: Canny edge detection is performed on the matching area, and the edge pixel ratio is compared with the average edge density of the corresponding template at the position. When the edge density difference is too large, the score is deducted.

[0054] Histogram similarity verification: the correlation coefficient of the color histogram of the matching area and all sample templates is calculated, and the maximum value is taken. The higher the correlation coefficient, the higher the similarity.

[0055] Comprehensive verification score: color score × 0.3 + edge score × 0.2 + histogram score × 0.5. When the comprehensive verification score is lower than 0.35, it is determined that the verification fails, and the detection fails.

[0056] S206, a supplementary detection frame is generated at the matching successful position area, and the class label, matching degree, and position coordinates of the supplementary detection frame are output, and the supplementary detection frame is marked.

[0057] For the matching result passing all verifications, a supplementary detection frame is generated at the matching successful matching area, the width and height use the average size of the template at the position, the Y coordinate is aligned with the center of the adjacent detected target, the right boundary does not exceed the left boundary of the right detected target to avoid overlapping. Finally, the class label, matching degree, and position coordinates of the supplementary detection frame are output, and the supplementary detection frame is marked. The matching degree is the feature comprehensive score of the missing position.

[0058] The single-terminal line defect detection method based on image processing provided in the embodiments of the present application searches for a missing target in a position region where the missing position is located after detecting the missing position of the to-be-tested image according to a standard sequence, performs feature comprehensive matching on an image of the search region based on a plurality of scale-angle combined templates, fully covers template features, and improves detection accuracy. For a detection target with low image complexity, color segmentation is combined to assist positioning and feature matching, which can improve detection efficiency. The matching region with a feature comprehensive score exceeding a matching threshold is verified in multiple dimensions, so that the detection result is more accurate. Finally, a supplementary detection frame is generated in the position region where matching is successful, complete information of the missing position is recorded and displayed, which is convenient for archiving and viewing.

[0059] Referring to Figure 3 The embodiments of the present application also provide a single-terminal line defect detection system based on image processing. In one embodiment, the system comprises: A standard sequence extraction unit 10 is configured to extract a standard sequence in which each detection target appears from a template image of a single-terminal line detection frame, and extract feature information of each detection frame of the standard sequence and image feature information of the detection target in each detection frame.

[0060] A missing position detection unit 20 is configured to map each detection frame of a to-be-tested image after being detected by a deep learning model according to the standard sequence, and mark a position that is not mapped by the standard sequence as a missing position.

[0061] A position region determination unit 30 is configured to determine a position region of the missing position according to position information of a detection frame adjacent to the missing position and position information of each detection frame in the standard sequence.

[0062] A feature matching unit 40 is configured to extract image feature information of the position region, perform multi-dimensional feature matching on the image feature information of the position region and image feature information of the detection target in the corresponding detection frame of the standard sequence, and determine whether matching is successful.

[0063] In one embodiment, the single-terminal line defect detection system based on image processing further comprises: A matching threshold calculation unit is configured to perform mutual matching test on image feature information of each detection target of a plurality of qualified template images, set a matching threshold of each detection target of the to-be-tested image and a detection target in each detection frame of the standard sequence according to a matching result and image complexity of each detection target, respectively.

[0064] In one embodiment, the missing position detection unit 20 is specifically configured to: Filter each detection frame of the to-be-tested image after being detected by the deep learning model according to a confidence, and retain a detection frame with high confidence. sequencing the reserved detection frames in a horizontal direction to form a detection sequence; According to a greedy matching algorithm, each detection frame in the detection sequence is mapped to a detection frame corresponding to the standard sequence, and a position not mapped by the standard sequence is marked as a missing position.

[0065] In one of the embodiments, the multi-dimensional feature matching includes multi-scale matching, multi-angle matching, and multi-feature matching. In the multi-feature matching, image feature information of the position region is matched with template image feature information under each scale and angle combination, a matching score of each feature is weighted and summed based on image complexity of the corresponding detection target to obtain a feature comprehensive score, and the template image is an image of a detection target in a detection frame corresponding to the missing position of the standard sequence.

[0066] In one of the embodiments, the single-terminal line defect detection system based on image processing further includes: The color segmentation auxiliary unit is configured to perform color segmentation auxiliary positioning and feature matching on a detection target with image complexity lower than a preset value.

[0067] In one of the embodiments, the single-terminal line defect detection system based on image processing further includes: The matching verification unit is configured to perform multi-dimensional feature verification on a position region with successful matching and a template image, and the multi-dimensional verification includes color similarity verification, edge density verification, histogram similarity verification, and feature weighted comprehensive verification.

[0068] In one of the embodiments, the single-terminal line defect detection system based on image processing further includes: The frame supplement generation unit is configured to generate a supplementary detection frame in a position region with successful matching, output a class label, a matching degree, and a position coordinate of the supplementary detection frame, and mark the supplementary detection frame.

[0069] The specific processes of each unit for performing the corresponding steps are described in detail in the above method embodiments, and thus are not described herein again for simplicity.

[0070] Embodiments of the present application further provide a computer device including a processor and a memory, the memory storing a computer program, and the computer program, when loaded and executed by the processor, implements the method steps of any one of the method embodiments.

[0071] The embodiment of the present application further provides a computer storage medium, which stores a computer program. When the computer program is executed, the method steps of any of the method embodiments described above are implemented.

[0072] In the above-described embodiments of the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the described device embodiments are merely schematic. The division of the units is merely a logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between the units can be indirect couplings or communication connections through some interfaces, devices or units, and can be electrical, mechanical or in other forms.

[0073] The units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments.

[0074] In addition, each functional unit in the various embodiments of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware or in the form of software function unit. If the integrated unit is implemented in the form of software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium.

[0075] Based on such an understanding, the technical solutions of the present application, in essence or the parts that make contributions, or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a mobile terminal, a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0076] In summary, although the application has been disclosed with preferred embodiments as above, the scope of protection of the application is not limited thereto, and any person skilled in the art can make equivalent replacements or changes to the technical solutions of the application within the technical scope disclosed by the application, and such should be covered within the scope of protection of the application.

[0077] The technical features of the above-described embodiments can be combined in any manner. For brevity, not all possible combinations of the technical features in the above-described embodiments are described, but it should be understood that any combination of the technical features is within the scope of the present disclosure as long as the combination does not result in a contradiction.

Claims

1. A method for detecting defects in a single-terminal wire based on image processing, characterized in that, include: Extract the standard sequence of each detection target from the template image of the single-terminal wire detection box, and extract the feature information of each detection box in the standard sequence and the image feature information of the detection target in each detection box; The detection boxes of the image to be tested after detection by the deep learning model are mapped according to the standard sequence, and the positions not mapped by the standard sequence are marked as missing positions. The location region of the missing position is determined based on the position information of the detection boxes adjacent to the missing position and the position information of each detection box in the standard sequence; Extract the image feature information of the location region, and perform multi-dimensional feature matching between the image feature information of the location region and the image feature information of the target detected in the corresponding detection box of the standard sequence to determine whether the match is successful.

2. The method according to claim 1, characterized in that, The step of mapping the detection bounding boxes of the image to be tested after detection by the deep learning model according to the standard sequence of the occurrence of each detection target, before marking the positions not mapped by the standard sequence as missing positions, further includes: A mutual matching test is performed on the image feature information of each detection target in several qualified template images. Based on the matching results and the image complexity of each detection target, a matching threshold is set between each detection target in the image to be tested and the detection targets in each detection box of the standard sequence.

3. The method according to claim 1, characterized in that, The step of mapping each detection box of the image to be tested after detection by the deep learning model according to the standard sequence, and marking the positions not mapped by the standard sequence as missing positions, includes: The detection boxes of the image to be tested after detection by the deep learning model are filtered according to their confidence scores, and the detection boxes with high confidence scores are retained. The retained detection frames are sorted horizontally to form a detection sequence; According to the greedy matching algorithm, each detection box in the detection sequence is mapped to the corresponding detection box in the standard sequence, and the positions not mapped by the standard sequence are marked as missing positions.

4. The method according to claim 1, characterized in that, The multi-dimensional feature matching includes multi-scale matching, multi-angle matching, and multi-feature matching; Specifically, the multi-feature matching involves performing multi-feature matching between the image feature information of the location region and the template image feature information under each scale and angle combination, and then weighting and summing the matching scores of each feature based on the image complexity of the corresponding detection target to obtain a comprehensive feature score; the template image is the image of the detection target in the detection box corresponding to the missing position of the standard sequence.

5. The method according to claim 4, characterized in that, For detection targets with image complexity below a preset value, color segmentation-assisted localization and feature matching are also included.

6. The method according to claim 4, characterized in that, After extracting the image feature information of the location region, performing multi-dimensional feature matching between the image feature information of the location region and the image feature information of the target detected in the corresponding detection box of the standard sequence, and determining whether the match is successful, the method further includes: The successfully matched location region is subjected to multi-dimensional feature verification with the template image. The multi-dimensional feature verification includes color similarity verification, edge density verification, histogram similarity verification, and feature weighted comprehensive verification.

7. The method according to claim 1, characterized in that, After extracting the image feature information of the location region, performing multi-dimensional feature matching between the image feature information of the location region and the image feature information of the target detected in the corresponding detection box of the standard sequence, and determining whether the match is successful, the method further includes: A supplementary detection box is generated in the successfully matched location area. The category label, matching degree and location coordinates of the supplementary detection box are output and marked as a supplementary detection box.

8. A single-terminal wire defect detection system based on image processing, characterized in that, include: The standard sequence extraction unit is used to extract the standard sequence of each detection target from the template image of the single terminal wire detection box, and extract the feature information of each detection box and the image feature information of the detection target in each detection box of the standard sequence; The missing location detection unit is used to map each detection box of the image to be tested after detection by the deep learning model according to the standard sequence, and the location that is not mapped by the standard sequence is marked as a missing location. A location region determination unit is used to determine the location region of the missing location based on the location information of the detection boxes adjacent to the missing location and the location information of each detection box in the standard sequence; The feature matching unit is used to extract image feature information of the location region, perform multi-dimensional feature matching between the image feature information of the location region and the image feature information of the target detected in the corresponding detection box of the standard sequence, and determine whether the match is successful.

9. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing a computer program that, when loaded and executed by the processor, implements the method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed, implements the method as described in any one of claims 1 to 7.

Images