Power transmission line equipment detection method based on lossless data enhancement strategy

Abstract

The invention belongs to the technical field of image detection, and particularly relates to a power transmission line equipment detection method based on a lossless data enhancement strategy, and the method comprises the following steps: building an image detection model for detecting a plurality of types of power transmission line equipment according to different image types of the power transmission line equipment in an original image set in a database; the method comprises the following steps: preprocessing a to-be-detected power transmission line equipment image, and converting a target bounding box label in a COCO format into a YOLO format to adapt to the demand of a target detection task; training parameters are set, the established image detection model is trained, and an improved YOLOv10 model is obtained; and inputting the preprocessed image of the power transmission line equipment into the improved YOLOv10 model to obtain an image after data enhancement, and detecting the image to obtain the type of the power transmission line equipment. The method has excellent detection performance in a complex power transmission environment, the monitoring efficiency of the power transmission equipment can be effectively improved, and a reliable solution is provided for guaranteeing power supply.

Description

Technical Field

[0001] The invention belongs to the technical field of image detection, and in particular is a method for detecting power transmission line equipment based on a lossless data enhancement strategy. Background Art

[0002] As a key component of the power system, the stability and reliability of the operation of power transmission equipment are directly related to the safety of power supply. However, due to the complex structure of power transmission equipment and the variability of its operating environment, equipment detection faces many challenges: including target overlap, significant scale differences, difficulty in detecting small targets, and scarcity of training data. These technical difficulties have seriously affected the performance and generalization ability of the detection model, and it is urgent to optimize it comprehensively at the technical level.

[0003] In the problem of target overlap, transmission line equipment such as conductors, insulators and fasteners often overlap in images, which can lead to blurred feature expression and reduced detection accuracy. In terms of scale differences, the transmission equipment detection task involves both very large targets such as high-voltage towers and tiny targets such as bolts. Traditional models are difficult to adapt to the detection needs of large and small targets at the same time. In addition, the low accuracy of small target detection is a major pain point in transmission equipment detection. Small targets often lead to missed detections and false detections due to their low resolution and susceptibility to background interference. Data scarcity is also an important factor limiting the detection performance of transmission equipment. Due to the high collection cost and large labeling workload, the existing data is insufficient to support the efficient training of deep learning models.

[0004] Some optimization strategies have been proposed in the existing technology, such as using the attention mechanism to solve the problem of target overlap, enhancing the adaptability to targets of different sizes through a multi-scale network structure, and using data enhancement technology to alleviate the problem of insufficient data. However, these technologies still have performance bottlenecks in detection tasks under complex power transmission environments. Therefore, a power transmission equipment detection method that combines innovative feature extraction mechanisms, multi-scale feature processing capabilities, and lossless data enhancement strategies is needed to comprehensively improve detection accuracy, robustness, and generalization capabilities, thereby meeting the actual needs under complex power transmission environments. Summary of the invention

[0005] The object of the present invention is to provide a power transmission line equipment detection method based on a lossless data enhancement strategy to solve the problem of low accuracy of the power transmission line equipment method in the prior art.

[0006] The technical solution adopted by the present invention to achieve the above-mentioned purpose is: a transmission line equipment detection method based on a lossless data enhancement strategy, comprising the following steps:

[0007] Step S1: establishing an image detection model for detecting multiple types of transmission line equipment according to different image types of transmission line equipment in the original image set in the database;

[0008] Step S2: pre-process the image of the power transmission line equipment to be detected, and convert the target bounding box annotation in COCO format into YOLO format to meet the requirements of the target detection task;

[0009] Step S3: setting training parameters, training the established image detection model, and obtaining an improved YOLOv10 model;

[0010] Step S4: input the preprocessed transmission line equipment image into the improved YOLOv10 model to obtain a data-enhanced image, detect the data-enhanced image, obtain the type of the transmission line equipment, and complete the transmission line equipment detection.

[0011] The image types include: target overlap, large target size differences in multi-scale detection tasks, difficulty in detecting small targets, and data scarcity.

[0012] In step S1, the image detection model for detecting various types of power transmission line equipment is established, specifically:

[0013] 1) To address target overlap, the ACmix model is introduced into the image detection model. By combining the convolution mechanism with the self-attention mechanism, a feature processing module is formed to dynamically adjust the target feature extraction process to improve the detection accuracy in overlapping scenarios;

[0014] 2) In view of the large differences in target sizes in multi-scale detection tasks, the GELAN module is designed to enhance the robustness of the model in multi-scale target detection by separating the input features into multiple sub-features with different receptive fields and combining them with the PGI mechanism;

[0015] 3) To address the difficulty in detecting small targets, an adaptive spatial feature fusion module is used to extract key information from feature maps of different resolutions and perform dynamic weighted fusion to optimize the complementarity of low-resolution and high-resolution features, thereby improving the consistency and accuracy of small target detection in complex scenes.

[0016] 4) To address the problem of data scarcity, a lossless data augmentation strategy is used to generate diversified training data while maintaining the key information of the original image through geometric transformation and visual adjustment methods, thereby significantly improving the generalization and anti-interference capabilities of the model.

[0017] In step 1), the feature processing module performs the following steps:

[0018] 1-1) Perform a projection operation on the input feature map and generate multiple sets of intermediate feature maps through lightweight convolution, which provide a basis for subsequent feature aggregation;

[0019] 1-2) Process the intermediate feature map along the convolution path and self-attention path respectively;

[0020] 1-2-1) In the self-attention path, the feature graph is decomposed into query, key and value. The attention weight is generated by calculating the similarity between the query and the key. The weight is used to perform weighted summation on the value to obtain the global relationship between the dynamic capture targets.

[0021] 1-2-2) In the convolution path, the intermediate feature map further enhances the local features through lightweight convolution operations, and combines specific position adjustment strategies to retain the accurate expression of detail information;

[0022] 1-3) The ACmix model fuses the output of the self-attention path and the convolution path to obtain a feature map.

[0023] The step 2) is specifically:

[0024] 2-1) First, the input feature map undergoes an initial convolution operation to generate an intermediate feature map with a fixed number of channels;

[0025] 2-2) The intermediate feature map is divided into two parts, each with an equal number of channels to achieve division of labor for different channels; each part of the feature map is processed by the RepNCSP module respectively; the RepNCSP module combines convolution kernels of different sizes to capture features within different receptive fields and enhance the fine-grained expression capability of features;

[0026] 2-3) Re-integrate multiple feature blocks through splicing operations to form a comprehensive feature map;

[0027] 2-4) The fused feature map is subjected to convolution operation to generate a new feature map, providing high-quality feature expression for subsequent detection tasks.

[0028] The step 3) is specifically:

[0029] 3-1) The adaptive spatial feature fusion module extracts key information from feature maps of multiple resolutions;

[0030] 3-2) The adaptive spatial feature fusion module adjusts the weight of each type of feature map layer by layer, so that the feature contribution of each scale can be dynamically allocated according to the requirements of the detection task;

[0031] 3-3) Through weighted adjustment and fusion operations, the adaptive spatial feature fusion module generates a unified feature map.

[0032] In step 3), the lossless data enhancement strategy includes: vertical flipping and brightness contrast adjustment.

[0033] The vertical flipping is to enhance the detection capability of the model for symmetrical targets without changing the original pixel information by flipping the input image symmetrically along the vertical axis and adjusting the coordinate position of the target bounding box synchronously.

[0034] The brightness and contrast are adjusted by adjusting the brightness gain and contrast range of image pixels to simulate different lighting conditions, so as to enhance the adaptability of the model to complex lighting environments.

[0035] The step S2 is specifically:

[0036] Step S2-1: Calculate the center point coordinates according to the upper left corner coordinates and width and height of the bounding box, and convert them into relative coordinate form;

[0037] Step S2-2: The width and height are divided by the overall width and height of the image respectively to achieve normalization processing, thereby obtaining the relative coordinates of the center point and the normalized width and height;

[0038] Step S2-3: Through the conversion method of step S2-1, the absolute coordinate information in COCO format is efficiently mapped to the relative annotation coordinates in YOLO format to ensure the consistency and efficiency of the detection model when images of different resolutions are input.

[0039] In step S3, the training parameters are set as follows:

[0040] Step S3-1: The training process is performed for M rounds in total, with N images of 640×640 size processed in each batch, using the GSD optimizer, and setting the warm-up phase to 3 rounds to stabilize the initial training process;

[0041] Among them, the initial learning rate of the GSD optimizer is set to 0.01, the momentum is 0.937, and the weight decay is 0.0005;

[0042] Step S3-2: The weight distribution of the loss function is: bounding box loss 7.5, classification loss 0.5, distribution focus loss 1.5, to balance the impact of each loss item;

[0043] Step S3-3: The IoU threshold in the verification phase is 0.7, and the maximum number of detections is 300 targets;

[0044] Step S3-4: Mosaic data enhancement, random flipping and color transformation are used for data enhancement during training to increase data diversity and improve the generalization ability of the model;

[0045] Among them, the horizontal flip probability is 50%, the hue is adjusted to 0.015, the saturation is adjusted to 0.7, and the brightness is adjusted to 0.4;

[0046] Step S3-5: The training uses an early stopping mechanism, which terminates the training early when there is no significant performance improvement within T rounds. Mixed precision training is also enabled to prevent overfitting, reduce video memory usage, and improve efficiency.

[0047] The present invention has the following beneficial effects and advantages:

[0048] The present invention proposes a transmission line equipment detection method based on a lossless data enhancement strategy, which optimizes problems such as target overlap, scale difference, small target detection and insufficient data, improves the detection performance of the model and its ability to adapt to complex environments. Compared with other methods, the present invention has higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flow chart for detecting the power transmission equipment of the present invention;

[0050] Figure 2 It is a schematic diagram of the principle of the ACmix model of the present invention;

[0051] Figure 3 Schematic diagram of the principle of GELAN designed for the present invention;

[0052] Figure 4 Schematic diagram of the principle of the adaptive spatial feature fusion module of the present invention. DETAILED DESCRIPTION

[0053] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the invention. Therefore, the present invention is not limited to the specific implementation disclosed below.

[0055] like Figure 1 As shown, it is a flow chart of the power transmission equipment detection of the present invention. In order to achieve the above-mentioned purpose, the present invention provides a power transmission line equipment detection method based on a lossless data enhancement strategy, comprising the following steps:

[0056] Step S1: establishing an image detection model for detecting multiple types of transmission line equipment according to different image types of transmission line equipment in the original image set in the database;

[0057] Step S2: pre-process the image of the power transmission line equipment to be detected, and convert the target bounding box annotation in COCO format into YOLO format to meet the requirements of the target detection task;

[0058] Step S3: setting training parameters, training the established image detection model, and obtaining an improved YOLOv10 model;

[0059] Step S4: input the preprocessed transmission line equipment image into the improved YOLOv10 model to obtain a data-enhanced image, detect the data-enhanced image, obtain the type of the transmission line equipment, and complete the transmission line equipment detection.

[0060] like Figure 2 As shown, it is a schematic diagram of the principle of the ACmix model of the present invention. In view of the problem of target overlap, the ACmix model is introduced. By combining the convolution mechanism with the self-attention mechanism, the extraction process of the target feature is dynamically adjusted to ensure that the target position and shape can be accurately distinguished under complex backgrounds, thereby effectively improving the detection accuracy in overlapping scenes. In view of the problem of large differences in target size in multi-scale detection tasks, the GELAN module is designed. By separating the input features into sub-features of multiple different receptive fields, and combining lightweight design and programmable gradient information (PGI) mechanism, the robustness of the model in multi-scale target detection is enhanced. In view of the difficulty in detecting small targets, the adaptive spatial feature fusion (ASFF) module is adopted. By extracting key information from feature maps of different resolutions and performing dynamic weighted fusion, the complementarity of low-resolution and high-resolution features is optimized, and the detection consistency and accuracy of small targets in complex scenes are improved. In view of the problem of data scarcity, a lossless data enhancement strategy is proposed. Through methods such as geometric transformation (such as vertical flipping) and visual adjustment (such as brightness and contrast adjustment), diversified training data is generated while maintaining the key information of the original image, thereby significantly improving the generalization ability and anti-interference ability of the model.

[0061] Furthermore, the ACmix model combines the convolution mechanism with the self-attention mechanism to form an efficient feature processing module. The module is divided into two main stages: the first stage performs a projection operation on the input feature map and generates multiple sets of intermediate feature maps through lightweight convolution, which provides a basis for subsequent feature aggregation. The second stage processes the intermediate feature maps along the convolution path and the self-attention path respectively. In the self-attention path, the feature map is decomposed into three parts: query, key, and value. The attention weight is generated by calculating the similarity between the query and the key, and the value is weighted summed using the weight to dynamically capture the global relationship between the objects. In the convolution path, the intermediate feature map further enhances the local features through lightweight convolution operations, and at the same time, combined with a specific position adjustment strategy, the accurate expression of the detailed information is retained. Finally, the ACmix model fuses the outputs of the self-attention path and the convolution path, combining the advantages of the two mechanisms. On the one hand, the self-attention mechanism can effectively deal with the occlusion problem between objects and enhance the model's feature extraction ability in complex backgrounds; on the other hand, the convolution mechanism further improves the fine-grained expression ability of features by sensitively capturing local information.

[0062] like Figure 3 As shown, it is a schematic diagram of the principle of GELAN designed by the present invention. Furthermore, the GELAN module realizes efficient feature expression and fusion through a series of innovative designs. First, the input feature map undergoes an initial convolution operation to generate an intermediate feature map with a fixed number of channels. Subsequently, these intermediate feature maps are divided into two parts, each with an equal number of channels, to achieve division of labor for different channels. Each part of the feature map is processed by the RepNCSP module, wherein the RepNCSP module combines convolution kernels of different sizes (such as 3×3 convolution and 1×1 convolution) to capture features within different receptive fields, thereby enhancing the fine-grained expression capability of the features. At the same time, these operations effectively balance the feature extraction efficiency of global information and local information. Next, multiple feature blocks are re-fused through a splicing operation to form a comprehensive feature map. This fusion process improves the adaptability to multi-scale targets by merging the features of different receptive fields into a unified feature representation, especially in scenarios where the target size differences are significant. Finally, the fused feature map generates the final output through a convolution operation, providing high-quality feature expression for subsequent detection tasks.

[0063] like Figure 4As shown, it is a schematic diagram of the principle of the adaptive spatial feature fusion module of the present invention. Furthermore, the ASFF module performs weighted adjustment on each layer of feature maps by assigning adaptive weights to feature maps from different levels of the feature pyramid. Specifically, the ASFF module extracts key information from feature maps of multiple resolutions and adjusts their weights layer by layer so that the feature contribution of each scale can be dynamically allocated according to the needs of the detection task. In this process, the feature map is combined with the feature maps of each layer through element-level operations to ensure the spatial consistency of the features and the integrity of the information transfer. Finally, after weighted adjustment and fusion operations, the ASFF module generates a unified feature map, which retains the useful information of the feature maps of each layer, while effectively solving the problems of feature loss and information redundancy.

[0064] Furthermore, the lossless data enhancement strategy includes two main operations: vertical flipping and brightness contrast adjustment. Among them, vertical flipping enhances the model's detection ability for symmetrical targets without changing the original pixel information by flipping the input image symmetrically along the vertical axis and adjusting the coordinate position of the target bounding box synchronously; brightness and contrast adjustment simulates different lighting conditions by adjusting the brightness gain and contrast range of image pixels to enhance the model's adaptability to complex lighting environments. All enhancement operations keep the bounding box and image updated synchronously to ensure the accuracy of the annotation information. Through the above strategy, diversified training samples can be generated without destroying the original image information, improving the robustness and detection performance of the model under data scarcity and complex background conditions.

[0065] Furthermore, the power transmission equipment detection method also includes the following steps: in the data preprocessing stage, the target bounding box annotation in COCO format is converted to YOLO format to meet the needs of the target detection task. The conversion process first calculates the center coordinates of each target bounding box, including the horizontal center and the vertical center, and normalizes the target width and height to ensure its proportional consistency relative to the image size. The specific steps include: calculating the center point coordinates according to the coordinates of the upper left corner of the bounding box and the width and height, and converting them into relative coordinate form; then, the width and height are divided by the overall width and height of the image, respectively, to achieve normalization, thereby obtaining the relative coordinates of the center point and the normalized width and height. Through the above conversion method, the absolute coordinate information in COCO format is efficiently mapped to the relative annotation coordinates in YOLO format, ensuring the consistency and efficiency of the detection model when images of different resolutions are input.

[0066] Furthermore, the power transmission equipment detection method also includes the following steps: setting the training parameters, the training process is carried out for a total of 150 rounds, each batch processes 18 images of 640×640 size, using the stochastic gradient descent (SGD) optimizer, setting the initial learning rate to 0.01, the momentum to 0.937, the weight decay to 0.0005, and setting the warm-up phase to 3 rounds to stabilize the initial training process. The weight distribution of the loss function is: bounding box loss 7.5, classification loss 0.5, distribution focus loss 1.5 to balance the influence of each loss item; the IoU threshold in the verification phase is 0.7, and the maximum number of detections is 300 targets. In addition, data enhancement strategies such as Mosaic data enhancement, random flipping (horizontal flipping probability 50%) and color transformation (hue adjustment 0.015, saturation adjustment 0.7, brightness adjustment 0.4) are used in the training process to increase data diversity and improve the generalization ability of the model. To prevent overfitting, the training uses an early stopping mechanism, which terminates the training early if there is no significant performance improvement within 100 rounds. Mixed precision training is also enabled to reduce video memory usage and improve efficiency.

[0067] Specifically, each method is described in detail, such as Figure 1 As shown in the figure, the processed data is first input into the improved YOLOv10 model, mainly improving the backbone feature extraction network, replacing the original C2f with C2f-ELAN4. For the Neck architecture, the ACmix attention mechanism is added to the output positions of the large, medium, and small object detection layers. In the Head, the original v10Detect detection head is replaced with the ASFFDetect detection head.

[0068] The following is an introduction to the various modules introduced in this application:

[0069] 1. ACmix principle

[0070] like Figure 2 As shown in Figure 1, ACmix forms a hybrid module by combining convolution and self-attention mechanisms. In the first stage, the input feature map is projected through three 1×1 convolutions to generate multiple intermediate feature maps. In the second stage, these intermediate feature maps follow two different aggregation methods: convolution path and self-attention path.

[0071] For the self-attention path, the generated intermediate feature maps are divided into multiple groups, each containing three feature maps, used as query, key, and value. Through the self-attention mechanism, the similarity between the query and the key is calculated, and the value is weighted and summed using the attention weight. This process can be expressed by the following formula 1:

[0072]

[0073] Among them, q ij and k ab are the feature representations of query and key respectively, A(q ij ,k ab ) represents the attention weight, v ab is the value.

[0074] For the convolution path, a lightweight fully connected layer is used to generate k 2 The feature map is shifted and summed using the following formula 2:

[0075]

[0076] in, It represents the feature map obtained after applying the convolution kernel, and Shift(g,p,q) is the shift operation.

[0077] Finally, the outputs from the convolution path and the self-attention path are weighted and summed to form the final output. As shown in Formula 3:

[0078] F out =αF att +βF conv (3)

[0079] Among them, α and β are learnable scalars that control the output strength of the self-attention path and the convolution path respectively.

[0080] 2.GELAN schematic diagram

[0081] The GELAN module improves feature extraction capabilities by combining multiple convolution operations. Figure 3 First, the input feature map X undergoes an initial convolution operation to generate an output feature map with C channels, as shown in Formula 4:

[0082] Y=Conv(X) (4)

[0083] in H and W are the height and width of the feature map, respectively, and C is the number of channels. Next, the generated feature map Y is divided into two equal parts, Y1 and Y2, each with channels, as shown in Formula 5:

[0084] Y1, Y2 = Split(Y)(5)

[0085] in The segmented feature map Y2 is then processed by the RepNCSP module. The core of RepNCSP is to combine 3×3 and 1×1 convolution operations, as shown in Formula 6:

[0086] Z i=RepNCSP(Y i )=Conv 3×3 (Y i )+Conv 1×1 (Y i ) (6)

[0087] Among them, the design of i=2, 3 allows the model to process features of different receptive fields at the same time, enhancing the feature expression ability of complex scenes.

[0088] Next, the two features Z2, Z3 and Y1 processed by RepNCSP are re-fused through the concatenation operation Concat to form a complete feature map Z, as shown in Formula 7:

[0089] Z=Concat(Y1,Z2,Z3) (7)

[0090] in Finally, the fused feature map Z is convolved again to generate the final output X out , as shown in Formula 8:

[0091] X out =Conv(Z) (8)

[0092] 3.ASFF principle

[0093] like Figure 4 As shown in Figure 1, the ASFF module displays feature maps from different levels (Level 1, Level 2, Level 3) of the feature pyramid, which have different resolutions and strides. The ASFF module uses adaptive weights (α 3 , β 3 , γ 3 ) weights each scale feature and combines it with the feature maps of each layer (X1→3, X2→3, X3→3) through element-by-element multiplication to ensure that the feature contribution of each layer can be effectively integrated. Especially for small target detection, low-level high-resolution features are crucial, and the ASFF module effectively alleviates the problem of information loss of small targets in higher-level feature maps by adaptively fusing features of different scales. Finally, the features of each scale are fused through addition operations to generate unified features for subsequent target prediction.

[0094] 4. Lossless Data Enhancement Method

[0095] First, vertical flipping is a geometric enhancement method that flips the image symmetrically along the vertical axis and simultaneously adjusts the y coordinate of the target bounding box to ensure the accuracy of the annotation information. As shown in Formula 9:

[0096] y new =Hyoriginal (9)

[0097] Among them, H represents the height of the image, y original is the original vertical coordinate of the center of the bounding box. The vertical flip operation does not change the pixel information of the image, but can generate different perspectives, which helps the model learn the ability to detect vertically symmetric objects, thereby improving the robustness of detection in vertically changing scenes.

[0098] Secondly, brightness and contrast adjustment is a visual enhancement operation used to simulate different lighting environments, thereby enhancing the adaptability of the model under light changes. As shown in Formula 10:

[0099] I new =α·(I-mean)+β (10)

[0100] Among them, I is the pixel value, α is the contrast adjustment factor, β is the brightness adjustment factor, and mean is the pixel mean of the image. By generating samples under different lighting conditions, the robustness of the model has been effectively improved, especially in application scenarios with large changes in lighting conditions.

[0101] In order to ensure the consistency of bounding box annotations during the enhancement process, this paper uses the bounding box parameter synchronization update function of the Albumentations library. For example, for the vertical flip operation, the y coordinate of the center of the bounding box is updated synchronously, as shown in Formula 11:

[0102] y center,new =1-y center,original (11)

[0103] 5. Dataset and Preprocessing

[0104] In this study, the bounding box coordinates in COCO format are converted to YOLO format to meet the needs of the target detection task. In the COCO format, the bounding box is represented by the upper left corner coordinates and width and height, while the YOLO format requires the relative coordinates of the center point of the target and the relative values ​​of width and height. To achieve this conversion, the convert function in the code first calculates the coordinates of the center point of the bounding box, and then normalizes the width and height to a ratio relative to the image size. In this way, the absolute pixel coordinates in the COCO format are converted to relative coordinates in the YOLO format, allowing the model to maintain detection accuracy on images of different resolutions. This process is automated in the code to ensure the consistency and efficiency of the input data. The center point coordinates are shown in Formula 12, the normalized width and height are shown in Formula 13, and the normalized center point coordinates are shown in Formula 14:

[0105]

[0106] 6. Training parameter settings

[0107] The training parameter settings are shown in Table 1, which are the training process parameter settings of the present invention;

[0108] Table 1. Training process parameter settings

[0109]

[0110]

[0111] The training process was carried out for a total of 150 rounds, with 18 images of 640×640 size processed in each batch. The stochastic gradient descent (SGD) optimizer was used, with the initial learning rate set to 0.01, the momentum to 0.937, the weight decay to 0.0005, and the warm-up phase set to 3 rounds to stabilize the initial training process. The weight distribution of the loss function was: bounding box loss 7.5, classification loss 0.5, distribution focus loss 1.5 to balance the influence of each loss term; the IoU threshold in the validation phase was 0.7, and the maximum number of detections was 300 targets. In addition, data enhancement strategies such as Mosaic data enhancement, random flipping (horizontal flipping probability 50%) and color transformation (hue adjustment 0.015, saturation adjustment 0.7, brightness adjustment 0.4) were used in the training process to increase data diversity and improve the generalization ability of the model. To prevent overfitting, the training enabled the early stopping mechanism, which terminated early when the performance did not improve significantly within 100 rounds, and enabled mixed precision training to reduce video memory usage and improve efficiency.

[0112] In summary, the present invention proposes an improved YOLOv10 model to solve the problems of target overlap, large size difference, difficulty in detecting small targets and data scarcity in the process of power transmission equipment detection. To solve the problem of target overlap, the present invention introduces the Acmix model, integrates self-attention and convolution mechanisms, realizes dynamic feature extraction, and significantly improves the detection performance in overlapping scenarios; to solve the problem of large size difference in multi-scale detection, the GELAN structure is designed, and the detection robustness of the model is enhanced by combining lightweight design and programmable gradient information (PGI) mechanism; to improve the accuracy and consistency of small target detection, an adaptive spatial feature fusion module (ASFF) is adopted; in addition, to solve the problem of data scarcity, the present invention proposes a lossless data enhancement strategy, which increases data diversity through methods such as vertical flipping, brightness and contrast adjustment, and significantly improves the generalization ability of the model. Experimental results show that the method of the present invention has excellent detection performance in complex power transmission environments, can effectively improve the monitoring efficiency of power transmission equipment, and provides an efficient solution for ensuring the reliability of power supply.

[0113] Those skilled in the art will appreciate that the above are only preferred embodiments of the present invention, and the various embodiments of the present disclosure and / or the features described in the claims may be combined or combined in various ways, even if such combinations or combinations are not explicitly described in the present disclosure. It is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art may still modify the technical solutions described in the aforementioned embodiments, or perform equivalent substitutions on some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

[0114] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A power transmission line equipment detection method based on lossless data enhancement strategy, characterized in that: The following steps are involved: Step S1: establishing an image detection model for detecting multiple types of transmission line equipment according to different image types of transmission line equipment in the original image set in the database; Step S2: pre-process the image of the power transmission line equipment to be detected, and convert the target bounding box annotation in COCO format into YOLO format to meet the requirements of the target detection task; Step S3: setting training parameters, training the established image detection model, and obtaining an improved YOLOv10 model; Step S4: input the preprocessed transmission line equipment image into the improved YOLOv10 model to obtain a data-enhanced image, detect the data-enhanced image, obtain the type of the transmission line equipment, and complete the transmission line equipment detection.

2. A power transmission line equipment detection method based on lossless data enhancement strategy according to claim 1, characterized in that: The image types include: target overlap, large target size differences in multi-scale detection tasks, difficulty in detecting small targets, and data scarcity.

3. A power transmission line equipment detection method based on lossless data enhancement strategy according to claim 1, characterized in that: In step S1, the image detection model for detecting various types of power transmission line equipment is established, specifically: 1) To address target overlap, the ACmix model is introduced into the image detection model. By combining the convolution mechanism with the self-attention mechanism, a feature processing module is formed to dynamically adjust the target feature extraction process to improve the detection accuracy in overlapping scenarios; 2) In view of the large differences in target sizes in multi-scale detection tasks, the GELAN module is designed to enhance the robustness of the model in multi-scale target detection by separating the input features into multiple sub-features with different receptive fields and combining them with the PGI mechanism; 3) To address the difficulty in detecting small targets, an adaptive spatial feature fusion module is used to extract key information from feature maps of different resolutions and perform dynamic weighted fusion to optimize the complementarity of low-resolution and high-resolution features, thereby improving the consistency and accuracy of small target detection in complex scenes. 4) To address the problem of data scarcity, a lossless data augmentation strategy is used to generate diversified training data while maintaining the key information of the original image through geometric transformation and visual adjustment methods, thereby significantly improving the generalization and anti-interference capabilities of the model.

4. A power transmission line equipment detection method based on lossless data enhancement strategy according to claim 3, characterized in that: In step 1), the feature processing module performs the following steps: 1-1) Perform a projection operation on the input feature map and generate multiple sets of intermediate feature maps through lightweight convolution, which provide a basis for subsequent feature aggregation; 1-2) Process the intermediate feature map along the convolution path and self-attention path respectively; 1-2-1) In the self-attention path, the feature graph is decomposed into query, key and value. The attention weight is generated by calculating the similarity between the query and the key. The weight is used to perform weighted summation on the value to obtain the global relationship between the dynamic capture targets. 1-2-2) In the convolution path, the intermediate feature map further enhances the local features through lightweight convolution operations, and combines specific position adjustment strategies to retain the accurate expression of detail information; 1-3) The ACmix model fuses the output of the self-attention path and the convolution path to obtain a feature map.

5. The method for detecting power transmission line equipment based on lossless data enhancement strategy according to claim 3, characterized in that: The step 2) is specifically: 2-1) First, the input feature map undergoes an initial convolution operation to generate an intermediate feature map with a fixed number of channels; 2-2) The intermediate feature map is divided into two parts, each with an equal number of channels to achieve division of labor for different channels; each part of the feature map is processed by the RepNCSP module respectively; the RepNCSP module combines convolution kernels of different sizes to capture features within different receptive fields and enhance the fine-grained expression capability of features; 2-3) Re-integrate multiple feature blocks through splicing operations to form a comprehensive feature map; 2-4) The fused feature map is subjected to convolution operation to generate a new feature map, providing high-quality feature expression for subsequent detection tasks.

6. A power transmission line equipment detection method based on lossless data enhancement strategy according to claim 3, characterized in that: The step 3) is specifically: 3-1) The adaptive spatial feature fusion module extracts key information from feature maps of multiple resolutions; 3-2) The adaptive spatial feature fusion module adjusts the weight of each type of feature map layer by layer, so that the feature contribution of each scale can be dynamically allocated according to the requirements of the detection task; 3-3) Through weighted adjustment and fusion operations, the adaptive spatial feature fusion module generates a unified feature map.

7. A power transmission line equipment detection method based on lossless data enhancement strategy according to claim 3, characterized in that: In step 3), the lossless data enhancement strategy includes: vertical flipping and brightness contrast adjustment.

8. A power transmission line equipment detection method based on lossless data enhancement strategy according to claim 7, characterized in that: The vertical flipping is to enhance the detection capability of the model for symmetrical targets without changing the original pixel information by flipping the input image symmetrically along the vertical axis and adjusting the coordinate position of the target bounding box synchronously. The brightness and contrast are adjusted by adjusting the brightness gain and contrast range of image pixels to simulate different lighting conditions, so as to enhance the adaptability of the model to complex lighting environments.

9. The method for detecting power transmission line equipment based on lossless data enhancement strategy according to claim 1, characterized in that: The step S2 is specifically: Step S2-1: Calculate the center point coordinates according to the upper left corner coordinates and width and height of the bounding box, and convert them into relative coordinate form; Step S2-2: The width and height are divided by the overall width and height of the image respectively to achieve normalization processing, thereby obtaining the relative coordinates of the center point and the normalized width and height; Step S2-3: Through the conversion method of step S2-1, the absolute coordinate information in COCO format is efficiently mapped to the relative annotation coordinates in YOLO format to ensure the consistency and efficiency of the detection model when images of different resolutions are input.

10. The method for detecting power transmission line equipment based on lossless data enhancement strategy according to claim 1, characterized in that: In step S3, the training parameters are set as follows: Step S3-1: The training process is performed for M rounds in total, with N images of 640×640 size processed in each batch, using the GSD optimizer, and setting the warm-up phase to 3 rounds to stabilize the initial training process; Among them, the initial learning rate of the GSD optimizer is set to 0.01, the momentum is 0.937, and the weight decay is 0.0005; Step S3-2: The weight distribution of the loss function is: bounding box loss 7.5, classification loss 0.5, distribution focus loss 1.5, to balance the impact of each loss item; Step S3-3: The IoU threshold in the verification phase is 0.7, and the maximum number of detections is 300 targets; Step S3-4: Mosaic data enhancement, random flipping and color transformation are used for data enhancement during training to increase data diversity and improve the generalization ability of the model; Among them, the horizontal flip probability is 50%, the hue is adjusted to 0.015, the saturation is adjusted to 0.7, and the brightness is adjusted to 0.4; Step S3-5: The training uses an early stopping mechanism, which terminates the training early when there is no significant performance improvement within T rounds. Mixed precision training is also enabled to prevent overfitting, reduce video memory usage, and improve efficiency.

Images