A fault defect detection method and system for real-scene scheduling
Through the improved single-precision object detection algorithm, combined with the double-layer attention mechanism and data enhancement technology, the problems of inaccurate identification and long inspection time in power grid equipment detection are solved, and high-precision detection of power grid equipment defects is achieved, which meets the needs of power grid equipment. The demand for real-life scheduling of power grids.
Patent Information
- Application Number
- CN202111222524.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-20
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2041-10-20
AI Technical Summary
The existing power grid equipment defect detection methods have inaccurate identification, long inspection time, and high missed inspection error rate, which cannot meet the precise requirements of real-life scheduling of the power grid, especially when the drone inspection requires lightweight and real-time performance. Improve detection accuracy.
Using an improved single-precision object detection algorithm, image feature information at different scales is extracted through a two-layer attention mechanism, combined with light adjustment, sharpening and noise data enhancement, optimize the prior box parameters, and add non-maximum suppression and bidirectional feature fusion. , improve the accuracy of detection of power grid equipment defects.
It significantly improves the accuracy of defect detection of power grid equipment, reduces the missed detection and error detection rates, meets the accurate requirements of real-life scheduling of power grids, and improves the ability to small targets and noise immunity.
Smart Images

Figure CN113989487B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of intelligent recognition technology, and in particular relates to a fault defect detection method and system for real-scene scheduling. Background Art
[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.
[0003] Real-time grid dispatching refers to the dispatching personnel using information technology to collect relevant information in the power grid, and then obtain the required information, and combine its operation status and recent records to determine whether there is a problem with the equipment. With the continuous development of my country's economy, the mileage of the national power grid's transmission lines is increasing. According to the defects of the power grid equipment, reasonable real-time grid dispatching has become a basic prerequisite for ensuring the safety of high-voltage power supply and maintaining the normal operation of the power system. In order to avoid or reduce the adverse effects caused by defects in power grid equipment, it is particularly important to effectively monitor the power grid equipment and repair it in time.
[0004] At present, the defect detection of power grid equipment generally adopts manual methods, which is not only inaccurate in identification and takes a long time to inspect, but also may lead to large-scale power transmission interruption due to untimely maintenance of power grid equipment defects, causing serious economic losses. In addition, maintenance personnel are an important resource in the maintenance process, but also a scarce resource. If the fault point of power grid equipment cannot be found in a timely and accurate manner, it will be difficult to reasonably and effectively dispatch maintenance personnel, and the situation of "no one can be found at the serious fault point, and the normal point or minor fault point is crowded with people" may occur, resulting in a serious waste of dispatching personnel resources.
[0005] Therefore, timely and accurate detection of power grid equipment defects is one of the important prerequisites for reasonable real-time grid scheduling.
[0006] With the continuous development of drone technology and deep learning technology, drone-based intelligent target detection algorithms have begun to be used in the power field. The current target detection algorithms are divided into two parts: two-stage target detection algorithms and single-stage targets. Two-stage target detection methods, such as the R-CNN series algorithm, first generate several target candidate regions through the Region Proposals Strategy (RPS), extract features and use support vector machines to identify target categories, and finally output the target type in the image and mark its position. This method has high detection accuracy for power grid equipment, but the detection speed is slow and the weight file is large, which does not meet the lightweight and real-time requirements of drone inspections. Single-precision target detection methods, such as the YOLO series and SSD algorithms, use the features extracted by deep convolutional neural networks to directly identify and locate targets. This method meets the lightweight and real-time requirements of drones, but has low detection accuracy and high false detection rates, which does not meet the requirements for accurate scheduling of real-time power grids.
[0007] To sum up, how to improve the accuracy of defect identification of power grid equipment, reduce missed detection and false detection, and meet the requirements of accurate scheduling of the actual power grid while retaining the lightweight and real-time performance of the single-precision target detection algorithm has become an urgent problem to be solved by industry insiders. Summary of the invention
[0008] In order to solve the above problems, the present invention proposes a fault defect detection method and system for real-scene scheduling. By improving the single-precision target detection algorithm, the defect position in the power grid equipment map taken by drone aerial photography is accurately located, and the demand for real-scene precise scheduling of the power grid is further realized.
[0009] According to some embodiments, the first solution of the present disclosure provides a fault defect detection method for real-scene scheduling, which adopts the following technical solutions:
[0010] A method and system for fault defect detection oriented to real-scene scheduling, comprising the following steps:
[0011] Obtain image information of power grid equipment;
[0012] According to the acquired image information and the preset detection and recognition model, the operating status of the power grid equipment is judged and the degree of fault defects of the power grid equipment is identified;
[0013] Among them, the detection and recognition model adopts a double-layer attention mechanism to extract image feature information of different scales, performs detection and recognition of the acquired image information based on a single-precision target detection algorithm, obtains the detection and recognition results of the power grid equipment, judges the operating status of the power grid equipment based on the real-life scheduling of the power grid, and identifies the degree of defects of the power grid equipment.
[0014] As a further technical limitation, after acquiring the image information of the power grid equipment, the image information is enhanced by means of light adjustment, sharpening and noise addition, the image sample data set is expanded, and the prior frame parameters are optimized using a clustering algorithm to improve the matching degree between the prior frame and the target.
[0015] Furthermore, the process of using the clustering algorithm to optimize the prior frame parameters is: determining cluster samples, that is, a sample set in the power grid equipment data set photographed by a drone; clustering the two sample sets of the default candidate frame scale and aspect ratio based on the k-means++ clustering algorithm.
[0016] As a further technical limitation, the detection and recognition model adopts an improved single-precision target detection algorithm, and adds attention mechanisms after the positions of CSP1_3 and CSP2_1 of the backbone network, respectively, to construct a double-layer attention feature extraction mechanism for strengthening the channel features containing important information, so as to reduce the missed detection rate of the defective positions of the power grid equipment.
[0017] As a further technical limitation, non-maximum suppression is used in the single-precision target detection algorithm to use the overlapping area and center point distance between the bounding box and the real box as a criterion for eliminating redundant boxes, thereby accelerating the convergence speed.
[0018] As a further technical limitation, the acquired image information is identified based on a detection and recognition model to obtain damage conditions of the power grid equipment, and the operating status of the power grid equipment is determined based on the damage conditions.
[0019] As a further technical limitation, the damage conditions of the power grid equipment include general damage conditions and emergency damage conditions; wherein the general damage condition is the area where the power grid equipment with a confidence level less than a set threshold is located and the damage degree is relatively small, and the emergency damage condition is the area where the power grid equipment with a confidence level greater than the set threshold is located or the damage degree is seriously severe.
[0020] According to some embodiments, the second solution of the present disclosure provides a fault defect detection system for real-scene scheduling, which adopts the following technical solutions:
[0021] A fault defect detection system for real-scene scheduling, comprising:
[0022] An acquisition module is configured to acquire image information of power grid equipment;
[0023] The detection module is configured to determine the operating status of the power grid equipment and identify the degree of fault defects of the power grid equipment according to the acquired image information and a preset detection and recognition model;
[0024] Among them, the detection and recognition model adopts a double-layer attention mechanism to extract image feature information of different scales, performs detection and recognition of the acquired image information based on a single-precision target detection algorithm, obtains the detection and recognition results of the power grid equipment, judges the operating status of the power grid equipment based on the real-life scheduling of the power grid, and identifies the degree of defects of the power grid equipment.
[0025] According to some embodiments, a third solution of the present disclosure provides a computer-readable storage medium, which adopts the following technical solution:
[0026] A computer-readable storage medium stores a program, which, when executed by a processor, implements the steps of the fault defect detection method for real-scene scheduling as described in the first aspect of the present disclosure.
[0027] According to some embodiments, a fourth solution of the present disclosure provides an electronic device, which adopts the following technical solution:
[0028] An electronic device comprises a memory, a processor and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the fault defect detection method for real-scene scheduling as described in the first aspect of the present disclosure are implemented.
[0029] Compared with the prior art, the present invention has the following beneficial effects:
[0030] Compared with traditional detection and recognition models, the present invention adds a double-layer attention mechanism to the backbone network based on an improved single-precision target detection algorithm, generalizes the feature representations of power grid equipment defects in different scenarios and sizes, highlights the distinguishability between features, improves feature fusion capabilities, improves the detection of small targets and the ability to resist image noise, and significantly improves the detection accuracy of power grid equipment defects. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure.
[0032] Figure 1 is a flow chart of a fault defect detection method for real-scene scheduling in the first embodiment of the present disclosure;
[0033] Figure 2 is an overall structural diagram of the improved precision target detection algorithm in the first embodiment of the present disclosure;
[0034] Figure 3 is a structural diagram of the attention mechanism in the first embodiment of the present disclosure;
[0035] Figure 4This is a structural diagram of the backbone network after adding a double-layer attention mechanism in the first embodiment of the present disclosure;
[0036] Figure 5 is a structural diagram of the BiFPN in the first embodiment of the present disclosure;
[0037] Figure 6 is a structural block diagram of a fault defect detection system for real-scene scheduling in the second embodiment of the present disclosure; DETAILED DESCRIPTION
[0038] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.
[0039] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.
[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0041] In the absence of conflict, the embodiments in the present disclosure and the features in the embodiments may be combined with each other.
[0042] Embodiment 1
[0043] Embodiment 1 of the present disclosure introduces a fault defect detection method for real-scene scheduling.
[0044] like Figure 1 A fault defect detection method for real-scene scheduling is shown, comprising the following steps:
[0045] Obtain image information of power grid equipment;
[0046] According to the acquired image information and the preset detection and recognition model, the operating status of the power grid equipment is judged and the degree of fault defects of the power grid equipment is identified;
[0047] Among them, the detection and recognition model adopts a double-layer attention mechanism to extract image feature information of different scales, performs detection and recognition of the acquired image information based on a single-precision target detection algorithm, obtains the detection and recognition results of the power grid equipment, judges the operating status of the power grid equipment based on the real-life scheduling of the power grid, and identifies the degree of defects of the power grid equipment.
[0048] The overall design of the program is as follows Figure 2 As shown:
[0049] 1. Creation of image dataset and correction of prior frames
[0050] In order to improve the generalization performance of the model and increase the robustness of the recognition of power grid equipment defects in the scene, this embodiment performs data enhancement on the normal pictures and defective pictures of power grid equipment collected by the drone by light adjustment, flipping, sharpening, and adding noise, where the light adjustment degree ranges from -5 to +5;
[0051] In an embodiment, the single-precision target detection algorithm adopts the YOLOv5 target detection algorithm.
[0052] The initial anchor box parameters of YOLOv5 [116, 90, 156, 198, 373, 326], [30, 61, 62, 45, 59, 119], [10, 13, 16, 30, 33, 23] are not suitable for the task detection in this embodiment, so the K-means++ algorithm is used to correct the initial anchor box. The corrected anchor boxes are [116, 90, 156, 198, 373, 326], [30, 61, 62, 45, 59, 119], [10, 13, 16, 30, 33, 23]. The steps of optimizing the prior box parameters using the K-means++ clustering algorithm are as follows:
[0053] First, determine the cluster samples, where the default candidate box scale cluster sample set is s = {s1,…,s k ,…,s t}, a total of t samples, namely, the t power grid equipment target scales marked in the power grid equipment dataset photographed by the drone, is the kth sample in the sample set, w k 、h k is the width and height information of the kth power grid equipment target bounding box marked in the power grid equipment dataset photographed by the drone, w o 、h o are the width and height information of the original image respectively. k ,…,r t}, which is the same as the scale sample set, with a total of t samples, namely the width-to-height ratio information of t power grid equipment targets marked in the data set, where r k =w k h k / w o h o ; Secondly, the k-means++ clustering algorithm is used to cluster the two sample sets of the default candidate box scale and aspect ratio. It includes two steps: initializing k cluster centers and performing standard K-means clustering iterations. The distance formula used in both steps is:
[0054]
[0055] Where μ sj and μ rj They are the j-th cluster center in the clustering process for the scale sample set s and the aspect ratio sample set r, respectively.
[0056] 2. Use LabelImg to annotate the image samples after data enhancement, use rectangular boxes to mark the locations of power grid equipment and defects, and divide the annotated data set into training set and test set in a ratio of 7:3.
[0057] 3. Build an improved YOLOv5l network
[0058] Attention mechanisms are added after the positions of CSP1_3 and CSP2_1 of the backbone network to construct a two-layer attention feature extraction mechanism.
[0059] Since the target of power grid equipment defect detection is small, the traditional backbone network has poor ability to extract the features of power grid equipment defects. This embodiment method proposes a double-layer attention mechanism structure to strengthen the channel features containing important information and improve the ability to extract important feature information, thereby reducing the missed detection rate of the defect location of the power grid equipment. Among them, the working mechanism of the attention mechanism is as follows: Figure 3 As shown, the backbone network flow chart after the double-layer attention mechanism is as follows Figure 4 As shown, the specific algorithm steps are as follows:
[0060] The feature map (h×w×c) input after passing through the CSP1_3 module and CSP2_1 is globally pooled. The pooling window size is (h×w), the number of channels remains unchanged, and the output is a feature description of (1×1×c). The relevant calculation formula is:
[0061]
[0062] Among them, Z c is the output feature map after pooling, H and W represent the width and height of the input feature map, μ c is the input feature map;
[0063] Then it passes through two fully connected layers, with c / r neurons and c neurons respectively, so as to better fit the complex correlation between channels. Finally, the output of the Sigmod activation function is a (1×1×c) weight factor and multiplied with the original input feature map to obtain output feature maps with different weights. The specific formula is:
[0064] S=F ex (z,W)=σ(g(z,W))=σ(W2δ(W1z))
[0065]
[0066] Among them, z is the global information obtained by the squeeze operation, δ represents the Relu function, W1 and W2 are the coefficients of the two fully connected layers, r is a scaling parameter used to reduce the computational complexity and number of parameters of the network. c is the initial weight of each input channel.
[0067] The BiFPN structure is used instead of the original PAN structure to better balance feature information of different scales through weights, thereby improving the feature fusion capability.
[0068] Although the PANet structure can bidirectionally fuse the strong semantic information of high-level feature maps with the positioning information of low-level feature maps, it directly adds the features of different levels during fusion, without considering the problem of unequal contributions to the final output. In order to make the network more efficient and allow the network to learn different input feature weights, this paper uses the BiFPN structure instead of the original PAN structure, which better balances feature information of different scales through weights and improves the feature fusion capability. The specific structure diagram is shown in the figure below. Figure 5 As shown, the relevant algorithm design steps are as follows:
[0069] The concat operation used in the YOLOv5 channel is changed to use pairwise add operation instead, and the horizontal channel is introduced. The weight can better balance the feature information of different scales. The specific process is to add the output feature map of the first layer of attention mechanism in the second concat module of YOLOv5 and the output feature map of the second layer of attention mechanism after CBL in the third concat module. The added and modified features are paired with the features to be fused. The relevant formulas are as follows:
[0070]
[0071] in, represents the output and input feature maps of the i-th layer, ω′ i (i=1,2,3) represents the weight when performing pairwise add operation, which is generated adaptively according to the relationship of feature graph. Resize(.) is the resolution matching of an upsampling or downsampling operation. Conv represents the convolution operation in the feature processing process. is a correction factor greater than zero to avoid the denominator being zero. Represents the intermediate output feature map of the i-th layer, and its calculation process is as follows:
[0072]
[0073] Among them, ω i (i=1,2) is also an adaptive weight factor, and the meanings of other symbols in the formula are consistent with the above formula.
[0074] 3. Use DIoU-NMS instead of NMS to improve the recognition accuracy of subsequent classification networks
[0075] NMS is a commonly used post-processing algorithm for target detection, which is used to remove redundant detection boxes. However, traditional NMS only considers the overlapping area of two boxes, and the description of the overlapping relationship between the bounding box and the real box is not comprehensive enough, which often leads to incorrect suppression. Therefore, this paper selects DIOU-NMS to replace the original NMS. DIOU-NMS uses the overlapping area and center point distance between the bounding box and the real box as the criteria for removing redundant boxes, which speeds up the convergence speed and improves the inspection accuracy. The specific formula is as follows:
[0076]
[0077] Among them, s i is the score of different categories, M is the bounding box with the highest score, IOU represents the intersection-over-union ratio between the bounding box and the pre-true box, ε is the set NMS threshold, and R DIoU (·) represents the DIOU penalty factor;
[0078]
[0079]
[0080] Among them, ρ 2 (b,b gt ) represents the distance between the center point of the bounding box and the true box, c 2 Indicates the diagonal length of the minimum circumscribed moment of the two boxes, B gt and B represent the true box and bounding box, respectively. 5. According to the server performance and to obtain better training results, this paper sets the hyperparameters required for model training, where the iteration batch size is set to 2, the decay coefficient is 0.0005, the total number of iterations is 300, and the initial learning rate is set to 0.001.
[0081] 6. The drone equipped with the improved YOLOv5 model will be used for regional inspections, and the relevant information of the detected defective power grid equipment will be input into the power grid real-time dispatching system for judgment. Then, reasonable and effective personnel dispatch will be carried out according to the degree of equipment failure. The specific steps are as follows:
[0082] 1) The drone uses the improved YOLOv5 model to transmit the real scene of the power grid equipment with defects, and marks the defect location, equipment category, defect recognition confidence, and location information of the drone shooting area on the returned defect image;
[0083] 2) According to the defect confidence and the damage size of the defect location in the returned image, the system automatically classifies the damage situation, and records the area where the power grid equipment with a confidence less than the set threshold and a minor degree of damage is located as a general damage situation. Conversely, the area where the power grid equipment with a confidence greater than the set threshold or a serious degree of damage is located is recorded as an emergency damage situation;
[0084] 3) For general damage, the system will automatically record the location information of the area in the background and issue an early warning. If the current personnel scheduling is tight, the incident will be postponed. If the current personnel are sufficient, the dispatcher will confirm and send relevant equipment category personnel to repair;
[0085] 4) In case of emergency, the system will send the real-life feedback pictures and the judgment information of the relevant drones to the dispatcher for further confirmation. If the dispatcher has doubts about the degree of damage to the equipment, he / she will issue a feedback command to the system. After receiving the command, the drone will take multi-directional photos of the defective power grid equipment and send back the shooting results for a second time in real time, waiting for the dispatcher to confirm again. If the dispatcher confirms that the power grid equipment is seriously damaged, he / she will issue an emergency dispatch command and dispatch relevant electrical equipment maintenance personnel for repair. If the dispatcher determines that the area is normal, the system will delete the emergency damage information of the area, and the drone will continue to inspect other equipment in the area.
[0086] Compared with the traditional detection and recognition model, the method introduced in this embodiment adds a double-layer attention mechanism to the backbone network based on the improved single-precision target detection algorithm, generalizes the feature representations of power grid equipment defects in different scenarios and sizes, highlights the distinguishability between features, improves the feature fusion capability, improves the detection of small targets and the ability to resist image noise, and significantly improves the detection accuracy of power grid equipment defects.
[0087] Embodiment 2
[0088] The second embodiment of the present disclosure introduces a fault defect detection system for real-scene scheduling, which adopts the fault defect detection method for real-scene scheduling introduced in the first embodiment.
[0089] like Figure 6 A fault defect detection system for real-scene scheduling is shown, comprising:
[0090] An acquisition module is configured to acquire image information of power grid equipment;
[0091] The detection module is configured to determine the operating status of the power grid equipment and identify the degree of fault defects of the power grid equipment according to the acquired image information and a preset detection and recognition model;
[0092] Among them, the detection and recognition model adopts a double-layer attention mechanism to extract image feature information of different scales, performs detection and recognition of the acquired image information based on a single-precision target detection algorithm, obtains the detection and recognition results of the power grid equipment, judges the operating status of the power grid equipment based on the real-life scheduling of the power grid, and identifies the degree of defects of the power grid equipment.
[0093] The detailed steps are the same as those of the fault defect detection method for real-scene scheduling provided in Example 1, and will not be repeated here.
[0094] Embodiment 3
[0095] Embodiment 3 of the present disclosure provides a computer-readable storage medium.
[0096] A computer-readable storage medium stores a program, which, when executed by a processor, implements the steps of the fault defect detection method for real-scene scheduling as described in the first embodiment of the present disclosure.
[0097] The detailed steps are the same as those of the fault defect detection method for real-scene scheduling provided in Example 1, and will not be repeated here.
[0098] Embodiment 4
[0099] A fourth embodiment of the present disclosure provides an electronic device.
[0100] An electronic device comprises a memory, a processor and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method and system for fault defect detection for real-scene scheduling as described in the first embodiment of the present disclosure are implemented.
[0101] The detailed steps are the same as the fault defect detection method and system for real-scene scheduling provided in Example 1, and will not be repeated here.
[0102] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.
Claims
1. A fault defect detection method for real-scene scheduling, characterized in that: The following steps are involved: Obtain image information of power grid equipment; According to the acquired image information and the preset detection and recognition model, the operating status of the power grid equipment is judged and the degree of fault defects of the power grid equipment is identified; Among them, the detection and recognition model adopts a double-layer attention mechanism to extract image feature information of different scales, and performs detection and recognition of the acquired image information based on a single-precision target detection algorithm to obtain the detection and recognition results of the power grid equipment. Specifically, attention mechanisms are added after the positions of CSP1_3 and CSP2_1 of the YOLOv5 backbone network, and a double-layer attention feature extraction mechanism is constructed to strengthen the channel features containing important information, so as to reduce the missed detection rate of the defective positions of the power grid equipment; the output feature map of the first-layer attention mechanism is added to the second concat module of YOLOv5 and the output feature map after CBL is used in the third concat module using the second-layer attention mechanism, and the added and modified features are paired with the features originally to be fused; the operating status of the power grid equipment is judged based on the real-scene scheduling of the power grid, and the degree of defects of the power grid equipment is identified.
2. A method for fault defect detection for real-scene scheduling as claimed in claim 1, characterized in that: After acquiring the image information of the power grid equipment, data enhancement of the image information is performed by light adjustment, sharpening and adding noise, the image sample data set is expanded, and the prior frame parameters are optimized using a clustering algorithm to improve the matching degree between the prior frame and the target.
3. A method for fault defect detection for real-scene scheduling as claimed in claim 2, characterized in that: The process of using the clustering algorithm to optimize the prior box parameters is: Determine cluster samples, that is, a sample set in a data set of power grid equipment photographed by a drone; Based on the k-means++ clustering algorithm, the two sample sets of default candidate box scale and aspect ratio are clustered.
4. A method for fault defect detection for real-scene scheduling as claimed in claim 1, characterized in that: In the single-precision target detection algorithm, non-maximum suppression is used to take the overlapping area and center point distance between the bounding box and the real box as the criterion for eliminating redundant boxes, thereby accelerating the convergence speed.
5. A method for fault defect detection for real-scene scheduling as claimed in claim 1, characterized in that: The acquired image information is identified based on the detection and recognition model to obtain the damage condition of the power grid equipment, and the operating status of the power grid equipment is judged based on the damage condition.
6. A method for fault defect detection for real-scene scheduling as claimed in claim 1, characterized in that: The damage conditions of the power grid equipment include general damage conditions and emergency damage conditions; wherein the general damage conditions are areas where power grid equipment with a confidence level less than a set threshold is located, and the emergency damage conditions are areas where power grid equipment with a confidence level greater than a set threshold is located.
7. A fault defect detection system for real-scene scheduling, characterized in that: include: An acquisition module is configured to acquire image information of power grid equipment; The detection module is configured to determine the operating status of the power grid equipment and identify the degree of fault defects of the power grid equipment according to the acquired image information and a preset detection and recognition model; Among them, the detection and recognition model adopts a double-layer attention mechanism to extract image feature information of different scales, and performs detection and recognition of the acquired image information based on a single-precision target detection algorithm to obtain the detection and recognition results of the power grid equipment. Specifically, attention mechanisms are added after the positions of CSP1_3 and CSP2_1 of the YOLOv5 backbone network, and a double-layer attention feature extraction mechanism is constructed to strengthen the channel features containing important information, so as to reduce the missed detection rate of the defective positions of the power grid equipment; the output feature map of the first-layer attention mechanism is added to the second concat module of YOLOv5 and the output feature map after CBL is used in the third concat module using the second-layer attention mechanism, and the added and modified features are paired with the features originally to be fused; the operating status of the power grid equipment is judged based on the real-scene scheduling of the power grid, and the degree of defects of the power grid equipment is identified.
8. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the fault defect detection method for real-scene scheduling as described in any one of claims 1 to 6 are implemented.
9. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the fault defect detection method for real-scene scheduling as described in any one of claims 1-6 are implemented.
