Fault identification method for small components in power transmission lines

Abstract

The present application discloses a method for identifying small parts faults in power transmission lines, including: obtaining a field image containing a target small part category on the power transmission line, performing multi-scale processing on the field image to obtain a multi-scale image feature map and a multi-scale sub-region feature map; performing enhancement processing on the multi-scale sub-region feature map, performing feature fusion on the processed multi-scale sub-region feature map and the multi-scale image feature map to obtain a multi-scale target region enhancement feature map, and then processing it to obtain a feature-optimized multi-scale fusion feature map; using the created image data set as a training set to train the constructed initial model, and using the trained model as a small part fault feature analysis model; analyzing the feature-optimized multi-scale fusion feature map based on the small part fault feature analysis model to obtain small part fault identification information. The method can realize accurate identification of small part faults in power transmission lines and meet the inspection needs under different terrains and landforms.

Description

Technical Field

[0001] The present disclosure relates to the technical field of power system monitoring, and in particular to a method for identifying faults of small components of a power transmission line. Background Art

[0002] In recent years, with the increasing dependence of modern society on electricity, uninterrupted power supply inspection of power lines has become an indispensable element of power supply guarantee. Under the vision of maintaining quality and increasing efficiency in the power industry, the original manual inspection method of transmission lines will gradually be replaced by new technologies such as drone inspection and robot inspection; however, there are still the following problems in image detection in transmission line drone inspection: the transmission line spans a wide range of terrain, so the aerial images basically include various terrains and landforms in my country, with large background differences and are relatively messy. Various targets, including the background, are densely distributed and easily blocked from each other, making it difficult to identify and locate the required targets. In addition, the height of drone shooting is variable, making the target objects sparsely distributed and with large scale differences. The same detection target will show different size scales and shape characteristics at different shooting distances, and the demand for detection of multi-scale similar targets in the image has increased. Summary of the invention

[0003] In view of this, an embodiment of the present disclosure provides a method for identifying faults of small components in power transmission lines, which can solve the problem in the prior art that it is impossible to accurately identify faults of small components in power transmission lines.

[0004] In a first aspect, an embodiment of the present disclosure provides a method for identifying a small component fault in a power transmission line, which specifically includes the following schemes:

[0005] Acquire a field image containing a target widget class on a transmission line;

[0006] Performing multi-scale processing on the scene image to obtain a multi-scale image feature map and a multi-scale sub-region feature map;

[0007] Performing enhancement processing on the multi-scale sub-region feature map to obtain a sub-region enhanced feature map;

[0008] Performing feature fusion on the sub-region enhancement feature map and the multi-scale image feature map to obtain a multi-scale target region enhancement feature map;

[0009] Processing the multi-scale target area enhanced feature map based on a preset strategy to obtain a feature-optimized multi-scale fusion feature map;

[0010] The created image data set is used as a training set to train the constructed initial model, and the trained model is used as a small component fault feature analysis model; the image data set includes various defect images and standard images containing the target small component category;

[0011] The feature-optimized multi-scale fusion feature map is analyzed based on the small component fault feature analysis model to obtain small component fault identification information.

[0012] Optionally, performing multi-scale processing on the scene image to obtain a multi-scale image feature map and a multi-scale sub-region feature map includes:

[0013] Converting the on-site image into a standard-sized aerial image;

[0014] Performing multi-scale transformation on the standard-size aerial image to obtain a multi-scale transformed image set, wherein the multi-scale transformed image set includes a plurality of scale transformed images;

[0015] Performing feature extraction on all the scale-transformed images to obtain a multi-scale image feature map;

[0016] Performing image cropping on the scale-transformed image to obtain a sub-region image containing the target widget category;

[0017] Feature extraction is performed on all the sub-region images to obtain a multi-scale sub-region feature map.

[0018] Optionally, the standard-sized aerial image is a square whose side length is an integer multiple of 128;

[0019] The size of the cropping is a square whose side length is an integer multiple of 32;

[0020] The image scales of the multiple scale-transformed images are respectively 1, 1 / 2, 1 / 4, and 1 / 8 of the original image.

[0021] Optionally, the performing enhancement processing on the multi-scale sub-region feature map to obtain the sub-region enhanced feature map includes:

[0022] Performing convolution processing on the multi-scale sub-region feature map to obtain a feature space matrix;

[0023] Performing spatial pooling encoding on the feature space matrix to obtain an X-direction encoding matrix and a Y-direction encoding matrix;

[0024] Performing weight correction processing on the X-direction encoding matrix and the Y-direction encoding matrix to obtain a weight correction fusion matrix;

[0025] Splitting the weight correction fusion matrix to obtain a one-dimensional tensor in the X direction and a one-dimensional tensor in the Y direction;

[0026] Performing fusion encoding processing on the one-dimensional tensor in the X direction and the one-dimensional tensor in the Y direction to obtain a local feature enhancement feature map;

[0027] The local feature enhancement feature map is processed based on an edge enhancement strategy to obtain a sub-region enhancement feature map.

[0028] Optionally, the processing of the local feature enhancement feature map based on the edge enhancement strategy to obtain the sub-region enhancement feature map includes:

[0029] Performing edge enhancement processing on the local feature enhancement feature map to obtain an edge enhancement feature map;

[0030] The edge enhancement feature map and the local feature enhancement feature map are element-wise accumulated, and then multiplied with the multi-scale sub-region feature map to obtain a sub-region enhancement feature map.

[0031] Optionally, the processing of the multi-scale target area enhanced feature map based on a preset strategy to obtain a feature-optimized multi-scale fusion feature map includes:

[0032] Determine the improved feature fusion pyramid;

[0033] Based on the n improved feature fusion pyramids, the multi-scale target area enhanced feature map is processed in parallel to obtain n sub-output feature maps;

[0034] Each of the sub-output feature maps is processed into blocks to obtain k sub-feature information blocks;

[0035] Obtain the sub-feature information blocks in the same column of n sub-output feature maps, recorded as a first subset;

[0036] Performing mean processing on the first subset to obtain a sub-feature optimization information block;

[0037] Based on the k sub-feature optimization information blocks, a feature-optimized multi-scale fusion feature map is obtained.

[0038] Optionally, the method for obtaining the sub-output feature map includes:

[0039] Performing top-down feature fusion on the multi-scale target region enhanced feature map to obtain a first feature map;

[0040] Performing bottom-up feature fusion on the first feature map to obtain a second feature map;

[0041] Based on the first convolutional layer, a convolution operation is performed on the features of the lowest feature level in the second feature map to obtain the first enhanced features of the lowest level;

[0042] Put the bottom layer first enhancement feature into the first feature class to obtain the first class enhancement feature;

[0043] Performing a convolution operation on the features of the lowest feature level in the second feature map based on the second convolution layer to obtain the bottom second enhanced features;

[0044] Put the bottom second enhanced feature into the second feature class to obtain the second class enhanced feature;

[0045] Based on the third convolutional layer, a convolution operation is performed on the features of the lowest feature level in the second feature map to obtain the third enhanced features of the bottom layer;

[0046] Put the bottom third enhancement feature into the third feature class to obtain the third class enhancement feature;

[0047] A sub-output feature map is obtained based on the features of the lowest feature class, the enhanced features of the first class, the enhanced features of the second class, and the enhanced features of the third class.

[0048] Optionally, the acquiring of the on-site image containing the target widget category comprises: acquiring on-site images of different angles containing the target widget category on the transmission line;

[0049] The number of the on-site images is M, where M≥2;

[0050] The target small component category is a bolt, an insulator, a suspension wire clamp or a shock-absorbing hammer.

[0051] Optionally, the analyzing the feature-optimized multi-scale fusion feature map based on the small component fault feature analysis model to obtain small component fault identification information includes:

[0052] Based on the small component failure feature analysis model, a plurality of the multi-scale fusion feature maps are analyzed to obtain small component failure identification information.

[0053] Optionally, analyzing the plurality of multi-scale fusion feature maps based on the small component fault feature analysis model to obtain small component fault identification information includes:

[0054] Analyze the plurality of multi-scale fusion feature maps based on the small component failure feature analysis model to obtain a plurality of analysis results;

[0055] Determine whether the plurality of analysis results are the same, and if so, use the analysis results as the small component fault identification information; if not, obtain the largest number of analysis results and record them as the first data set;

[0056] Determine whether the analysis results contained in the first data set are the same. If so, use the analysis result as the small component fault identification information; if not, determine a unique analysis result in combination with a preset experience library, and use the analysis result as the small component fault identification information.

[0057] In a second aspect, the embodiment of the present disclosure further provides a transmission line small component fault identification system, comprising:

[0058] An acquisition module, used for acquiring on-site images containing target small component categories on the transmission line;

[0059] An image processing module, used for performing multi-scale processing on the scene image to obtain a multi-scale image feature map and a multi-scale sub-region feature map;

[0060] An enhancement module is used to enhance the multi-scale sub-region feature map to obtain a sub-region enhanced feature map;

[0061] A fusion module is used to perform feature fusion on the sub-region enhanced feature map and the multi-scale image feature map to obtain a multi-scale target region enhanced feature map;

[0062] An optimization module processes the multi-scale target area enhanced feature map based on a preset strategy to obtain a feature-optimized multi-scale fusion feature map;

[0063] A training module, using the created image data set as a training set to train the constructed initial model, and using the trained model as a small component fault feature analysis model; the image data set includes various defect images and standard images containing the target small component category;

[0064] The identification module analyzes the feature-optimized multi-scale fusion feature map based on the small component fault feature analysis model to obtain small component fault identification information.

[0065] In a third aspect, the embodiments of the present disclosure further provide a computer device, which adopts the following technical solution:

[0066] The computer device comprises:

[0067] at least one processor; and,

[0068] a memory communicatively connected to the at least one processor; wherein,

[0069] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform any of the above-mentioned power transmission line widget fault identification methods.

[0070] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute any of the above-mentioned methods for identifying faults of small components in power transmission lines.

[0071] In a fifth aspect, an embodiment of the present disclosure further provides a computer program product, including a computer program / instruction, which implements the steps of any of the above methods when executed by a processor.

[0072] The method for identifying faults of small components of power transmission lines disclosed in the present application performs multi-scale processing on the acquired on-site images to generate a multi-scale image feature map and a multi-scale sub-region feature map, so as to capture small components of different sizes; performs enhancement processing on the multi-scale sub-region feature map to generate a sub-region enhancement feature map, and then performs feature fusion of the sub-region enhancement feature map with the multi-scale image feature map to obtain a multi-scale target region enhancement feature map, which can effectively enhance the feature expression of key areas in the image, help improve the recognition accuracy, and at the same time reduce the occlusion effect between targets and the interference of background clutter, so as to make target recognition more stable and accurate; then use the improved The feature fusion pyramid is used to process the multi-scale target area enhanced feature map to generate a feature-optimized multi-scale fusion feature map, which can effectively combine high- and low-level features to capture the details and overall information of the target at the same time; an initial model is constructed and trained using an image dataset containing defects and standard images to form a small component fault feature analysis model to ensure that the model can generalize to handle a variety of fault types; based on the trained small component fault feature analysis model, the feature-optimized multi-scale fusion feature map is analyzed to output small component fault identification information, that is, the feature extraction and classification capabilities learned during the training process are used to achieve accurate identification of small component faults.

[0073] The method disclosed in the present application solves the problem that the same target presents different scales and shapes at different distances through multi-scale processing and feature fusion; feature enhancement and fusion reduce the interference between targets and the background, and improve the target recognition ability in complex backgrounds; through the extensiveness of the training data set and the generalization ability of the model, it can achieve accurate identification of small component faults of transmission lines and adapt to the transmission line inspection needs under different terrains in my country.

[0074] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following preferred embodiments are specifically cited and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0076] Figure 1 A schematic flow chart of a method for identifying a small component fault in a power transmission line provided in an embodiment of the present disclosure.

[0077] Figure 2 A schematic flow chart of a method for performing multi-scale processing of scene images provided in an embodiment of the present disclosure.

[0078] Figure 3 A schematic diagram of the flow of a method for acquiring a sub-region enhanced feature map provided in an embodiment of the present disclosure.

[0079] Figure 4 A schematic flow chart of a method for processing a local feature enhancement feature map based on an edge enhancement strategy provided in an embodiment of the present disclosure.

[0080] Figure 5 A schematic diagram of a flow chart of a method for acquiring a feature-optimized multi-scale fusion feature map provided in an embodiment of the present disclosure.

[0081] Figure 6 A schematic flowchart of a method for obtaining a sub-output feature map provided in an embodiment of the present disclosure.

[0082] Figure 7 A schematic flow chart of a method for analyzing multiple multi-scale fusion feature maps based on a small component fault feature analysis model provided in an embodiment of the present disclosure.

[0083] Figure 8 A schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0084] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0085] It should be clear that the following embodiments of the present disclosure are described by specific specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other in the absence of conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0086] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0087] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0088] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0089] Reference Figure 1 The present application discloses a method for identifying faults of small components in a power transmission line, which specifically comprises the following steps:

[0090] S100, acquiring a field image containing a target small component category on a transmission line.

[0091] In this embodiment, aerial photography is performed by using a drone to obtain on-site images.

[0092] Among them, the target small component categories can be bolts, insulators, suspension clamps, connectors, lightning arresters or shock-absorbing hammers and other common small components of transmission lines.

[0093] S200, performing multi-scale processing on the scene image to obtain a multi-scale image feature map and a multi-scale sub-region feature map.

[0094] Multi-scale processing can capture the features of target widgets at different scales. The extraction of multi-scale sub-region feature maps helps to focus on the key areas of the target widgets and reduce the interference of background noise.

[0095] S300, performing enhancement processing on the multi-scale sub-region feature map to obtain a sub-region enhanced feature map;

[0096] The sub-region enhanced feature map is fused with the multi-scale image feature map to obtain a multi-scale target region enhanced feature map.

[0097] In this step, sub-region enhancement processing helps to highlight the key features of the target small parts and improve the recognition sensitivity of the model; the feature fusion operation can integrate global and local features and enhance the discrimination ability, so that the faults of small parts can be accurately identified even in complex backgrounds.

[0098] S400, processing the multi-scale target area enhanced feature map based on a preset strategy to obtain a feature-optimized multi-scale fusion feature map.

[0099] In this step, the improved feature fusion pyramid can better integrate multi-scale features and enhance the ability to express global and local information.

[0100] S500, build the initial model.

[0101] Specifically, you can choose a suitable deep learning model architecture (such as YOLO, SSD, etc.) as the initial model.

[0102] S600, using the created image data set as a training set to train the initial model, and using the trained model as a small component fault feature analysis model; the image data set includes various defect images and standard images containing the target small component category.

[0103] In this step, a rich image dataset can provide diverse training samples, enhance the generalization ability of the model, and reduce the risk of overfitting; data enhancement technology can simulate various changes in actual scenes and improve the recognition performance of the model in complex environments.

[0104] By using the created image dataset as a training set, it can flexibly meet the recognition needs in different scenarios, has strong adaptability, and can effectively ensure the reliability of the recognition results.

[0105] S700, analyzing the feature-optimized multi-scale fusion feature map based on the small component fault feature analysis model to obtain small component fault identification information.

[0106] In this step, the small component fault feature analysis model can quickly and accurately identify the fault status of various small components on the transmission line; the output fault identification information can be used for subsequent maintenance decisions to improve the operation and maintenance efficiency and safety of the transmission line.

[0107] The method for identifying faults of small components of power transmission lines disclosed in the present application performs multi-scale processing on the acquired on-site images to generate a multi-scale image feature map and a multi-scale sub-region feature map, so as to capture small components of different sizes; performs enhancement processing on the multi-scale sub-region feature map to generate a sub-region enhancement feature map, and then performs feature fusion of the sub-region enhancement feature map with the multi-scale image feature map to obtain a multi-scale target region enhancement feature map, which can effectively enhance the feature expression of key areas in the image, help improve the recognition accuracy, and at the same time reduce the occlusion effect between targets and the interference of background clutter, so as to make target recognition more stable and accurate; then use the improved The feature fusion pyramid is used to process the multi-scale target area enhanced feature map to generate a feature-optimized multi-scale fusion feature map, which can effectively combine high- and low-level features to capture the details and overall information of the target at the same time; an initial model is constructed and trained using an image dataset containing defects and standard images to form a small component fault feature analysis model to ensure that the model can generalize to handle a variety of fault types; based on the trained small component fault feature analysis model, the feature-optimized multi-scale fusion feature map is analyzed to output small component fault identification information, that is, the feature extraction and classification capabilities learned during the training process are used to achieve accurate identification of small component faults.

[0108] The method disclosed in the present application solves the problem that the same target presents different scales and shapes at different distances through multi-scale processing and feature fusion; feature enhancement and fusion reduce the interference between targets and the background, and improve the target recognition ability in complex backgrounds; through the extensiveness of the training data set and the generalization ability of the model, it can achieve accurate identification of small component faults of transmission lines and adapt to the transmission line inspection needs under different terrains in my country.

[0109] Reference Figure 2 The method for multi-scale processing of the scene image (i.e., the method for obtaining the multi-scale image feature map and the multi-scale sub-region feature map) specifically includes the following steps:

[0110] S210, converting the on-site image into an aerial image of a standard size.

[0111] Among them, the standard size aerial image is a square with a side length that is an integer multiple of 128.

[0112] Through this step, the influence of different shooting conditions (such as lighting, angle, and distance) on the image can be eliminated, and the stability of feature extraction can be improved.

[0113] S220, performing multi-scale transformation on the standard-sized aerial image to obtain a multi-scale transformed image set, where the multi-scale transformed image set includes a plurality of scale transformed images.

[0114] Specifically, a standard-sized aerial image is transformed into four different scales, and the scales of the multiple scale-transformed images obtained after the transformation are 1, 1 / 2, 1 / 4, and 1 / 8 of the original image, respectively.

[0115] Through this step, representations of the image at different scales can be provided, which can capture the characteristics of the target widgets at different scales.

[0116] S230, performing feature extraction on all scale-transformed images to obtain a multi-scale image feature map.

[0117] Specifically, all scale-transformed images can be input into the Resnet feature extraction network for feature extraction to obtain a multi-scale image feature map.

[0118] For example, if all scale-transformed images include P1, P2, P3, and P4, the multi-scale image feature map is obtained as follows: .

[0119] S240, performing image cropping on the scale-transformed image to obtain a sub-region image containing the target widget category.

[0120] The cropped size is a square whose side length is an integer multiple of 32.

[0121] Specifically, the target widget category is included in the area where the target widget category is located in the four scale-transformed images of different scales with a square size whose side length is an integer multiple of 32, and the sub-area image corresponding to the target widget category is obtained, that is, the corresponding sample area of ​​interest in the preprocessed sample image is obtained.

[0122] Focus on processing the local area containing the target widget to reduce the impact of background noise; the sub-region image is helpful to extract the fine features of the target widget and improve the accuracy of classification and recognition.

[0123] S250, extracting features from all sub-region images to obtain multi-scale sub-region feature maps.

[0124] Specifically, all sub-region images are input into the Resnet feature extraction network for feature extraction to obtain multi-scale sub-region feature maps.

[0125] For example, all sub-region images include: , then the obtained multi-scale sub-region feature map is: .

[0126] In this embodiment, the method for multi-scale processing of the scene image disclosed in S210-S250 provides a comprehensive feature representation of the target widget at different scales through multi-scale processing and feature extraction; the sub-region feature map captures the fine features of the target widget, reduces the influence of background noise, and improves the accuracy of classification and recognition; the multi-scale processing and feature extraction method enhances the scale invariance of the model and is suitable for target detection and recognition tasks of different sizes and shapes. In addition, the scheme can adjust the scale and feature extraction method according to specific task requirements, and has high flexibility and scalability.

[0127] Reference Figure 3 The method for enhancing the multi-scale sub-region feature map (i.e., the method for obtaining the sub-region enhanced feature map) specifically includes the following steps:

[0128] A100, performs convolution processing on the multi-scale sub-region feature maps to obtain the feature space matrix.

[0129] Specifically, two 1*1 convolutional layers are used to perform two convolutions on the multi-scale sub-region feature maps to convert them into potential feature spaces, that is, to obtain feature space matrices.

[0130] In this step, the convolution operation can effectively extract local features and improve the expressiveness of the feature space; through convolution processing, the high-dimensional feature map can be converted into a low-dimensional feature space matrix, reducing the computational complexity.

[0131] A200 performs spatial pooling encoding on the feature space matrix to obtain an X-direction encoding matrix and a Y-direction encoding matrix.

[0132] Specifically, the two spatial pooling kernels encode the output of the convolutional layer (i.e., the feature space matrix) along the X and Y directions to obtain the X-direction encoding matrix and the Y-direction encoding matrix.

[0133] Specific encoding methods include:

[0134]

[0135] Wherein, c represents the cth channel.

[0136] Spatial pooling can be used to reduce feature dimensions and subsequent computational burdens; pooling operations can effectively integrate local features and enhance the ability to focus on key features.

[0137] A300 performs weight correction processing on the X-direction coding matrix and the Y-direction coding matrix to obtain a weight correction fusion matrix.

[0138] Specifically, after the X-direction encoding matrix and the Y-direction encoding matrix are concat, they are input into a 1*1 convolution layer, and operations are performed to accelerate the convergence of weights to ensure that the feature weights do not change significantly in the two directions, that is, to ensure stability.

[0139] Through weight correction, the information of important features can be enhanced and the influence of irrelevant information can be reduced. The weights can be dynamically adjusted according to the importance of the features to improve the adaptability and expressiveness of the model.

[0140] A400, splits the weight correction fusion matrix to obtain a one-dimensional tensor in the X direction and a one-dimensional tensor in the Y direction.

[0141] Specifically, the weight correction fusion matrix can be divided into two independent tensors, namely:

[0142] ;

[0143] In this way, the original two-dimensional features are converted into one-dimensional features encoded in different directions, reducing the computational space.

[0144] A500 performs fusion encoding processing on the one-dimensional tensor in the X direction and the one-dimensional tensor in the Y direction to obtain a local feature enhanced feature map.

[0145] Specifically, perform matrix product of the one-dimensional tensor in the X direction and the one-dimensional tensor in the Y direction, and then input the product into a The activation function is used, and a spatial pooling kernel is used to encode it on the channel to obtain the local feature enhancement feature map .

[0146] The specific calculation process includes:

[0147] ,

[0148] Among them, c is the number of the cth channel, and C is the total number of channels.

[0149] Through the fusion encoding processing of local features, the expressiveness of feature maps can be effectively enhanced, and the performance of target detection and classification can be improved; the fusion operation allows the integration of information from two dimensions, comprehensively capturing the characteristics of the target and improving accuracy.

[0150] A600, processes the local feature enhancement feature map based on the edge enhancement strategy to obtain the sub-region enhancement feature map.

[0151] Through the edge enhancement strategy, the edge information of the image is further improved and the discrimination ability of the feature map is enhanced; edge enhancement can highlight the important edge information in the image, so that the feature map performs better in classification and recognition tasks.

[0152] In this embodiment, the method disclosed in A100-A600 for enhancing the multi-scale sub-region feature map (i.e., the method for obtaining the sub-region enhanced feature map) can extract deep features in the image and enhance the feature recognition ability through multi-scale convolution processing and spatial pooling encoding; weight correction and fusion coding processing can improve the representativeness of the features and make the classification task more accurate; spatial pooling and splitting operations can reduce the computational complexity and make the entire processing process more efficient; the edge enhancement strategy can highlight the important edge information in the image, so that the feature map performs better in classification and recognition tasks; this method is suitable for processing complex images and multi-scale features, and can effectively extract and enhance the key information in the image.

[0153] Reference Figure 4 The method for processing the local feature enhancement feature map based on the edge enhancement strategy specifically includes:

[0154] A610, performs edge enhancement processing on the local feature enhancement feature map to obtain an edge enhancement feature map.

[0155] Specifically, the edge enhancement feature map is :

[0156] .

[0157] Through edge enhancement processing, the edge information in the image is highlighted, making the boundaries of objects in the image clearer; edge enhancement processing can improve the contrast of the edges in the image, making subsequent feature fusion more effective and enhancing the recognition ability of the feature map.

[0158] A620, element-wise accumulation of the edge enhancement feature map and the local feature enhancement feature map, and then multiplication of the feature map with the multi-scale sub-region feature map to obtain the sub-region enhancement feature map.

[0159] Specifically, the sub-region enhanced feature map is :

[0160] ;in, Multi-scale sub-region feature map.

[0161] In order to standardize the whole process, the output Output is recorded as: .

[0162] Through element accumulation and multiplication operations, edge information and local feature information can be effectively combined to generate richer and more accurate sub-region enhanced feature maps.

[0163] In this embodiment, the method disclosed in A610-A620 for processing a local feature enhancement feature map based on an edge enhancement strategy can highlight the edge information in the image through edge enhancement processing, making the object boundaries clearer, which helps to improve the accuracy of subsequent classification and recognition; through element accumulation operation, the information of the edge enhancement feature map and the local feature enhancement feature map is fused together to generate a more comprehensive feature expression; the multiplication operation can further enhance the expression ability of the feature, so that the feature map performs better in classification and recognition tasks; by multiplying with the multi-scale sub-region feature map, the multi-scale information can be effectively integrated into the final feature map, so that the feature map has good performance at different scales.

[0164] The overall method can effectively improve the feature recognition ability of the image, thereby improving the accuracy of classification and recognition, and is suitable for various complex image processing tasks; through edge enhancement and feature fusion, the robustness of the feature map can be improved, making it more stable in the face of noise and interference.

[0165] Reference Figure 5 , a method for obtaining a feature-optimized multi-scale fusion feature map includes:

[0166] S410, determining an improved feature fusion pyramid.

[0167] S420, performing parallel processing on the multi-scale target region enhanced feature map based on the n improved feature fusion pyramids to obtain n sub-output feature maps.

[0168] Among them, the number of basic units n can be dynamically adjusted according to the actual training effect.

[0169] Parallel processing can improve computational efficiency and obtain n sub-output feature maps at the same time, each of which contains information of different scales and perspectives.

[0170] S430, each sub-output feature map is divided into blocks to obtain k sub-feature information blocks.

[0171] Specifically include:

[0172]

[0173] in, It represents the kth sub-feature information block obtained after splitting the sub-output of the nth basic processing unit.

[0174] Each sub-output feature map is divided into k sub-feature information blocks for more fine-grained processing; block processing can better analyze and optimize local feature information, providing a basis for subsequent feature optimization.

[0175] S440, obtaining sub-feature information blocks in the same column of n sub-output feature maps, and recording them as a first subset.

[0176] By extracting sub-feature information blocks in the same column, multi-scale information at the same position can be fused to improve the richness and accuracy of features.

[0177] S450, performing mean processing on the first subset to obtain a sub-feature optimization information block.

[0178] Specifically, the sub-feature optimization information block is , ,in, .

[0179] Mean processing can smooth and enhance feature information, reduce the impact of noise and outliers, and obtain more stable and accurate sub-feature optimization information blocks.

[0180] S460, obtaining a feature-optimized multi-scale fusion feature map based on the k sub-feature optimization information blocks.

[0181] The obtained feature optimized multi-scale fusion feature map is:

[0182] .

[0183] The combined optimized sub-feature information blocks can generate a more comprehensive and accurate multi-scale fusion feature map, improving the overall feature expression capability. This processing can effectively alleviate the problem of feature information loss and feature information ambiguity.

[0184] In this embodiment, the method for acquiring the feature optimized multi-scale fusion feature map disclosed in S410-S460 can better capture and fuse multi-scale feature information and enhance the expressiveness of features by improving the feature fusion pyramid and parallel processing; mean processing of sub-feature information blocks can reduce the influence of noise and outliers and improve the stability and robustness of features; fine-grained feature optimization can be achieved through block processing and subset mean processing, so that each local feature is fully enhanced and fused; parallel processing of n feature fusion pyramids can significantly improve computing efficiency and reduce processing time; by combining k sub-feature optimization information blocks, a comprehensive and accurate multi-scale fusion feature map can be generated, thereby improving the performance of overall image processing and analysis.

[0185] Reference Figure 6 , a method for obtaining a sub-output feature map includes:

[0186] B100, perform feature fusion of the multi-scale target area enhanced feature map from top to bottom to obtain the first feature map.

[0187] Through the top-down feature fusion path, the abstract features of the high-level layers are transferred to the low-level layers, enhancing the overall expressiveness of the feature map.

[0188] The top-down path can help low-level feature maps retain more high-level semantic information, making the feature maps richer and more comprehensive.

[0189] B200, perform feature fusion of the first feature map from bottom to top to obtain the second feature map.

[0190] Through the bottom-up feature fusion path, the detailed features of the low layer are transferred to the high layer, enhancing the detail expression ability of the feature map.

[0191] The bottom-up path can help high-level feature maps retain more detail information, so that the feature maps have higher spatial resolution while maintaining semantic information.

[0192] B300, based on the first convolutional layer, a convolution operation is performed on the features of the lowest feature class in the second feature map to obtain the first enhanced features of the bottom layer.

[0193] Specifically, based on the first convolution layer with a kernel of 3x3 and a step size of 2, a convolution operation is performed on the features of the lowest feature class in the second feature map (ie, the features of the 0th feature class) to obtain the first enhanced features of the bottom layer.

[0194] B400, put the bottom first enhanced feature into the first feature class to obtain the first class enhanced feature.

[0195] Specifically, the bottom layer first enhanced feature is placed into the first feature class in the second feature map to obtain the first class enhanced feature.

[0196] The enhanced underlying features are placed into the first feature class to enhance the feature expression capability of this class.

[0197] By enhancing the underlying features, the quality of the first-order features can be improved, making them perform better in subsequent processing.

[0198] B500, based on the second convolutional layer, performs a convolution operation on the features of the lowest feature class in the second feature map to obtain the bottom second enhanced features.

[0199] Specifically, based on the second convolutional layer with a kernel of 3x3 and a stride of 4, a convolution operation is performed on the features of the lowest feature class in the second feature map to obtain the bottom second enhanced features.

[0200] B600, put the bottom second enhancement feature into the second feature class to obtain the second class enhancement feature.

[0201] Specifically, the bottom-level second enhanced features are placed into the second feature class in the second feature map to obtain the second-class enhanced features.

[0202] By enhancing the underlying features, the quality of the second-order features can be improved, making them perform better in subsequent processing.

[0203] B700, based on the third convolutional layer, a convolution operation is performed on the features of the lowest feature class in the second feature map to obtain the third enhanced features of the bottom layer.

[0204] Based on the third convolutional layer with a kernel of 3x3 and a stride of 8, a convolution operation is performed on the features of the lowest feature class in the second feature map to obtain the third enhanced features of the bottom layer.

[0205] B800, put the bottom third enhancement feature into the third feature class to obtain the third class enhancement feature.

[0206] The bottom third enhanced feature is placed into the third feature class in the second feature map to obtain the third class enhanced feature.

[0207] B900, obtains a sub-output feature map based on the features of the lowest feature class, the enhanced features of the first class, the enhanced features of the second class, and the enhanced features of the third class.

[0208] Specifically, the features of the 0th feature class are passed through three convolutional layers with a kernel of 3x3 and step sizes of 2, 4, and 8, and then merged with the features of the 1st, 2nd, and 3rd feature classes, respectively. That is, three additional short branches from bottom to top are expanded to improve the detection accuracy of small objects.

[0209] By combining the bottom-level features and multi-level enhanced features, the final sub-output feature map is generated. Through this step, combining multi-level features can generate a more comprehensive and rich feature map, improving the expressiveness and complexity of the overall features.

[0210] In this embodiment, the method for obtaining the sub-output feature map disclosed by B100-B900, based on the path structure of the above traditional feature fusion pyramid, expands multiple bottom-up branches, makes multiple use of small-scale features, enhances the network's ability to focus on small component targets, and improves the recognition accuracy of similar targets of different scales. Specifically, through multi-level convolution and feature fusion, the expressiveness of the feature map can be significantly enhanced, making the feature map richer and more comprehensive; multi-layer convolution operations can extract feature information at different levels, making the feature map more complex and representative; through top-down and bottom-up feature fusion paths, multi-scale feature information can be retained, so that the feature map has good performance at different scales; through multi-level convolution operations, the expressiveness of the underlying features can be effectively enhanced, and the overall quality of the feature map can be improved; the overall method can improve the expressiveness and complexity of the feature map, thereby improving the accuracy of classification and recognition.

[0211] Furthermore, the method disclosed in the present application also includes: acquiring on-site images of different angles containing target widget categories on the transmission line;

[0212] The number of on-site images is M, where M≥2.

[0213] Through image acquisition at multiple angles and time points, the states of target widgets under different lighting conditions and environments can be captured, enhancing the diversity and robustness of the dataset.

[0214] In this embodiment, the feature-optimized multi-scale fusion feature map is analyzed based on the small component fault feature analysis model to obtain small component fault identification information, including: analyzing multiple multi-scale fusion feature maps based on the small component fault feature analysis model to obtain small component fault identification information.

[0215] specifically refer to Figure 7 , a method for analyzing multiple multi-scale fusion feature maps based on a small component fault feature analysis model, comprising:

[0216] Based on the small component fault feature analysis model, multiple multi-scale fusion feature maps are analyzed to obtain multiple analysis results;

[0217] Determine whether multiple analysis results are the same, if so, use the analysis results as the small component fault identification information; if not, obtain the largest number of analysis results and record them as the first data set;

[0218] Determine whether the analysis results included in the first data set are the same. If so, use the corresponding analysis result as the widget fault identification information; if not, determine a unique analysis result in combination with a preset experience library, and use the analysis result as the widget fault identification information.

[0219] Through comprehensive analysis of multiple multi-scale fusion feature maps, richer fault features can be obtained, thereby improving the accuracy and reliability of fault identification; the collection and comparison of multiple analysis results establishes a redundant mechanism. When certain features or parameters are abnormal, other analysis results can still support fault identification and enhance the robustness of the system; by obtaining the largest number of analysis results and recording them as the first data set, the possibility of misidentification can be effectively reduced, ensuring that fault identification is based on the most representative analysis results; when the results of the first data set are inconsistent, by combining the preset experience library, historical knowledge and experience can be used to make decisions to avoid identification problems caused by small sample errors.

[0220] This method realizes data-driven intelligent fault analysis. By comprehensively analyzing the information of multiple feature maps, it demonstrates the ability of intelligent fault detection and facilitates subsequent automated maintenance and decision-making. This implementation scheme is particularly suitable for failure situations of small components that require identification of multiple failure modes and complex features, and can handle highly complex and changeable fault scenarios.

[0221] The method of analyzing the feature-optimized multi-scale fusion feature graph based on the small component fault feature analysis model not only enhances the accuracy and robustness of fault identification through systematic information analysis and decision-making mechanisms, but also improves the quality and intelligence of decision-making by integrating the experience library. In practical applications, this method provides strong support for the fault identification and accurate analysis of small components, and can effectively cope with complex and changeable fault scenarios.

[0222] The transmission line small component fault identification method disclosed in this embodiment provides an innovative transmission line small component fault identification method, which can significantly improve the detection capability of multi-scale similar targets, especially the detection accuracy of small targets in drone inspections. This method is not only highly adaptable and can cope with the changes in the height of drone shooting and the scale differences of target objects, but also realizes the real-time transmission and intelligent analysis of inspection data through intelligent detection methods, providing comprehensive and timely decision-making support for power operation and maintenance departments, significantly improving the efficiency and response speed of transmission line inspections, and at the same time providing strong support for the safe operation and intelligent management of power systems.

[0223] In a second aspect, the present application discloses a transmission line small component fault identification system, comprising:

[0224] An acquisition module, used for acquiring on-site images containing target small component categories on the transmission line;

[0225] An image processing module is used to perform multi-scale processing on the scene image to obtain a multi-scale image feature map and a multi-scale sub-region feature map;

[0226] The enhancement module performs enhancement processing on the multi-scale sub-region feature map to obtain the sub-region enhanced feature map;

[0227] A fusion module is used to fuse the sub-region enhanced feature map with the multi-scale image feature map to obtain a multi-scale target region enhanced feature map;

[0228] The optimization module processes the multi-scale target area enhancement feature map based on a preset strategy to obtain a feature-optimized multi-scale fusion feature map;

[0229] A training module uses the created image dataset as a training set to train the constructed initial model, and uses the trained model as a small component fault feature analysis model; the image dataset includes various defect images and standard images containing the target small component category;

[0230] The recognition module uses a small component fault feature analysis model to analyze the feature-optimized multi-scale fusion feature map to obtain small component fault recognition information.

[0231] It should be noted that the solutions disclosed in the power transmission line widget fault identification method disclosed in the first aspect of the present application are all applicable to the power transmission line widget fault identification system disclosed in the second aspect of the present application, so they will not be repeated here.

[0232] The computer device according to the embodiment of the present disclosure includes a memory and a processor. The memory is used to store non-temporary computer-readable instructions. Specifically, the memory may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc.

[0233] The processor may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the computer device to perform desired functions. In one embodiment of the present disclosure, the processor is used to run the computer-readable instructions stored in the memory, so that the computer device performs all or part of the steps of the power transmission line widget fault identification method of each embodiment of the present disclosure.

[0234] Those skilled in the art should be able to understand that in order to solve the technical problem of how to obtain a good user experience, the present embodiment may also include well-known structures such as a communication bus and an interface, and these well-known structures should also be included in the protection scope of the present disclosure.

[0235] like Figure 8 A schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure is shown, which is a schematic diagram of the structure of a computer device suitable for implementing the embodiment of the present disclosure. Figure 8 The computer device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0236] like Figure 8 As shown, the computer device may include a processor (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) or a program loaded from a storage device into a random access memory (RAM). In the RAM, various programs and data required for the operation of the computer device are also stored. The processor, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0237] Typically, the following devices can be connected to the I / O interface: input devices such as sensors or visual information acquisition devices; output devices such as display screens; storage devices such as tapes, hard disks, etc.; and communication devices. The communication device can allow the computer device to communicate with other devices (such as edge computing devices) wirelessly or by wire to exchange data. Figure 8 A computer device having various devices is shown, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead.

[0238] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processor, all or part of the steps of the power transmission line widget fault identification method of the embodiment of the present disclosure are executed.

[0239] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0240] The computer-readable storage medium according to the embodiment of the present disclosure stores non-transitory computer-readable instructions, and when the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the power transmission line widget fault identification method of each embodiment of the present disclosure are executed.

[0241] The above-mentioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (e.g., memory card) and media with built-in ROM (e.g., ROM box).

[0242] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0243] The basic principles of the present disclosure are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present disclosure are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present disclosure. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, and are not limitations. The above details do not limit the present disclosure to the necessity of adopting the above specific details to be implemented.

[0244] In the present disclosure, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. The block diagrams of the devices, devices, equipment, and systems involved in the present disclosure are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The words "such as" used here refer to the phrase "such as but not limited to", and can be used interchangeably with them.

[0245] Additionally, as used herein, "or" used in a list of items beginning with "at least one" indicates a separate list, so that, for example, a list of "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not mean that the example described is preferred or better than other examples.

[0246] It should also be noted that in the system and method of the present disclosure, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present disclosure.

[0247] Various changes, substitutions, and modifications of the techniques described herein may be made without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of the present disclosure is not limited to the specific aspects of the processes, machines, manufactures, compositions of events, means, methods, and actions described above. Currently existing or later to be developed processes, machines, manufactures, compositions of events, means, methods, or actions that perform substantially the same functions or achieve substantially the same results as the corresponding aspects described herein may be utilized. Thus, the appended claims include such processes, machines, manufactures, compositions of events, means, methods, or actions within their scope.

[0248] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0249] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A method for identifying faults of small components in a power transmission line, characterized in that: include: Acquire a field image containing a target widget class on a transmission line; Performing multi-scale processing on the scene image to obtain a multi-scale image feature map and a multi-scale sub-region feature map; Performing enhancement processing on the multi-scale sub-region feature map to obtain a sub-region enhanced feature map; Performing feature fusion on the sub-region enhancement feature map and the multi-scale image feature map to obtain a multi-scale target region enhancement feature map; Processing the multi-scale target area enhanced feature map based on a preset strategy to obtain a feature-optimized multi-scale fusion feature map; The created image data set is used as a training set to train the constructed initial model, and the trained model is used as a small component fault feature analysis model; the image data set includes various defect images and standard images containing the target small component category; Analyze the feature-optimized multi-scale fusion feature map based on the small component fault feature analysis model to obtain small component fault identification information; The method processes the multi-scale target area enhanced feature map based on a preset strategy to obtain a feature-optimized multi-scale fusion feature map, including: determining an improved feature fusion pyramid; performing parallel processing on the multi-scale target area enhanced feature map based on n improved feature fusion pyramids to obtain n sub-output feature maps; performing block processing on each of the sub-output feature maps to obtain k sub-feature information blocks; obtaining the sub-feature information blocks in the same column of the n sub-output feature maps, recorded as a first subset; performing mean processing on the first subset to obtain a sub-feature optimized information block; and obtaining a feature-optimized multi-scale fusion feature map based on the k sub-feature optimized information blocks.

2. The method for identifying faults of small components in power transmission lines according to claim 1, characterized in that: The performing multi-scale processing on the scene image to obtain a multi-scale image feature map and a multi-scale sub-region feature map includes: Converting the on-site image into a standard-sized aerial image; Performing multi-scale transformation on the standard-size aerial image to obtain a multi-scale transformed image set, wherein the multi-scale transformed image set includes a plurality of scale transformed images; Performing feature extraction on all the scale-transformed images to obtain a multi-scale image feature map; Performing image cropping on the scale-transformed image to obtain a sub-region image containing the target widget category; Feature extraction is performed on all the sub-region images to obtain a multi-scale sub-region feature map.

3. The method for identifying faults of small components in power transmission lines according to claim 2, characterized in that: The standard-size aerial image is a square whose side length is an integer multiple of 128; The size of the cropping is a square whose side length is an integer multiple of 32; The image scales of the multiple scale-transformed images are respectively 1, 1 / 2, 1 / 4, and 1 / 8 of the original image.

4. The method for identifying faults of small components in power transmission lines according to claim 1, characterized in that: The step of performing enhancement processing on the multi-scale sub-region feature map to obtain a sub-region enhanced feature map includes: Performing convolution processing on the multi-scale sub-region feature map to obtain a feature space matrix; Performing spatial pooling encoding on the feature space matrix to obtain an X-direction encoding matrix and a Y-direction encoding matrix; Performing weight correction processing on the X-direction encoding matrix and the Y-direction encoding matrix to obtain a weight correction fusion matrix; Splitting the weight correction fusion matrix to obtain a one-dimensional tensor in the X direction and a one-dimensional tensor in the Y direction; Performing fusion encoding processing on the one-dimensional tensor in the X direction and the one-dimensional tensor in the Y direction to obtain a local feature enhancement feature map; The local feature enhancement feature map is processed based on an edge enhancement strategy to obtain a sub-region enhancement feature map.

5. The method for identifying faults of small components in power transmission lines according to claim 4, characterized in that: The step of processing the local feature enhancement feature map based on the edge enhancement strategy to obtain the sub-region enhancement feature map includes: Performing edge enhancement processing on the local feature enhancement feature map to obtain an edge enhancement feature map; The edge enhancement feature map and the local feature enhancement feature map are element-wise accumulated, and then multiplied with the multi-scale sub-region feature map to obtain a sub-region enhancement feature map.

6. The method for identifying faults of small components in power transmission lines according to claim 1, characterized in that: The method for obtaining the sub-output feature map comprises: Performing top-down feature fusion on the multi-scale target region enhanced feature map to obtain a first feature map; Performing bottom-up feature fusion on the first feature map to obtain a second feature map; Based on the first convolutional layer, a convolution operation is performed on the features of the lowest feature level in the second feature map to obtain the first enhanced features of the lowest level; Put the bottom layer first enhancement feature into the first feature class to obtain the first class enhancement feature; Performing a convolution operation on the features of the lowest feature level in the second feature map based on the second convolution layer to obtain the bottom second enhanced features; Put the bottom second enhanced feature into the second feature class to obtain the second class enhanced feature; Based on the third convolutional layer, a convolution operation is performed on the features of the lowest feature level in the second feature map to obtain the third enhanced features of the bottom layer; Put the bottom third enhancement feature into the third feature class to obtain the third class enhancement feature; A sub-output feature map is obtained based on the features of the lowest feature class, the enhanced features of the first class, the enhanced features of the second class, and the enhanced features of the third class.

7. The method for identifying faults of small components in a power transmission line according to any one of claims 1 to 6, characterized in that: The obtaining of the on-site image containing the target widget category comprises: obtaining on-site images at different angles containing the target widget category on the transmission line; The number of the on-site images is M, where M≥2; The target small component category is a bolt, an insulator, a suspension wire clamp, or an anti-vibration hammer.

8. The method for identifying faults of small components in power transmission lines according to claim 7, characterized in that: The step of analyzing the feature-optimized multi-scale fusion feature graph based on the small component fault feature analysis model to obtain small component fault identification information includes: Based on the small component failure feature analysis model, a plurality of the multi-scale fusion feature maps are analyzed to obtain small component failure identification information.

9. The method for identifying faults of small components in power transmission lines according to claim 8, characterized in that: The step of analyzing the plurality of multi-scale fusion feature maps based on the small component fault feature analysis model to obtain small component fault identification information includes: Analyze the plurality of multi-scale fusion feature maps based on the small component failure feature analysis model to obtain a plurality of analysis results; Determine whether the plurality of analysis results are the same, and if so, use the analysis results as the small component fault identification information; if not, obtain the largest number of analysis results and record them as the first data set; Determine whether the analysis results contained in the first data set are the same. If so, use the analysis result as the small component fault identification information; if not, determine a unique analysis result in combination with a preset experience library, and use the analysis result as the small component fault identification information.

Images