Multi-dimensional evaluation method and device for insulator shackle corrosion detection model

Abstract

The invention relates to the technical field of power equipment detection, in particular to a multi-dimensional evaluation method and device for an insulator shackle corrosion detection model, and aims to evaluate the insulator shackle corrosion detection model based on a detection result of the insulator shackle corrosion detection model by integrating a plurality of evaluation indexes. Carrying out classification and standardization processing on values of the evaluation indexes based on characteristics of the evaluation indexes so as to eliminate dimensional differences and then convert the evaluation indexes into uniform interval values; and the weight value of each evaluation index is calculated by using an automatic weight value optimization method, so that the importance degree of each evaluation index can be objectively quantified, the one-sidedness of subjective weighting is avoided, and the scientificity and credibility of an evaluation result are improved. According to the method, multi-dimensional index information is integrated, limitation of a single index is reduced, scientific and comprehensive evaluation of the insulator shackle corrosion detection model is realized, accurate evaluation of the performance of the insulator shackle corrosion detection model can be realized, and a basis is provided for researchers to further optimize the detection model.

Description

Technical Field

[0001] This invention relates to the field of power equipment testing technology, and in particular to a multi-dimensional evaluation method and device for an insulator shackle corrosion detection model. Background Technology

[0002] Insulators are a crucial component of power transmission lines, and their operational status directly impacts the safety and stability of the power system. Shackles, as insulator connectors, are susceptible to corrosion during long-term service due to exposure to wind, sun, rain, snow, and electrochemical reactions. Severe corrosion of shackles can lead to insulator detachment, causing power line tripping or even large-scale power outages. Therefore, timely detection and identification of insulator shackle corrosion is a vital aspect of power operation and maintenance.

[0003] In recent years, with the development of computer vision and deep learning technologies, researchers have proposed a variety of image recognition-based methods for detecting corrosion of insulator shackles. Existing corrosion detection models for insulator shackles typically use convolutional neural networks and object detection networks (such as Faster R-CNN, YOLO, SSD, etc.) to analyze acquired transmission line images to identify the location and extent of corrosion.

[0004] While existing insulator shackle corrosion detection models have achieved certain results in terms of detection accuracy, shortcomings remain in the model selection and optimization process, particularly the lack of unified and comprehensive evaluation standards. For example, existing studies often use a single performance index as the evaluation criterion, making it difficult to comprehensively measure the model's detection capability and practical application value; existing studies tend to overlook computational efficiency and complexity, with some high-precision models having a large number of parameters and excessive computational overhead, thus limiting their deployment on edge devices; and existing studies are not fair in their model evaluation, often testing model performance on different datasets and with different evaluation indicators, lacking a unified standardized processing and comprehensive comparison mechanism.

[0005] To achieve a comprehensive evaluation of the multi-dimensional performance of models, some studies employ the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS). This method determines the order of performance by calculating the distances of each model to the ideal solution and the negative ideal solution, and it is simple to operate and widely used. However, in the actual evaluation of detection models, TOPSIS often fails to fully consider the directionality of different metrics, typically assuming that larger metric values ​​are always better. This leads to metrics such as mAP, FPS and GFLOPs, and the number of parameters being treated in the same direction, resulting in biased results. Furthermore, metrics such as precision, recall, and F1 score are correlated and cannot be completely independent, and the weight allocation is difficult to dynamically adjust according to the task. Ultimately, this results in an inaccurate comprehensive evaluation, failing to truly reflect the overall performance of the detection model.

[0006] Therefore, a method is needed to evaluate insulator shackle corrosion detection models from multiple dimensions, conduct objective, fair, and comprehensive performance comparisons of the detection models, and provide quantifiable comprehensive evaluation indices to guide model optimization and selection, thereby improving the practicality and reliability of insulator shackle corrosion detection. Summary of the Invention

[0007] Therefore, the technical problem to be solved by the present invention is to overcome the difficulty in the existing technology of unifying and comprehensively evaluating the corrosion detection model of insulator shackles, and the difficulty in optimizing the detection model in a targeted manner, thereby reducing the practicality and reliability of corrosion detection of insulator shackles.

[0008] To address the aforementioned technical problems, this invention provides a multi-dimensional evaluation method for an insulator shackle corrosion detection model, comprising: The corrosion sample images of the insulator shackles were input into the insulator shackle corrosion detection model to obtain the detection results; Based on the detection results, calculate the initial values ​​of multiple preset evaluation indicators; The initial values ​​of each evaluation indicator are normalized to obtain the standardized values ​​of each evaluation indicator. The weight values ​​of each evaluation indicator are calculated based on the automatic weight value optimization method, and the standardized values ​​of each evaluation indicator are weighted and summed according to the weight values ​​of each evaluation indicator to obtain the comprehensive effectiveness index. The comprehensive performance index was used to evaluate the corrosion detection model of insulator shackles.

[0009] Preferably, the preset multiple evaluation indicators include positive indicators and negative indicators, wherein the positive indicators include mAP@50-95, recall rate, F1 score and FPS, and the negative indicators include the number of parameters and GFLOPs.

[0010] Preferably, the initial values ​​of each evaluation indicator are normalized to obtain the standardized values ​​of each evaluation indicator. The specific steps are as follows: If the j-th evaluation indicator is a positive indicator, then the initial value of the positive indicator is normalized using the following formula to obtain the standardized value of the positive indicator: ; If the j-th evaluation indicator is a negative indicator, then the initial value of the negative indicator is normalized using the following formula to obtain the standardized value of the negative indicator: ; in, Let be the initial value of the j-th evaluation index in the corrosion detection model of the i-th insulator shackle. Let j be the standardized value of the evaluation index of the i-th insulator shackle corrosion detection model. and Let be the maximum and minimum values ​​of the j-th evaluation index in all models, respectively. This serves as an index for the corrosion detection model of insulator shackles. , The number of evaluation indicators.

[0011] Preferably, the weight values ​​of each evaluation index are calculated based on the automatic weight value optimization method, and the specific steps include: Initialize the weight values ​​of each evaluation indicator; Construct an objective function and calculate the partial derivative of the objective function with respect to the weight values ​​of each evaluation index to obtain the weight gradient of each evaluation index. The weight values ​​of each evaluation indicator are iteratively updated based on the weight gradient of each evaluation indicator until the iteration termination condition is met, thus obtaining the target weight value of each evaluation indicator.

[0012] Preferably, the weight values ​​of each evaluation indicator are iteratively updated based on the weight gradient of each evaluation indicator, as follows: Set the learning rate; In the (t+1)th iteration, based on the learning rate, the weight gradient of each evaluation metric, and the weight values ​​of each evaluation metric in the tth iteration, the weight values ​​of each evaluation metric in the current iteration are calculated using the following formula: ; in, Let j be the weight value of the j-th evaluation index in the (t+1)-th iteration. Let j be the weight value of the j-th evaluation index in the t-th iteration. This is the index of the iteration round; For learning rate, Let be the objective function. Let be the weight gradient of the j-th evaluation index.

[0013] Preferably, in the (t+1)th iteration, after calculating the weight values ​​of each evaluation indicator in the current iteration based on the learning rate, the weight gradient of each evaluation indicator, and the weight values ​​of each evaluation indicator in the tth iteration, the method further includes: truncating and normalizing the weight values ​​of each evaluation indicator in the current iteration.

[0014] Preferably, the learning rate is in the range of 0.005 to 0.02, and more preferably 0.01.

[0015] Preferably, the iteration termination condition is: the weight difference between the current iteration and the previous iteration is less than the convergence threshold; The weight difference between the current iteration and the previous iteration is calculated using the following formula: ; in, The weight difference between the (t+1)th iteration and the tth iteration. The number of evaluation indicators, Let j be the weight value of the j-th evaluation index in the (t+1)-th iteration. Let j be the weight value of the j-th evaluation index in the t-th iteration. This is the index of the iteration round.

[0016] Preferably, the convergence threshold ranges from 10. -6 ~10 -4 The preferred value is 10. -5 .

[0017] This invention also provides a multi-dimensional evaluation device for an insulator shackle corrosion detection model, comprising: The detection module is used to input the corrosion sample images of insulator shackles into the insulator shackle corrosion detection model to obtain the detection results; The initial index calculation module is used to calculate the initial values ​​of multiple preset evaluation indicators based on the detection results. The normalization module is used to normalize the initial values ​​of each evaluation indicator to obtain the standardized values ​​of each evaluation indicator. The fusion module is used to calculate the weight values ​​of each evaluation indicator based on the automatic weight value optimization method, and to sum the standardized values ​​of each evaluation indicator according to the weight values ​​of each evaluation indicator to obtain the comprehensive efficiency index. The evaluation module is used to evaluate the corrosion detection model of insulator shackles using a comprehensive performance index.

[0018] Compared with the prior art, the above-described technical solution of the present invention has the following advantages: This invention discloses a multi-dimensional evaluation method for an insulator shackle corrosion detection model. Based on the detection results of the insulator shackle corrosion detection model, it comprehensively evaluates the model using multiple evaluation indicators. The numerical values ​​of each evaluation indicator are classified and standardized based on their characteristics to eliminate dimensional differences and transform them into uniform interval values. An automatic weight value optimization method is used to calculate the weight values ​​of each evaluation indicator, objectively quantifying the importance of each indicator and avoiding the bias of subjective weighting, thus improving the scientific rigor and credibility of the evaluation results. This invention integrates multi-dimensional indicator information, reducing the limitations of single indicators, making the evaluation of detection results more comprehensive, objective, and aligned with practical needs. The method achieves a scientific and comprehensive evaluation of the insulator shackle corrosion detection model, enabling accurate judgment of its performance and providing a basis for researchers to further optimize the detection model. This invention provides a reliable evaluation basis for detection work, helps improve the detection accuracy of insulator shackles, and has strong engineering application value. Attached Figure Description

[0019] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart of a multi-dimensional evaluation method for an insulator shackle corrosion detection model according to the present invention; Figure 2 It is a performance comparison chart based on multiple indicators under different models; Figure 3 This is a comparison chart of the precision, recall, and mAP of the YOLOv5n, YOLOv6n, and YOLOv8n models. Figure 3 (a) shows the comparison results of the accuracy of each model. Figure 3 (b) shows the comparison results of the recall rates of each model. Figure 3 (c) shows the comparison results of each model on mAP at IoU=0.5. Figure 3 (d) in the figure represents the comparison results of mAP for each model when the IoU value ranges from 0.5 to 0.95; Figure 4 This is a comparison chart of precision, recall, and mAP between YOLOv8n and the YOLOv8n-Change model optimized based on the comprehensive performance index. Figure 4 (a) shows the comparison results of the accuracy of each model. Figure 4 (b) shows the comparison results of the recall rates of each model. Figure 4 (c) shows the comparison results of each model on mAP at IoU=0.5. Figure 4(d) in the figure represents the comparison results of mAP for each model when the IoU value ranges from 0.5 to 0.95. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0021] Reference Figure 1 As shown, this invention provides a multi-dimensional evaluation method for an insulator shackle corrosion detection model, comprising the following steps:

[0022] S1: Input the corrosion sample images of the insulator shackles into the insulator shackle corrosion detection model to obtain the detection results.

[0023] In this embodiment, the YOLO series target detection models (such as YOLOv5, YOLOv8, etc.) are used as the corrosion detection model for insulator shackles.

[0024] S2: Based on the detection results, calculate the initial values ​​of multiple preset evaluation indicators.

[0025] The model evaluation metrics are selected based on the detection results output by the YOLO model. The preset evaluation metrics may include F1 score, mAP@50, mAP@50-95, precision, recall, number of parameters, GFLOPs, and FPS.

[0026] Preferably, in this embodiment, the preset evaluation indicators are divided into positive indicators and negative indicators, wherein the higher the value of the positive indicators, the better, and the lower the value of the negative indicators, the better. Specifically, the positive indicators include mAP@50-95, recall rate, F1 score, and FPS, while the negative indicators include the number of parameters and GFLOPs.

[0027] S3: Normalize the initial values ​​of each evaluation indicator to obtain the standardized values ​​of each evaluation indicator.

[0028] Specifically, if the j-th evaluation indicator is a positive indicator, then the initial value of the positive indicator is normalized using the following formula to obtain the standardized value of the positive indicator: ; If the j-th evaluation indicator is a negative indicator, then the initial value of the negative indicator is normalized using the following formula to obtain the standardized value of the negative indicator: ; in, Let be the initial value of the j-th evaluation index in the corrosion detection model of the i-th insulator shackle. Let j be the standardized value of the evaluation index of the i-th insulator shackle corrosion detection model. and Let be the maximum and minimum values ​​of the j-th evaluation index in all models, respectively. This serves as an index for the corrosion detection model of insulator shackles. , N represents the number of evaluation metrics. In this embodiment, N is set to 6.

[0029] S4: Calculate the weight values ​​of each evaluation index based on the automatic weight value optimization method, and sum the standardized values ​​of each evaluation index according to the weight values ​​of each evaluation index to obtain the comprehensive efficiency index (Insulator Corrosion Detection Comprehensive Efficiency Index, ICDCEI).

[0030] Preferably, the weight values ​​of each evaluation index are calculated based on the automatic weight value optimization method, and the specific steps include:

[0031] S401: Initialize the weight values ​​of each evaluation indicator.

[0032] Construct the initial weight vector for each evaluation index as follows: And the initial weights satisfy ,and .

[0033] In this embodiment, the initial weights of each evaluation index are set as follows: mAP@50-95: 0.25, Recall: 0.25, F1 score: 0.2, FPS: 0.15, Number of parameters: 0.10, GFLOPs: 0.05.

[0034] S402: Construct the objective function and calculate the partial derivative of the objective function with respect to the weight values ​​of each evaluation index to obtain the weight gradient of each evaluation index.

[0035] The objective function is expressed as: ; in, Let be the objective function. For the weight vector of the evaluation indicators, Let j be the weight value of the j-th evaluation index. Let be the standardized value of the j-th evaluation index in the corrosion detection model of the i-th insulator shackle.

[0036] Calculate the partial derivatives of the objective function with respect to the weights of each evaluation index. The weight gradients of each evaluation index are obtained.

[0037] S403: Iteratively update the weight values ​​of each evaluation indicator based on the weight gradient of each indicator until the iteration termination condition is met, and obtain the target weight values ​​of each evaluation indicator. The specific steps include: Set learning rate ; In the (t+1)th iteration, based on the learning rate, the weight gradient of each evaluation metric, and the weight values ​​of each evaluation metric in the tth iteration, the weight values ​​of each evaluation metric in the current iteration are calculated using the following formula: ; in, Let j be the weight value of the j-th evaluation index in the (t+1)-th iteration. Let j be the weight value of the j-th evaluation index in the t-th iteration. This is the index of the iteration round; For learning rate, Let be the objective function. Let be the weight gradient of the j-th evaluation index; When the iteration termination condition is met, the weight values ​​of each evaluation indicator in the current iteration are output as the target weight values ​​of each evaluation indicator. .

[0038] The iteration termination condition is: the weight difference between the current iteration and the previous iteration is less than the convergence threshold, and the specific calculation formula is as follows: ; in, The weight difference between the (t+1)th iteration and the tth iteration. The number of evaluation indicators, Let j be the weight value of the j-th evaluation index in the (t+1)-th iteration. Let j be the weight value of the j-th evaluation index in the t-th iteration. This is the index of the iteration round.

[0039] Preferably, in order to limit the weight values ​​of each evaluation index to the [0,1] interval and avoid exceeding a reasonable range, in the (t+1)th iteration, after calculating the weight values ​​of each evaluation index in the current iteration based on the learning rate, the weight gradient of each evaluation index and the weight values ​​of each evaluation index in the tth iteration, the method further includes: truncating and normalizing the weight values ​​of each evaluation index in the current iteration.

[0040] The weight values ​​of each evaluation index in the current iteration are truncated and constrained as follows: If Then let ;like Then let .

[0041] After truncating the constraints, normalization is performed using the following formula: ; in, Let j be the normalized weight value of the evaluation index in the t-th iteration. The weight value of the j-th evaluation index after truncation constraint in the t-th iteration round.

[0042] Preferably, the learning rate ranges from 0.005 to 0.02, and more preferably is 0.01. The convergence threshold ranges from 10. -6 ~10 -4 The preferred value is 10. -5 This is to ensure the stability of the weights.

[0043] In one embodiment of the present invention, after the 1234th iteration, the weights of each evaluation index are as follows: mAP@50-95: 0.247, recall: 0.246, F1 score: 0.199, FPS: 0.151, number of parameters: 0.102, GFLOPs: 0.055.

[0044] The comprehensive effectiveness index is obtained by weighting and summing the standardized values ​​of each evaluation indicator according to their respective weights. The calculation formula is as follows: ; in, As a comprehensive performance index, Let be the target weight value of the j-th evaluation indicator, and ; Let be the standardized value of the j-th evaluation index in the corrosion detection model of the i-th insulator shackle.

[0045] S5: Evaluation of the corrosion detection model for insulator shackles using the comprehensive performance index.

[0046] The comprehensive performance index ICDCEI ranges from [0,1]. The closer the value is to 1, the better the comprehensive performance of the detection model in terms of detection accuracy and engineering practicality.

[0047] The Integrated Performance Index (ICDCEI) can be used to evaluate the performance of multiple detection models in a unified manner, and to select or optimize the construction of detection models based on the Integrated Performance Index.

[0048] Preferably, the output format of the Integrated Performance Index (ICDCEI) includes charts, reports, or visualization interfaces to facilitate intuitive comparison and selection between different detection models.

[0049] This invention standardizes the performance of different detection models across multiple dimensions, including mAP@50-95, recall, F1 score, FPS, number of parameters, and GFLOPs, and obtains a comprehensive performance index based on weighted fusion calculation. The results are referenced... Figure 2 As shown. From Figure 2 It can be seen that different models may exhibit differences in a single metric. For example, YOLOv5n has an advantage in inference speed (FPS), but ranks low in the overall index; YOLOv8n performs well in detection accuracy, but its ICDCEI is somewhat affected by its high computational complexity; while YOLOv6n strikes a relatively good balance between accuracy and complexity, with an overall index at a moderate level. ICDCEI does not simply reflect overall performance, but rather achieves a unified and objective evaluation of the overall performance of different models by weighting and balancing multiple dimensions such as detection accuracy, computational complexity, and running speed. The method of this invention can provide a quantitative basis for model selection and optimization, avoiding the imbalance of evaluation results caused by relying on a single metric.

[0050] This invention further compares and analyzes the main performance indicators of different detection models during the training phase, and the results are referenced. Figure 3 and Figure 4 As shown. Figure 3 This is a comparison chart of the precision, recall, and mAP of the YOLOv5n, YOLOv6n, and YOLOv8n models. Figure 3 (a) shows the comparison results of the accuracy of each model. Figure 3 (b) shows the comparison results of the recall rates of each model. Figure 3 (c) shows the comparison results of each model on mAP at IoU=0.5. Figure 3 (d) shows the comparison results of mAP for each model when the IoU range is 0.5-0.95. It can be seen that YOLOv8n outperforms YOLOv5n and YOLOv6n in terms of precision, recall and mAP, showing stronger detection accuracy and convergence stability.

[0051] Figure 4 This is a comparison chart of precision, recall, and mAP between YOLOv8n and the YOLOv8n-Change model optimized based on the comprehensive performance index. Figure 4 (a) shows the comparison results of the accuracy of each model. Figure 4 (b) shows the comparison results of the recall rates of each model. Figure 4 (c) shows the comparison results of each model on mAP at IoU=0.5. Figure 4In the figure, (d) represents the comparison results of mAP for each model when the IoU range is 0.5-0.95. It can be seen that the YOLOv8n-Change model optimized by the method of this invention shows improvements in all performance metrics compared to the original YOLOv8n model during training. Combined with... Figure 2 The comprehensive performance index results show that the training performance trends of each model are consistent with the comprehensive evaluation results, further verifying the effectiveness and rationality of ICDCEI as a multi-dimensional comprehensive performance index. This result demonstrates that the evaluation method proposed in this invention not only reflects the overall model performance numerically but also maintains consistency with the model's training behavior, providing a reliable basis for model optimization and performance improvement.

[0052] In summary, the multi-dimensional evaluation method for insulator shackle corrosion detection models described in this invention evaluates the models based on their detection results, integrating multiple evaluation indicators. The methods classify and standardize the values ​​of each indicator based on their characteristics to eliminate dimensional differences and transform them into uniform interval values. Furthermore, an automatic weight optimization method is used to calculate the weight values ​​of each indicator, objectively quantifying their importance and avoiding the bias of subjective weighting, thus improving the scientific rigor and credibility of the evaluation results. This invention integrates multi-dimensional indicator information, reducing the limitations of single indicators and making the evaluation of detection results more comprehensive, objective, and aligned with practical needs. The method achieves a scientific and comprehensive evaluation of insulator shackle corrosion detection models, enabling accurate assessment of their performance and providing a basis for researchers to further optimize the detection models. This invention provides a reliable evaluation basis for detection work, helps improve the detection accuracy of insulator shackles, and has strong engineering application value.

[0053] Based on the above-mentioned multi-dimensional evaluation method for insulator shackle corrosion detection model, the present invention also provides a multi-dimensional evaluation device for insulator shackle corrosion detection model, comprising: The detection module is used to input the corrosion sample images of insulator shackles into the insulator shackle corrosion detection model to obtain the detection results; The initial index calculation module is used to calculate the initial values ​​of multiple preset evaluation indicators based on the detection results. The normalization module is used to normalize the initial values ​​of each evaluation indicator to obtain the standardized values ​​of each evaluation indicator. The fusion module is used to calculate the weight values ​​of each evaluation indicator based on the automatic weight value optimization method, and to sum the standardized values ​​of each evaluation indicator according to the weight values ​​of each evaluation indicator to obtain the comprehensive efficiency index. The evaluation module is used to evaluate the corrosion detection model of insulator shackles using a comprehensive performance index.

[0054] The device proposed in this invention is not only suitable for corrosion detection of insulator shackles, but can also be extended to various corrosion or defect detection scenarios, including rust corrosion. The application scenarios include, but are not limited to, the detection and evaluation of rust corrosion on the surface of metal components, bolt corrosion, pitting corrosion on the surface of equipment, coating peeling, and other surface defects generated during the operation of power equipment.

[0055] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0056] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0057] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0058] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0059] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for multi-dimension evaluation of a model for detecting corrosion of insulator de-energizing, characterized in that, The method comprises the steps of: inputting the insulator unclamping corrosion sample images into the insulator unclamping corrosion detection model respectively to obtain detection results; calculating initial values of a plurality of preset evaluation indexes according to the detection results; normalizing the initial values of the evaluation indexes to obtain standardized values of the evaluation indexes; calculating weight values of the evaluation indexes based on an automatic weight value optimization method, and performing weighted summation on the standardized values of the evaluation indexes according to the weight values of the evaluation indexes to obtain a comprehensive performance index; evaluating the insulator unclamping corrosion detection model by using the comprehensive performance index.

2. The multi-dimensional evaluation method of the insulator disconnector corrosion detection model according to claim 1, characterized in that, The plurality of preset evaluation indexes comprise positive indexes and negative indexes, wherein the positive indexes comprise mAP@50-95, recall rate, F1 score and FPS, and the negative indexes comprise parameter quantity and GFLOPs.

3. The multi-dimensional evaluation method of the insulator disconnector corrosion detection model according to claim 2, characterized in that, The initial values of the evaluation indexes are normalized to obtain the standardized values of the evaluation indexes, and the specific steps are as follows: If the jth evaluation index is a positive index, the initial value of the positive index is normalized by the following formula to obtain the standardized value of the positive index: ； If the jth evaluation index is a negative index, the initial value of the negative index is normalized by the following formula to obtain the standardized value of the negative index: ； wherein, is the initial value of the jth evaluation index of the ith insulator unloading corrosion detection model, is the standardized value of the jth evaluation index of the ith insulator unloading corrosion detection model, and are the maximum value and the minimum value of the jth evaluation index in all models, respectively, is the index of the insulator unloading corrosion detection model, , is the number of evaluation indexes.

4. The multi-dimensional evaluation method of the insulator disconnector corrosion detection model according to claim 1, characterized in that, The weight values of the evaluation indexes are calculated based on the automatic weight value optimization method, and the specific steps comprise: initializing the weight values of the evaluation indexes; constructing an objective function and calculating the partial derivatives of the objective function with respect to the weight values of the evaluation indexes to obtain weight gradients of the evaluation indexes; iteratively updating the weight values of the evaluation indexes based on the weight gradients of the evaluation indexes until the iteration termination condition is reached to obtain target weight values of the evaluation indexes.

5. The multi-dimensional evaluation method of the insulator disconnector corrosion detection model according to claim 1, characterized in that, The weight values of the evaluation indexes are iteratively updated based on the weight gradients of the evaluation indexes, and the method comprises: setting a learning rate; in the (t+1)th iteration round, the weight values of the evaluation indexes in the current iteration round are calculated based on the learning rate, the weight gradients of the evaluation indexes and the weight values of the evaluation indexes in the tth iteration round by the following formula: ； wherein, is a weight value of the jth evaluation index for the t+1th iteration round, is a weight value of the jth evaluation index for the tth iteration round, is an index of the iteration round; is a learning rate, is an objective function, is a weight gradient of the jth evaluation index.

6. The multi-dimensional evaluation method of the insulator disconnector corrosion detection model according to claim 5, characterized in that, In the (t+1)th iteration round, after calculating the weight values of the evaluation indexes in the current iteration round based on the learning rate, the weight gradients of the evaluation indexes and the weight values of the evaluation indexes in the tth iteration round, the method further comprises: performing truncation constraint and normalization on the weight values of the evaluation indexes in the current iteration round.

7. The multi-dimensional evaluation method of the insulator disconnector corrosion detection model according to claim 5, characterized in that, The learning rate is in the range of 0.005 to 0.02, and the preferred value is 0.

01.

8. The multi-dimensional evaluation method of the insulator clamp corrosion detection model according to claim 1, characterized in that, The iteration termination condition is that the weight difference between the current iteration round and the last iteration round is less than a convergence threshold; the weight difference between the current iteration round and the last iteration round is calculated by the following formula: ； wherein, is the weight difference of the t+1th iteration round and the tth iteration round, is the number of evaluation metrics, is the weight value of the jth evaluation metric of the t+1th iteration round, is the weight value of the jth evaluation metric of the tth iteration round, is the index of the iteration round.

9. The multi-dimensional evaluation method of the insulator disconnector corrosion detection model according to claim 8, characterized in that, The convergence threshold has a value in the range 10 -6 10 -4 , preferably 10 -5 .

10. A device for multi-dimensional evaluation of an insulator disconnection corrosion detection model, characterized by, The method comprises the steps of: a detection module configured to input the insulator unclamping corrosion sample images into the insulator unclamping corrosion detection model respectively to obtain detection results; an initial index calculation module configured to calculate initial values of a plurality of preset evaluation indexes according to the detection results; a normalization module configured to normalize the initial values of the evaluation indexes to obtain standardized values of the evaluation indexes; a fusion module configured to calculate weight values of the evaluation indexes based on an automatic weight value optimization method, and perform weighted summation on the standardized values of the evaluation indexes according to the weight values of the evaluation indexes to obtain a comprehensive performance index; An evaluation module is configured to evaluate the insulator unbuckling corrosion detection model by using a comprehensive performance index.

Images