Defect analysis method and system based on time domain stability and multi-graph cross recognition

Abstract

The invention discloses a defect analysis method and system based on time domain stability and multi-graph cross recognition, and relates to the technical field of defect analysis. The method comprises the following steps: acquiring a multi-cycle inspection image, extracting defect data, associating the defect data, generating a defect time sequence track fragment set, extracting a defect time sequence track fragment of a defect to be judged from the defect time sequence track fragment set, fusing the defect time sequence track fragment with the defect to be judged, generating a defect high-dimensional feature fusion vector, a time domain stability quantitative index is generated through a time domain stability quantitative algorithm, an optimal cross recognition confidence coefficient is generated through a multi-graph cross recognition algorithm, a defect association mapping space is constructed based on the time domain stability quantitative index and the optimal cross recognition confidence coefficient, a defect association mode is divided, and a matching decision is made for a defect to be judged. And performing factor disassembly on matching failure defects, and matching targeted optimization strategy recommendation. According to the invention, an efficient solution can be provided for intelligent defect management of industrial equipment and other scenes, and the accuracy, efficiency and intelligent level of industrial inspection are improved.

Description

Technical Field

[0001] This invention relates to the field of defect analysis technology, and in particular to a defect analysis method and system based on temporal stability and multi-graph cross-identification. Background Technology

[0002] In the field of industrial equipment inspection, such as power transmission lines, rail transit, and petrochemical pipelines, accurate defect detection and efficient management are crucial to ensuring the safe and stable operation of equipment. Traditional defect analysis methods often have several limitations: First, they do not make sufficient use of historical inspection data, focusing mainly on the static feature extraction of defects in single-cycle images, neglecting the temporal evolution of the same defect in different inspection cycles. This results in a lack of effective temporal dimension basis for judging the correlation between newly discovered defects and historical defects, making it difficult to achieve full lifecycle tracking and management of defects. Second, the misjudgment rate of duplicate defects is high. Because similarity comparison is based on only a single dimension (such as location or appearance), it is difficult to distinguish between the "evolutionary state of the same defect" and "newly appearing defects," resulting in a large amount of duplicate annotation, increasing the cost of data storage and subsequent analysis. Third, the cross-validation capability of multi-source heterogeneous inspection data is weak. Industrial inspections often involve acquiring images from multiple devices and from multiple perspectives, but traditional methods fail to fully integrate the features of these multiple images. They are easily affected by environmental factors such as changes in lighting, occlusion, and differences in equipment parameters, leading to distortion of defect features and thus misjudgment. Fourth, when defect matching errors occur, there is a lack of a quantitative analysis mechanism for the reasons for failure, making it impossible to optimize detection strategies in a targeted manner and difficult to form a closed-loop system of "detection-analysis-optimization". In addition, traditional methods are poorly adaptable to the dynamic evolution of defects and are unable to cope with changes in the shape and location of defects over time, thus limiting their effectiveness in complex dynamic scenarios.

[0003] Therefore, there is an urgent need for a defect analysis method that can integrate temporal stability analysis and multi-graph cross-identification to solve the above-mentioned technical pain points and improve the intelligence and accuracy of industrial inspection. Summary of the Invention

[0004] The purpose of this invention is to overcome at least one technical problem existing in the prior art and to provide a defect analysis method and system based on time-domain stability and multi-graph cross-identification.

[0005] On one hand, embodiments of the present invention provide a defect analysis method based on temporal stability and multi-graph cross-identification, including: Step S1, acquiring multi-period inspection images, extracting defect data from each inspection image, and associating defect data of the same defect in different periods through a trajectory segment continuity detection algorithm to generate a defect temporal trajectory segment set; Step S2, for the defect to be judged, extracting historical feature sequences from its corresponding defect temporal trajectory segments, and performing spatiotemporal feature fusion of the historical feature sequences with the current features of the defect to be judged to generate a high-dimensional feature fusion vector of the defect; based on the high-dimensional feature fusion vector of the defect and its corresponding historical feature sequences, calculating the temporal stability quantification index of the defect to be judged through a temporal stability quantification algorithm; Step S3, based on the high-dimensional feature fusion vector of the defect, retrieving candidate defect features from a dynamic evolutionary historical defect knowledge base, and using multi-graph cross-identification... The cross-identification algorithm calculates the cross-identification confidence between the defect to be judged and the features of each candidate defect, and selects the optimal cross-identification confidence. Based on the high-dimensional feature fusion vector of the defect and the candidate defect features in the preset dynamic evolutionary historical defect knowledge base, the optimal cross-identification confidence is generated by the multi-graph cross-identification algorithm. Step S4: Using the time-domain stability quantification index as the first dimension and the optimal cross-identification confidence as the second dimension, a two-dimensional defect association mapping space is constructed, and the mapping space is divided into multiple defect association pattern regions according to the preset stability threshold and confidence threshold. According to the region where the defect to be judged is located in the defect association mapping space, the corresponding defect association pattern is matched, and the defect decision operation corresponding to the pattern is executed. Step S5: The factors of the failed matching defects are decomposed, and targeted optimization strategy recommendations are matched from the preset optimization knowledge graph.

[0006] Furthermore, step S1 includes the following sub-steps: Step S11: Obtain inspection images and their image metadata of the same batch of inspection points in multiple historical cycles over the entire time dimension, and extract defect data from each inspection image through a defect target detection model; Step S12: Based on the defect data of each inspection image, using the inspection point where the defect is located as a spatial anchor point, use a trajectory judgment continuity detection algorithm to connect the same defect located in different cycles with a correlation according to the order of its appearance, forming a single defect time-series trajectory segment, and all single defect time-series trajectory segments constitute the defect time-series trajectory segment set.

[0007] Further, step S12 includes: step S120, based on the defect data of each inspection image, using the inspection point where the defect is located as the spatial anchor point, spatially clustering defects detected at the same inspection point in different cycles; step S121, for the target defect in the current cycle, calculating the defect connectivity value between it and all spatially adjacent defects to be associated in the historical cycle using the trajectory association scoring formula; step S122, based on the defect connectivity value, determining the executability of the association series of defects to be associated using the connectivity constraint decision formula, including: when the result calculated by the connectivity constraint decision formula is 1, determining that the target defect in the current cycle and the defect to be associated are the same defect in different cycles, and establishing a cross-cycle association; step S123, connecting the defects with the association relationship in the order of their appearance to form a single defect time-series trajectory segment, and all single defect time-series trajectory segments constitute the defect time-series trajectory segment set.

[0008] Further, step S2 includes the following sub-steps: Step S21: Based on the inspection point and defect type of the defect to be determined, retrieve relevant defect time-series trajectory segments from the defect time-series trajectory segment set, and match the defect time-series trajectory segment that uniquely corresponds to the defect to be determined; Step S22: Perform dimensionless and normalized mapping processing on the defect to be determined and the defect time-series trajectory segment, and output the feature vector of the defect to be determined and the feature vector of the defect time-series trajectory segment; Step S23: Fuse the feature vector of the defect to be determined and the feature vector of the defect time-series trajectory segment through the defect multi-layer feature fusion formula to generate a high-dimensional feature fusion vector of the defect to be determined; Step S24: Based on the high-dimensional feature fusion vector of the defect to be determined and the defect time-series trajectory segment, generate a time-domain stability quantification index through a time-domain stability quantification algorithm.

[0009] Furthermore, step S3 includes the following sub-steps: Step S31, constructing a dynamic evolutionary historical defect knowledge base based on the defect high-dimensional feature fusion vector of each historical inspection image; Step S32, selecting candidate defect features from the dynamic evolutionary historical defect knowledge base based on the inspection point and defect category of the defect to be determined; Step S33, generating the optimal cross-identification confidence score based on the defect high-dimensional feature fusion vector of the defect to be determined and the candidate defect features through a multi-image cross-identification algorithm.

[0010] Furthermore, step S33 includes: step S330, generating the cross-identification confidence of the defect to be determined and each candidate defect feature based on the defect high-dimensional feature fusion vector of the defect to be determined and the candidate defect features through a multi-graph cross-identification algorithm; step S331, removing candidate defect features whose cross-identification confidence is less than the confidence screening threshold, and taking the candidate defect feature with the highest confidence from the remaining candidate defect features as the optimal cross-identification confidence. If the high-confidence candidate defect feature is 0, then it is marked as no matching candidate defect feature, and the optimal cross-identification confidence is 0.

[0011] Furthermore, the defect association pattern region in step S4 includes four decision quadrants: the first quadrant: a high-stability, high-confidence region, corresponding to recurring defect patterns; the second quadrant: a high-stability, low-confidence region, corresponding to new defect patterns; the third quadrant: a low-stability, high-confidence region, corresponding to changing defect patterns; and the fourth quadrant: a low-stability, low-confidence region, corresponding to matching failure patterns. The stability threshold is used to distinguish between high and low stability, and the confidence threshold is used to distinguish between high and low confidence.

[0012] Furthermore, the defect decision-making operation includes: if the defect to be determined falls into the repeated defect pattern region, it is marked as a repeated occurrence of a known defect, and only the last detection time and status of the defect are updated without creating a new defect work order; if the defect to be determined falls into the new defect pattern region, it is identified as a newly discovered defect, a new defect work order is created, and its feature vector is added to the dynamic evolutionary historical defect knowledge base; if the defect to be determined falls into the changing defect pattern region, it is determined that the corresponding known defect has evolved, the feature vector and severity level of the defect in the knowledge base are updated, and a defect change alarm is generated; if the defect to be determined falls into the matching failure pattern region, a failure analysis process is triggered to decompose the reasons for the matching failure in multiple dimensions.

[0013] Furthermore, step S5 includes the following sub-steps: Step S51: Extract the full data of the matching failure defects and decompose them into factors to generate a failure factor decomposition report; Step S52: Based on the failure factor decomposition report, match a unique targeted optimization strategy recommendation from the preset optimization knowledge graph.

[0014] Secondly, embodiments of the present invention provide a defect analysis system based on temporal stability and multi-image cross-identification. The system is implemented using the aforementioned defect analysis method based on temporal stability and multi-image cross-identification, comprising: a defect temporal trajectory fragment set generation module, suitable for acquiring multi-period inspection images, extracting defect data from each inspection image, and associating defect data of the same defect in different periods through a trajectory fragment continuity detection algorithm to generate a defect temporal trajectory fragment set; a temporal stability quantification index calculation module, suitable for extracting historical feature sequences from the corresponding defect temporal trajectory fragments for a defect to be judged, and performing spatiotemporal feature fusion of the historical feature sequences with the current features of the defect to be judged to generate a high-dimensional defect feature fusion vector; calculating the temporal stability quantification index of the defect to be judged based on the high-dimensional defect feature fusion vector and its corresponding historical feature sequences using a temporal stability quantification algorithm; and an optimal cross-identification confidence generation module, suitable for generating a dynamic cross-identification confidence index based on the high-dimensional defect feature fusion vector. Candidate defect features are retrieved from an evolutionary historical defect knowledge base. A multi-graph cross-recognition algorithm is used to calculate the cross-recognition confidence between the defect to be judged and each candidate defect feature, and the optimal cross-recognition confidence is selected. Based on the defect high-dimensional feature fusion vector and the candidate defect features in the preset dynamic evolutionary historical defect knowledge base, the optimal cross-recognition confidence is generated using a multi-graph cross-recognition algorithm. A defect matching decision module is used to construct a two-dimensional defect association mapping space with the temporal stability quantification index as the first dimension and the optimal cross-recognition confidence as the second dimension. The mapping space is divided into multiple defect association pattern regions according to preset stability and confidence thresholds. Based on the region where the defect to be judged is located in the defect association mapping space, the corresponding defect association pattern is matched, and a defect decision operation corresponding to that pattern is executed. A defect optimization strategy generation module is used to decompose the factors of the defect that failed to match and match targeted optimization strategy recommendations from a preset optimization knowledge graph.

[0015] Thirdly, embodiments of the present invention also provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the above-described defect analysis method based on time-domain stability and multi-graph cross-identification.

[0016] Fourthly, embodiments of the present invention also provide a readable storage medium, which, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to execute the aforementioned defect analysis method based on temporal stability and multi-graph cross-identification.

[0017] The beneficial effects achieved by this invention are as follows: By organically integrating multiple innovative technical links such as time-series trajectory modeling, multi-image cross recognition, temporal stability quantification, intelligent spatial decision-making and closed-loop optimization recommendation, this invention not only solves the core pain points of existing technologies such as "repeated misjudgment", "poor environmental robustness", "lack of time-series insight" and "inability to self-optimize", but also brings comprehensive and significant technological progress and practical value to the field of intelligent inspection of industrial equipment in multiple dimensions such as improving detection accuracy, reducing operation and maintenance costs, enhancing security and promoting knowledge accumulation. Attached Figure Description

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Figure 1 This is a flowchart of a defect analysis method based on temporal stability and multi-graph cross-identification provided in Embodiment 1 of this application.

[0020] Figure 2 This is a flowchart of a defect time-series trajectory fragment set generation method provided in Embodiment 1 of this application.

[0021] Figure 3 This is another flowchart for generating a set of defective time-series trajectory segments provided in Embodiment 1 of this application.

[0022] Figure 4 This is a flowchart of the calculation of a time-domain stability quantification index for a defect to be determined, provided in Embodiment 1 of this application.

[0023] Figure 5 This is a flowchart of an optimal cross-identification confidence generation process provided in Embodiment 1 of this application.

[0024] Figure 6 This is a flowchart of a strategy for generating matching failure defects provided in Embodiment 1 of this application.

[0025] Figure 7 This is a schematic diagram of a defect analysis system based on temporal stability and multi-graph cross-identification provided in Embodiment 2 of this application.

[0026] Figure 8 This is a partial block diagram of the electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Example 1

[0029] like Figure 1 The flowchart shown is a defect analysis method based on temporal stability and multi-graph cross-identification provided in this application embodiment.

[0030] As an example, the method includes: Step S1, acquiring multi-period inspection images, extracting defect data from each inspection image, and associating defect data of the same defect in different periods using a trajectory segment continuity detection algorithm to generate a defect time-series trajectory segment set; Step S2, for the defect to be judged, extracting historical feature sequences from its corresponding defect time-series trajectory segments, and performing spatiotemporal feature fusion of the historical feature sequences with the current features of the defect to be judged to generate a high-dimensional defect feature fusion vector; based on the high-dimensional defect feature fusion vector and its corresponding historical feature sequences, calculating the temporal stability quantification index of the defect to be judged using a temporal stability quantification algorithm; Step S3, based on the high-dimensional defect feature fusion vector, retrieving candidate defect features from a dynamic evolutionary historical defect knowledge base, and calculating the defect to be judged using a multi-image cross-recognition algorithm. The cross-identification confidence score between the defect and each candidate defect feature is calculated, and the optimal cross-identification confidence score is selected. Based on the defect high-dimensional feature fusion vector and the candidate defect features in the preset dynamic evolutionary historical defect knowledge base, the optimal cross-identification confidence score is generated through a multi-graph cross-identification algorithm. Step S4: Using the time-domain stability quantification index as the first dimension and the optimal cross-identification confidence score as the second dimension, a two-dimensional defect association mapping space is constructed, and the mapping space is divided into multiple defect association pattern regions according to the preset stability threshold and confidence threshold. According to the region where the defect to be judged is located in the defect association mapping space, the corresponding defect association pattern is matched, and the defect decision operation corresponding to the pattern is executed. Step S5: The factors of the failed matching defects are decomposed, and targeted optimization strategy recommendations are matched from the preset optimization knowledge graph.

[0031] In some feasible implementations, combined with Figure 2 As shown, step S1 includes the following sub-steps: Step S11: Obtain inspection images and their image metadata of the same batch of inspection points in multiple historical cycles over the entire time dimension, and extract defect data of each inspection image through a defect target detection model; Step S12: Based on the defect data of each inspection image, take the inspection point where the defect is located as the spatial anchor point, and use a trajectory judgment continuity detection algorithm to connect the same defect located in different cycles with a correlation according to the order of its appearance to form a single defect time-series trajectory segment. All single defect time-series trajectory segments constitute the defect time-series trajectory segment set.

[0032] Preferably, step S11 includes: accessing inspection images of the same batch of inspection points across multiple periods within the target inspection area, collected in real time by inspection equipment such as inspection drones and high-definition cameras, and synchronously acquiring metadata for each inspection image. This metadata includes, but is not limited to, inspection point identifiers, acquisition timestamps, equipment parameters, and environmental parameters. The inspection images are then standardized using standardization techniques such as aligning images from different periods to the same coordinate system and removing invalid data. Defect data from each standardized inspection image is extracted using a defect target detection model. This defect data includes, but is not limited to, image, defect location, category, and appearance features. Corresponding metadata is associated with the defect data in each inspection image to construct a defect dataset for each inspection image, including associated information such as "image-inspection point ID-timestamp-defect category-defect location-acquisition parameters-appearance features."

[0033] Preferred, combined Figure 3 As shown, step S12 includes: step S120, based on the defect data of each inspection image, using the inspection point where the defect is located as the spatial anchor point, performing spatial clustering on defects detected at the same inspection point in different cycles; step S121, for the target defect in the current cycle, calculating the defect connectivity value between it and all spatially adjacent defects to be associated in the historical cycle using the trajectory association scoring formula; step S122, based on the defect connectivity value, determining the executability of the association and chaining of the defects to be associated using the connectivity constraint decision formula, including: when the result calculated by the connectivity constraint decision formula is 1, determining that the target defect in the current cycle and the defect to be associated are the same defect in different cycles, and establishing a cross-cycle association; step S123, chaining the defects with association relationship in the order of their appearance to form a single defect time-series trajectory segment, and all single defect time-series trajectory segments constitute the defect time-series trajectory segment set.

[0034] Specifically, the trajectory continuity detection algorithm works by clustering the defect dataset based on the defect dataset from each inspection image, using the inspection point where the defect is located as the spatial anchor point, and then using a trajectory association scoring formula. Calculate the defect connectivity values ​​between the defects to be associated. For the first The first defect and the first The defect can be linked to a value. The first The first defect and the first One defect, and The range of values ​​is and , The number of defects at the same inspection point. This represents the number of score factors that can be used to assess defects. The range of values ​​is , For the first The joint contribution index of each defect-related scoring factor. For the first The first defect and the first The first defect The connectivity score of each defect connectivity factor; the connectivity constraint decision formula based on the defect connectivity scores between defects to be associated. Determine the executability of the associated chain of defects. Indicates the first The first defect and the first Each defect can be associated and chained. Indicates the first The first defect and the first This defect may prevent the execution of associated chaining. For defect connectivity threshold, For the first The time point for defect inspection. For the first The time point for defect inspection. This is the maximum permissible time interval set based on the inspection cycle. For the first Inspection points for each defect, For the first The inspection points for each defect. The defect datasets of defects to be associated are linked and concatenated periodically to generate multiple single-defect time-series trajectory segments. These single-defect time-series trajectory segments are then collected to construct a defect time-series trajectory segment set.

[0035] In some feasible implementations, combined with Figure 4 As shown, step S2 includes the following sub-steps: Step S21: Based on the inspection point and defect type of the defect to be determined, retrieve relevant defect time-series trajectory segments from the defect time-series trajectory segment set, and match the defect time-series trajectory segment that uniquely corresponds to the defect to be determined; Step S22: Perform dimensionless and normalized mapping processing on the defect to be determined and the defect time-series trajectory segment, and output the feature vector of the defect to be determined and the feature vector of the defect time-series trajectory segment; Step S23: Fuse the feature vector of the defect to be determined and the feature vector of the defect time-series trajectory segment through the defect multi-layer feature fusion formula to generate a high-dimensional feature fusion vector of the defect to be determined; Step S24: Based on the high-dimensional feature fusion vector of the defect to be determined and the defect time-series trajectory segment, generate a time-domain stability quantification index through a time-domain stability quantification algorithm.

[0036] Specifically, based on the inspection points and defect types of the defects to be determined, relevant defect time-series trajectory segments are retrieved from the defect time-series trajectory segment set. A unique defect time-series trajectory segment corresponding to the defect to be determined is matched. The defects to be determined and the defect time-series trajectory segments are then subjected to dimensionless and normalized mapping processing. The feature vectors of the defects to be determined and the defect time-series trajectory segments are output, and then the feature vectors of the defects to be determined and the defect time-series trajectory segments are output. This is then processed using a multi-layer feature fusion formula for defects. The feature vector of the defect to be determined is fused with the feature vector of the defect's temporal trajectory segment to generate a high-dimensional feature fusion vector of the defect to be determined. This is the high-dimensional feature fusion vector of the defect to be determined. For layer normalization, The feature aggregation weight matrix, The number of features to be fused. The range of values ​​is , For the first Temporal context fusion function for each feature to be fused. For the first The collaborative amplification fusion coefficient matrix of the features to be fused. The first defect feature to be determined One feature to be fused The first feature of the time-series trajectory segment of the defect One feature to be fused For the first Feature enhancement interaction index of each feature to be fused For the first Temporal enhancement interaction bias vectors of the features to be fused This is the aggregate bias vector.

[0037] Specifically, based on the high-dimensional feature fusion vector of the defect to be determined and the feature vector of the defect's temporal trajectory segment, a temporal stability quantification index is generated through a temporal stability quantification algorithm. Its mathematical expression is as follows: ; in, As a quantitative indicator of time-domain stability, The feature consistency quantification coefficient, This represents the number of cycles in the defect time-series trajectory segment. The range of values ​​is , This is the high-dimensional feature fusion vector of the defect to be determined. The first segment of the defect time trajectory Eigenvectors of each period The threshold for determining feature consistency. These are the time series smoothness quantization coefficients. The number of defect feature dimensions. The first segment of the defect time trajectory The first cycle Feature vectors with 1 feature dimension The first defect to be determined Feature vectors with 1 feature dimension This is the threshold for determining the smoothness of the time series.

[0038] In some feasible implementations, combined with Figure 5 As shown, step S3 includes the following sub-steps: Step S31: Construct a dynamic evolutionary historical defect knowledge base based on the defect high-dimensional feature fusion vector of each historical inspection image; Step S32: Filter candidate defect features from the dynamic evolutionary historical defect knowledge base based on the inspection point and defect category of the defect to be determined; Step S33: Generate the optimal cross-identification confidence score based on the defect high-dimensional feature fusion vector of the defect to be determined and the candidate defect features through a multi-image cross-identification algorithm.

[0039] Preferably, step S33 includes: step S330, generating the cross-identification confidence of the defect to be determined and each candidate defect feature based on the defect high-dimensional feature fusion vector of the defect to be determined and the candidate defect features through a multi-graph cross-identification algorithm; step S331, removing candidate defect features whose cross-identification confidence is less than the confidence screening threshold, and taking the candidate defect feature with the highest confidence from the remaining candidate defect features as the optimal cross-identification confidence. If the high-confidence candidate defect feature is 0, then it is marked as no matching candidate defect feature, and the optimal cross-identification confidence is 0.

[0040] Specifically, step S31 includes obtaining the high-dimensional feature fusion vector of defects in the identified inspection images, establishing a defect feature mapping of "image-inspection point ID-category-timestamp-defect high-dimensional feature fusion vector" for each identified inspection image defect, and constructing a dynamic evolutionary historical defect knowledge base.

[0041] Specifically, step S32 includes searching the inspection points and defect categories of the defects to be judged in a dynamic evolutionary historical defect knowledge base, excluding defect features of irrelevant inspection points and defect categories, finely filtering based on the spatial distance between historical defects and defects to be judged being less than a preset distance, and retaining a preset number of candidate defect features.

[0042] Specifically, step S33 includes: generating a cross-identification confidence score between the defect to be determined and each candidate defect feature based on the defect high-dimensional feature fusion vector of the defect to be determined and the candidate defect features through a multi-graph cross-identification algorithm, the mathematical expression of which is: ; in, For the defect to be determined and the first Cross-identification confidence of candidate defect features The range of values ​​is , The number of candidate defect features. Defect to be determined For the first One candidate defect feature, The number of cross-identification factors, The range of values ​​is , For the first Spatial distance indicator index of cross-identification factors, For the first The time decay coefficient of each cross-identification factor For the defect to be determined and the first The first candidate defect feature The cross-identification value of each cross-identification factor. For the first The correlation factor correction index of each cross-identification factor. Candidate defect features with cross-identification confidence scores below the confidence screening threshold are removed, and high-confidence candidate defect features are retained. The candidate defect feature with the highest confidence score among the high-confidence candidate defect features is selected as the optimal cross-identification confidence score. If the high-confidence candidate defect feature score is 0, then it is marked as no matching candidate defect feature, and the optimal cross-identification confidence score is 0.

[0043] In some feasible implementations, the defect association pattern region in step S4 includes four decision quadrants: the first quadrant: a high-stability, high-confidence region, corresponding to recurring defect patterns; the second quadrant: a high-stability, low-confidence region, corresponding to new defect patterns; the third quadrant: a low-stability, high-confidence region, corresponding to changing defect patterns; and the fourth quadrant: a low-stability, low-confidence region, corresponding to matching failure patterns. The stability threshold is used to distinguish between high and low stability, and the confidence threshold is used to distinguish between high and low confidence.

[0044] Specifically, using the temporal stability quantification index as the X-axis and the optimal cross-identification confidence as the Y-axis, the spatial grid is divided according to the preset mapping space dimension rules to construct the defect association mapping space. The value of "temporal stability quantification index - optimal cross-identification confidence" is used as the coordinate point and falls into the corresponding grid of the defect association mapping space.

[0045] In some feasible implementations, the defect decision-making operation includes: if the defect to be determined falls into the repeated defect pattern region, it is marked as a repeated occurrence of a known defect, and only the last detection time and status of the defect are updated without creating a new defect work order; if the defect to be determined falls into the new defect pattern region, it is identified as a newly discovered defect, a new defect work order is created, and its feature vector is added to the dynamic evolutionary historical defect knowledge base; if the defect to be determined falls into the changing defect pattern region, it is determined that the corresponding known defect has evolved, the feature vector and severity level of the defect in the knowledge base are updated, and a defect change alarm is generated; if the defect to be determined falls into the matching failure pattern region, a failure analysis process is triggered to decompose the reasons for the matching failure in multiple dimensions.

[0046] Specifically, based on the inspection points, categories, and other defect attributes of each defect, a time-domain stability threshold and a cross-identification confidence threshold are set for each defect, and the defect association mapping space is divided into high-stability high-confidence association mode, high-stability low-confidence association mode, low-stability high-confidence association mode, and low-stability low-confidence association mode. Based on the numerical matching of the "temporal stability quantification index - optimal cross-identification confidence level" of the defect to be judged, the following steps are taken: When the correlation pattern of the defect to be judged is a high-stability, high-confidence correlation pattern, it is judged as a duplicate defect, and redundant records of the current defect to be judged are deleted; when the correlation pattern of the defect to be judged is a high-stability, low-confidence correlation pattern, it is judged as a new defect, and data such as the high-dimensional feature fusion vector of the current defect and the time-series trajectory fragment of the defect are added to the time-series trajectory fragment set of the defect and the dynamic evolutionary historical defect knowledge base as needed; when the correlation pattern of the defect to be judged is a low-stability, high-confidence correlation pattern, it is judged as a changing defect, and the current periodic defect data is processed to update the corresponding data information in the time-series trajectory fragment set of the defect and the dynamic evolutionary historical defect knowledge base; when the correlation pattern of the defect to be judged is a low-stability, low-confidence correlation pattern, it is judged as a defect with a failed match.

[0047] In some feasible implementations, combined with Figure 6 As shown, step S5 includes the following sub-steps: Step S51: Extract the full data of the matching failure defects and decompose the factors to generate a failure factor decomposition report; Step S52: Based on the failure factor decomposition report, match a unique targeted optimization strategy recommendation from the preset optimization knowledge graph.

[0048] Specifically, step S51 includes: the full data of the failed matching defects includes inspection images, defect time-series trajectory segments, defect high-dimensional feature fusion vectors, decision-making process data, etc., and calculates the defect matching failure contribution based on factors such as data environment location and the full data, and generates a failure factor decomposition report including matching failure factors, matching failure contribution, matching failure main cause, etc.

[0049] Specifically, step S52 includes: searching for the main causes of failure in the preset optimization knowledge graph based on the failure factor decomposition report, matching the corresponding optimization strategy recommendations, and performing multi-objective optimization ranking of the optimization strategy recommendations according to comprehensive effects such as low implementation cost and good effect, and generating a unique and executable optimization strategy recommendation.

[0050] For example, consider the scenario of an insulator pollution defect detected during a power transmission line inspection. However, during the matching decision, it is judged as a "matching failure defect" (i.e., falling into a "low-stability, low-confidence association pattern"). Step S51: Failure Factor Decomposition: The system extracts and analyzes the full data of the matching failure defect, including: inspection images (multi-angle, multi-cycle); defect time-series trajectory fragments (such contaminants may not have been recorded before); high-dimensional feature fusion vector of the defect; temporal stability index and cross-identification confidence during the decision-making process. By calculating the "matching failure contribution," the system generates a failure factor decomposition report as shown in Table 1: Table 1:

[0051] The main reason for the matching failure is that changes in lighting cause feature distortion.

[0052] Step S52: Matching Optimization Strategy Recommendation: Based on the primary cause of "feature distortion due to illumination changes," the system searches within a pre-defined optimization knowledge graph. This knowledge graph stores the mapping relationship between various failure reasons and corresponding optimization strategies, as shown in Table 2: Table 2:

[0053] The system matched the corresponding strategy: "Enable illumination normalization enhancement module".

[0054] Example of optimization strategy execution: When the system inspects the same area again, it automatically calls the "illuminance normalization enhancement module" to perform illumination compensation processing, shadow elimination processing, and contrast adaptive adjustment on the acquired image.

[0055] The above implementation can tap the value of historical inspection data, improve the discriminative power of defect features, reduce the false positive rate of repeated defects, avoid redundant labeling and misjudgment of evolution status, improve the accuracy and robustness of defect identification, reduce the false positive rate of repeated defects, and enhance robustness in complex environments. It can be adapted to various industrial scenarios, break through the scenario limitations of traditional methods, achieve targeted optimization when matching fails, improve the system's self-iteration efficiency, realize full-chain support from problem discovery to problem solving, promote the upgrade of defect management from passive detection to proactive prevention, form a complete closed loop of "detection-analysis-decision-optimization", provide efficient solutions for intelligent defect management in industrial equipment and other scenarios, improve the accuracy, efficiency and intelligence level of industrial inspection, and has important engineering application value.

[0056] Example 2

[0057] Please see Figure 7 This embodiment provides a structural diagram of a defect analysis system based on temporal stability and multi-graph cross-identification.

[0058] As an example, the system is implemented using the defect analysis method based on temporal stability and multi-graph cross-identification described in Example 1, including: The defect time-series trajectory fragment set generation module 700 is suitable for acquiring multi-period inspection images, extracting defect data from each inspection image, and generating a defect time-series trajectory fragment set by associating defect data of the same defect in different periods through a trajectory fragment continuity detection algorithm.

[0059] The temporal stability quantification index calculation module 710 is suitable for extracting historical feature sequences from the corresponding defect time-series trajectory segments for a defect to be judged, and performing spatiotemporal feature fusion of the historical feature sequences with the current features of the defect to be judged to generate a high-dimensional feature fusion vector of the defect; based on the high-dimensional feature fusion vector of the defect and its corresponding historical feature sequences, the temporal stability quantification index of the defect to be judged is calculated by a temporal stability quantification algorithm.

[0060] The optimal cross-identification confidence generation module 720 is suitable for retrieving candidate defect features from a dynamic evolutionary historical defect knowledge base based on the defect high-dimensional feature fusion vector, calculating the cross-identification confidence between the defect to be judged and each candidate defect feature through a multi-graph cross-identification algorithm, and selecting the optimal cross-identification confidence; and generating the optimal cross-identification confidence based on the defect high-dimensional feature fusion vector and the candidate defect features in the preset dynamic evolutionary historical defect knowledge base through a multi-graph cross-identification algorithm.

[0061] The defect matching decision module 730 is suitable for constructing a two-dimensional defect association mapping space with the time-domain stability quantification index as the first dimension and the optimal cross-identification confidence as the second dimension, and dividing the mapping space into multiple defect association mode regions according to the preset stability threshold and confidence threshold; matching the corresponding defect association mode according to the region where the defect to be determined is located in the defect association mapping space, and performing the defect decision operation corresponding to the mode.

[0062] The matching failure defect optimization strategy generation module 740 is suitable for decomposing matching failure defects into factors and matching targeted optimization strategy recommendations from a preset optimization knowledge graph.

[0063] It is not difficult to see that this embodiment is a system implementation corresponding to the first embodiment, and this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the first embodiment.

[0064] It is worth mentioning that all modules involved in this embodiment are logical units. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this invention, this embodiment does not introduce units that are not closely related to solving the technical problem proposed by this invention; however, this does not mean that other units are absent from this embodiment.

[0065] Example 3

[0066] Please see Figure 8 The present invention also provides an electronic device, including: a memory and a processor; the memory stores at least one program instruction; the processor loads and executes the at least one program instruction to implement the defect analysis method based on time-domain stability and multi-graph cross-identification provided in Embodiment 1.

[0067] The memory 802 and processor 801 are connected via a bus, which may include any number of interconnecting buses and bridges, connecting various circuits of one or more processors 801 and memory 802 together. The bus may also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface provides an interface between the bus and the transceiver. The transceiver may be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 801 is transmitted over a wireless medium via an antenna, which further receives data and transmits it to processor 801.

[0068] The processor 801 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. The memory 802 can be used to store data used by the processor 801 during operation.

[0069] Example 4

[0070] This invention also proposes a storage medium storing a defect analysis method based on temporal stability and multi-graph cross-identification. When executed by a processor, the defect analysis program based on temporal stability and multi-graph cross-identification implements the steps of the defect analysis method described above. Since this storage medium employs all the technical solutions of the above embodiments, it possesses at least all the beneficial effects brought about by the technical solutions of the above embodiments, which will not be elaborated upon further here.

[0071] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A defect analysis method based on time domain stability and multi-graph cross identification, characterized in that, The method comprises the following steps: Step S1, acquiring multi-period inspection images, extracting defect data of each inspection image, associating defect data of the same defect in different periods through a track segment continuity detection algorithm, and generating a defect time sequence track segment set; Step S2, for a defect to be judged, extracting a historical feature sequence from the corresponding defect time sequence track segment, and performing spatio-temporal feature fusion on the historical feature sequence and the current feature of the defect to be judged to generate a defect high-dimensional feature fusion vector; based on the defect high-dimensional feature fusion vector and the corresponding historical feature sequence, calculating a time domain stability quantization index of the defect to be judged through a time domain stability quantization algorithm; Step S3, based on the defect high-dimensional feature fusion vector, retrieving candidate defect features from a dynamic evolutionary historical defect knowledge base, calculating the cross-recognition confidence between the defect to be judged and each candidate defect feature through a multi-graph cross-recognition algorithm, and selecting the optimal cross-recognition confidence; Based on the defect high-dimensional feature fusion vector and the candidate defect features in the preset dynamic evolutionary historical defect knowledge base, the multi-graph cross-recognition algorithm generates the optimal cross-recognition confidence; Step S4, constructing a two-dimensional defect correlation mapping space with the time domain stability quantization index as the first dimension and the optimal cross-recognition confidence as the second dimension, and dividing the mapping space into multiple defect correlation mode regions according to a preset stability threshold and a confidence threshold; according to the region where the defect to be judged is located in the defect correlation mapping space, matching the corresponding defect correlation mode, and performing a defect decision operation corresponding to the mode; Step S5, factor disassembly is performed on the matching failed defect, and a targeted optimization strategy recommendation is matched from a preset optimization knowledge graph. 2.The defect analysis method based on time domain stability and multi-diagram cross identification according to claim 1, wherein, The step S1 comprises the following sub-steps: Step S11, acquiring inspection images and image metadata of the same batch of inspection points in multiple historical periods, and extracting defect data of each inspection image through a defect target detection model; Step S12, based on the defect data of each inspection image, taking the inspection point where the defect is located as a spatial anchor point, and connecting the same defects in different periods having a correlation in the time order according to the track judgment continuity detection algorithm to form a single defect time sequence track segment, and all single defect time sequence track segments constitute the defect time sequence track segment set.

3. The defect analysis method based on time-domain stability and multi-pattern cross identification according to claim 2, characterized in that, The step S12 comprises: Step S120, based on the defect data of each inspection image, taking the inspection point where the defect is located as a spatial anchor point, and performing spatial clustering on the defects detected at the same inspection point in different periods; Step S121, for a target defect in the current period, calculating the defect association value between the target defect and all spatially adjacent defects to be associated in the historical period through a track association scoring formula; Step S122, based on the defect association value, determining the association concatenation executability of the defect to be associated through an association constraint decision formula, including: when the result calculated by the association constraint decision formula is 1, it is determined that the target defect in the current period and the defect to be associated are the same defect in different periods, and a cross-period association is established. Step S123: defects with the association relationship are concatenated in time sequence to form a single defect time sequence track segment, and all single defect time sequence track segments constitute the defect time sequence track segment set.

4. The defect analysis method based on time domain stability and multi-pattern cross identification according to claim 1, characterized in that, The step S2 comprises the following sub-steps: Step S21: based on the inspection point and the defect type of the defect to be judged, the relevant defect time sequence track segment is searched from the defect time sequence track segment set, and the defect time sequence track segment corresponding to the defect to be judged is matched; Step S22: the defect to be judged and the defect time sequence track segment are de-dimensioned and normalized to output the defect feature vector and the defect time sequence track segment feature vector; Step S23: the defect feature vector and the defect time sequence track segment feature vector are fused by a defect multi-layer feature fusion formula to generate a high-dimensional feature fusion vector of the defect to be judged; Step S24: based on the high-dimensional feature fusion vector of the defect to be judged and the defect time sequence track segment, a time domain stability quantification index is generated by a time domain stability quantification algorithm.

5. The defect analysis method based on time-domain stability and multi-pattern cross identification according to claim 1, characterized in that, The step S3 comprises the following sub-steps: Step S31: based on the defect high-dimensional feature fusion vector of each historical inspection image, a dynamic evolution type historical defect knowledge base is constructed; Step S32: based on the inspection point and the defect type of the defect to be judged, candidate defect features are screened from the dynamic evolution type historical defect knowledge base; Step S33: based on the defect high-dimensional feature fusion vector of the defect to be judged and the candidate defect features, an optimal cross-recognition confidence is generated by a multi-image cross-recognition algorithm.

6. The defect analysis method based on time-domain stability and multi-pattern cross identification according to claim 5, characterized in that, The step S33 comprises: Step S330: based on the defect high-dimensional feature fusion vector of the defect to be judged and the candidate defect features, cross-recognition confidences of the defect to be judged and each candidate defect feature are generated by a multi-image cross-recognition algorithm; Step S331: candidate defect features with a cross-recognition confidence less than a confidence screening threshold are removed, and the candidate defect feature with the highest confidence is taken from the remaining candidate defect features as the optimal cross-recognition confidence, if the high-confidence candidate defect feature is 0, it is marked as no matching candidate defect feature, and the optimal cross-recognition confidence is 0.

7. The defect analysis method based on time-domain stability and multi-pattern cross identification according to claim 1, characterized in that, The defect association mode region in the step S4 comprises four decision quadrants: First quadrant: high stability high confidence region, corresponding to a repeated defect mode; Second quadrant: high stability low confidence region, corresponding to a new defect mode; Third quadrant: low stability high confidence region, corresponding to a change defect mode; Fourth quadrant: low stability low confidence region, corresponding to a matching failure mode; Wherein, the stability threshold is used to distinguish high stability and low stability, and the confidence threshold is used to distinguish high confidence and low confidence.

8. The defect analysis method based on time-domain stability and multi-pattern cross identification according to claim 7, characterized in that, The defect decision operation comprises: If the defect to be judged falls into the repeated defect mode region, it is marked as a repeated appearance of a known defect, only the last detection time and state of the defect are updated, and no new defect work order is created; If the defect to be judged falls into the new defect mode region, it is identified as a newly discovered defect, a new defect work order is created, and the feature vector is added to the dynamic evolution type historical defect knowledge base; If the to-be-judged defect falls into the change defect mode region, it is determined that the corresponding known defect has evolved, the feature vector and severity level of the defect in the knowledge base are updated, and a defect change warning is generated; If the to-be-judged defect falls into the matching failure mode region, a failure analysis process is triggered to analyze the matching failure reason in multiple dimensions.

9. The defect analysis method based on time-domain stability and multi-pattern cross identification according to claim 1, characterized in that, The step S5 includes the following sub-steps: Step S51, full-amount data of the matching failure defect is extracted for factor disassembly, and a failure factor disassembly report is generated; Step S52, based on the failure factor disassembly report, a unique and targeted optimization strategy recommendation is matched from a preset optimization knowledge graph.

10. A defect analysis system based on time-domain stability and multi-graph cross-identification, the system is implemented by the defect analysis method based on time-domain stability and multi-graph cross-identification according to any one of claims 1-9, characterized in that, It comprises: A defect time sequence trajectory segment set generation module is suitable for acquiring multi-period inspection images, extracting defect data of each inspection image, correlating defect data of the same defect in different periods through a trajectory segment continuity detection algorithm, and generating a defect time sequence trajectory segment set; A time domain stability quantitative index calculation module is suitable for extracting a historical feature sequence from the corresponding defect time sequence trajectory segment of the to-be-judged defect, and performing spatio-temporal feature fusion on the historical feature sequence and the current feature of the to-be-judged defect to generate a defect high-dimensional feature fusion vector; based on the defect high-dimensional feature fusion vector and its corresponding historical feature sequence, a time domain stability quantitative algorithm is used to calculate the time domain stability quantitative index of the to-be-judged defect; An optimal cross-recognition confidence generation module is suitable for retrieving candidate defect features from a dynamic evolutionary historical defect knowledge base based on the defect high-dimensional feature fusion vector, calculating the cross-recognition confidence between the to-be-judged defect and each candidate defect feature through a multi-graph cross-recognition algorithm, and selecting the optimal cross-recognition confidence; Based on the defect high-dimensional feature fusion vector and the candidate defect features in the preset dynamic evolutionary historical defect knowledge base, an optimal cross-recognition confidence is generated through a multi-graph cross-recognition algorithm; A to-be-judged defect matching decision module is suitable for constructing a two-dimensional defect correlation mapping space with the time domain stability quantitative index as the first dimension and the optimal cross-recognition confidence as the second dimension, and dividing the mapping space into multiple defect correlation mode regions according to preset stability threshold and confidence threshold; according to the region where the to-be-judged defect is located in the defect correlation mapping space, the corresponding defect correlation mode is matched, and the defect decision operation corresponding to the mode is performed; A matching failure defect optimization strategy generation module is suitable for factor disassembly of the matching failure defect and matching of a targeted optimization strategy recommendation from a preset optimization knowledge graph.

Images