Electric power high-altitude operation safety belt wearing identification method based on YOLOv10n and product

Abstract

The invention provides an electric power high-altitude operation safety belt wearing identification method based on YOLOv10n and a product. The method comprises the steps that an actual electric power high-altitude operation scene image is acquired; based on the trained safety belt wearing recognition model, feature extraction and recognition are carried out on the actual electric power high-altitude operation scene image, and a safety belt wearing recognition result of the operator is obtained; wherein the safety belt wearing identification model is constructed and obtained by introducing a StarNet module, a bidirectional feature pyramid BiFPN module and an LSCD detection head based on a YOLOv10n model. According to the invention, the StarNet module is introduced into the head network, the BiFPN module is introduced into the neck network and is combined with the LSCD detection head, and the constructed safety belt wearing recognition model is more targeted for the wearing mode, and is more accurate and lighter than the original target detection network in detection result.

Description

Technical Field

[0001] The present application relates to the fields of computer vision technology and power safety, and in particular to a method and product for identifying the wearing of a safety belt for high-altitude power operations based on YOLOv10n. Background Art

[0002] In power construction operations, the safety of workers working at heights has always been an important issue that needs urgent attention. According to relevant statistics, failure to wear a safety belt is one of the important factors leading to power construction accidents. Such accidents not only cause casualties, but also bring huge economic losses to enterprises. Studies have shown that accidents caused by failure to wear a safety belt account for a considerable proportion of the total number of power construction accidents, further confirming the necessity of strengthening safety management in high-risk environments. As a basic personal protective equipment, the correct wearing of a safety belt can significantly reduce the risk of falling during high-altitude operations. Although relevant laws and safety standards have clearly required workers to wear safety belts when working at heights, in the actual construction process, workers fail to comply with this regulation for various reasons, resulting in an increase in safety hazards. Therefore, how to effectively monitor and ensure that workers wear safety belts correctly has become one of the key measures to improve the level of construction safety management.

[0003] Safety belt wearing detection is an important part of ensuring the safety of workers in power construction operations. Traditionally, this detection mainly relies on manual monitoring and observation. Although this method is highly flexible and can detect problems in a timely manner, its strong subjectivity and low efficiency make it difficult to fully cover large construction sites and there is a risk of missed detection. With the development of technology, traditional image processing technology has gradually been applied to safety belt wearing detection. Real-time monitoring through cameras and algorithms can reduce the interference of human factors and improve the objectivity of detection. However, existing models often face large volume and computing resource consumption, which limits their promotion in practical applications, especially in resource-constrained construction sites. In addition, challenges such as the adaptability of image processing technology in complex environments, data privacy issues, and high equipment costs also need to be addressed. Therefore, future research needs to focus on how to optimize the model to reduce the consumption of computing resources, and combine artificial intelligence and Internet of Things technologies to achieve more accurate and intelligent safety belt detection, thereby further improving the safety management level of power construction operations.

[0004] In order to address the above challenges, this paper focuses on developing a seat belt object detection model based on YOLOv10. This model aims to provide an innovative and efficient solution for seat belt wearing detection at power construction sites by combining lightweight network architecture, scenario-specific optimization strategies, and efficient training methods. This solution can not only meet practical application needs, but also promote technological progress in the safety and reliability of power systems. Summary of the invention

[0005] The purpose of this application is to provide an ultra-lightweight electric high-altitude work safety belt wearing identification method and product based on YOLOv10n, which can at least solve the problems of insufficient detection model accuracy and high resource consumption in related technologies.

[0006] In order to solve the above technical problems, the first aspect of the embodiment of the present application provides a method for identifying the wearing of a safety belt for electric power aerial work based on YOLOv10n, comprising:

[0007] Acquire images of actual power aerial work scenes;

[0008] Based on the trained seat belt wearing recognition model, feature extraction and recognition are performed on the actual power high-altitude operation scene image to obtain the seat belt wearing recognition result of the operator; wherein, the seat belt wearing recognition model is constructed based on the YOLOv10n model by introducing the StarNet module, the bidirectional feature pyramid BiFPN module and the LSCD detection head.

[0009] The second aspect of the embodiment of the present application provides a power high-altitude work safety belt wearing recognition system based on YOLOv10n, including:

[0010] An acquisition module is used to acquire images of actual power aerial work scenes;

[0011] The recognition module is used to extract and recognize features of the actual power high-altitude operation scene image based on the trained seat belt wearing recognition model to obtain the seat belt wearing recognition results of the operators; wherein the seat belt wearing recognition model is constructed based on the YOLOv10n model by introducing the StarNet module, the bidirectional feature pyramid BiFPN module and the LSCD detection head.

[0012] The third aspect of the present application provides an electronic device, comprising: a memory and a processor, wherein the processor is used to execute a computer program stored in the memory, and when the processor executes the computer program, it implements each step of the method for identifying the wearing of a safety belt for high-altitude power work described in the first aspect of the embodiment of the present application.

[0013] The fourth aspect of the present application provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for identifying the wearing of a safety belt for high-altitude power work described in the first aspect of the embodiment of the present application are implemented.

[0014] It can be seen from the above that the embodiment of the present application first obtains an actual image of an electric power high-altitude operation scene, and then extracts and identifies features of the actual electric power high-altitude operation scene image based on the trained seat belt wearing recognition model to obtain the seat belt wearing recognition result of the operator, wherein the seat belt wearing recognition model is constructed based on the YOLOv10n model by introducing the StarNet module, the bidirectional feature pyramid BiFPN module and the LSCD detection head. The seat belt wearing recognition model constructed based on the YOLOv10n model by introducing the StarNet module, the bidirectional feature pyramid BiFPN module and the LSCD detection head is more targeted at the wearing method, and is more accurate and lighter than the detection result of the original target detection network.

[0015] It should be understood that the content described in this section is not intended to identify the key or important features of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the related technologies or the embodiments of the present application, the drawings required for use in the description of the related technologies or the embodiments of the present application will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, not all embodiments. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A flowchart of a method for identifying the wearing of a safety belt for electric power aerial work based on YOLOv10n is provided in the first aspect of an embodiment of the present application;

[0018] Figure 2 A main architecture diagram of a method for identifying the wearing of a safety belt for electric power aerial work based on YOLOv10n provided in the first aspect of an embodiment of the present application;

[0019] Figure 3 A star-operation phased architecture diagram in a StarNet module in a method for identifying the wearing of a safety belt for electric power aerial work based on YOLOv10n provided in the first aspect of an embodiment of the present application;

[0020] Figure 4 A structural diagram of LSCD in a method for identifying the wearing of a safety belt for electric power aerial work based on YOLOv10n provided in the first aspect of an embodiment of the present application;

[0021] Figure 5An example diagram of the recognition effect of a method for identifying the wearing of a safety belt for electric power aerial work based on YOLOv10n provided in the first aspect of an embodiment of the present application;

[0022] Figure 6 A detailed flow chart of a method for identifying the wearing of a safety belt for power aerial work based on YOLOv10n provided in the first aspect of an embodiment of the present application;

[0023] Figure 7 A schematic diagram of a program module of a device for identifying the wearing of a safety belt for electric power aerial work provided in the second aspect of an embodiment of the present application;

[0024] Figure 8 A module block diagram of an electronic device provided in the third aspect of an embodiment of the present application;

[0025] Fig. 9 A module block diagram of a computer-readable storage medium provided in the fourth aspect of an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the present application more obvious and easy to understand, the present application will be clearly and completely described below in conjunction with the embodiments of the present application and the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. It should be understood that the various embodiments of the present application described below are only used to explain the present application and are not used to limit the present application, that is, based on the various embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present application. In addition, the technical features involved in the various embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0027] See also Figure 1 , Figure 1 A flow chart of a method for identifying the wearing of a safety belt for high-altitude power work based on YOLOv10n is provided in the first aspect of an embodiment of the present application. The method for identifying the wearing of a safety belt for high-altitude power work includes the following steps.

[0028] Step 101: Acquire an image of an actual power aerial work scene.

[0029] Step 102: Based on the trained seat belt wearing recognition model, feature extraction and recognition are performed on the actual power high-altitude operation scene image to obtain the seat belt wearing recognition result of the operator; wherein the seat belt wearing recognition model is constructed based on the YOLOv10n model by introducing the StarNet module, the bidirectional feature pyramid BiFPN module and the LSCD detection head.

[0030] In an embodiment of the present application, a specially tuned StarNet module is used as a feature extractor of the model to extract input image features, and then the improved Neck part of the BiFPN module is introduced to perform feature fusion to obtain a fused multi-scale feature map. Finally, the multi-scale feature map is processed by an innovative LSCD detection head to obtain the final type of safety belts worn by high-altitude workers. The present invention combines the improved StarNet with the Neck part and the innovative lightweight detection head LSCD to construct a target detection model that is more targeted to the wearing method, and is more accurate and lighter than the original target detection network detection results. The specific construction of the model will be described in detail in subsequent steps.

[0031] In an optional embodiment of this embodiment, before extracting and identifying features of an actual electric power aerial work scene image based on the trained safety belt wearing recognition model, the method further includes:

[0032] Obtaining a training sample set and a test sample set; wherein the training sample set and the test sample set include a plurality of electric power aerial work scene image samples;

[0033] Randomly select power aerial work scene image samples from the training sample set and input them into the original safety belt wearing recognition model for target recognition to obtain training results;

[0034] Based on the training results and the preset loss function, the network parameters in the original seat belt wearing recognition model are adjusted to obtain an adjusted seat belt wearing recognition model; wherein the preset loss function is a multiple loss function including category loss, bounding box loss and distributed regression loss;

[0035] Randomly select power aerial work scene image samples from the test sample set and input them into the adjusted seat belt wearing recognition model for target recognition to obtain the test results;

[0036] If the test passes, the adjusted seat belt wearing recognition model is used as the trained seat belt wearing recognition model; if the test fails, the adjusted seat belt wearing recognition model continues to be trained.

[0037] In the embodiment of the present application, the image acquisition system is used to make the images of high-altitude workers wearing safety belts obtained in cooperation with the power system company into a model-specific data set, and 1800 images with a resolution of 1936*1296 are collected, and the ratio of the number of images in the training set to the number of images in the test set is 4:1. It is understandable that the images in the data set need to be preprocessed before training, including recording the location of the wearer in the data set picture and the type of wear, and then comparing with the output result to calculate the accuracy of target detection. Table 1 shows the number of various types of wear in the data set.

[0038] Table 1

[0039]

[0040] The data in the training set is input into the model in batches for training. The training data is loaded from dataset / data.yaml. The total number of training rounds is 300, and each batch trains 32 pictures with an image size of 640*640. The SGD optimizer is used in training, with an initial learning rate of 0.01, and is gradually adjusted after warmup_epochs=3. The model uses multiple loss functions including category loss (cls=0.5), bounding box loss (box=7.5) and distributed regression loss (dfl=1.5) to optimize the model's performance in classification and positioning.

[0041] In an optional embodiment of this embodiment, based on the trained seat belt wearing recognition model, feature extraction and recognition are performed on the actual power high-altitude working scene image to obtain the seat belt wearing recognition result of the operator, including:

[0042] The actual power aerial work scene image is input into the backbone network of the safety belt wearing recognition model for feature extraction to obtain the initial multi-scale feature map;

[0043] The initial multi-scale feature map is input into the neck network of the seat belt wearing recognition model for feature fusion to obtain a fused multi-scale feature map;

[0044] The fused multi-scale feature map is input into the head network of the seat belt wearing recognition model for target recognition, and the seat belt wearing recognition result of the operator is obtained.

[0045] Specifically, first, the actual image of the power high-altitude operation scene is input into the backbone network in the safety belt wearing recognition model for feature extraction to obtain an initial multi-scale feature map, wherein the backbone network is an improved backbone network after the introduction of the StarNet module, and the specific model will be described in subsequent embodiments. Then, the initial multi-scale feature map is input into the improved neck network after the introduction of the BiFPN module for feature fusion to obtain a fused multi-scale feature map. Finally, the fused multi-scale feature map is input into the head network after the introduction of the LSCD detection head for target recognition. The LSCD detection head in the model performs the final regression of the seat belt wearer's position and classification of the wearing type, and finally obtains the seat belt wearing recognition result of the operator.

[0046] In an optional embodiment of this embodiment, the backbone network in the seat belt wearing recognition model adopts a StarNet module, and the StarNet module includes:

[0047] The first operation module is configured to include an initialized convolutional layer and increase the number of channels of the input image from 3 to 16;

[0048] A second operation module, configured to include a first-stage star operation of a block, each block including a depthwise separable convolutional layer and two fully connected layers;

[0049] The third operation module is configured as a second-stage star operation including one block, each block includes a depthwise separable convolutional layer and two fully connected layers, and the number of channels is increased to 32;

[0050] The fourth operation module is configured as a third-stage star operation containing three blocks, each block includes a depthwise separable convolutional layer and two fully connected layers, and the number of channels is increased to 64;

[0051] a fifth operation module, configured as a fourth-stage star operation including one block, each block including a depthwise separable convolutional layer and two fully connected layers, and increasing the number of channels to 128;

[0052] A sixth operation module, configured as a SPPF module in YOLOv8;

[0053] The seventh operation module is configured as a PSA module in YOLOv10.

[0054] In an optional embodiment of this embodiment, each stage of the star operation in the StarNet module consists of a convolution downsampling module and stacked star blocks, and the formula for each star block to complete the nonlinear high-dimensional feature mapping is expressed as:

[0055]

[0056] in and are weight matrices of two sets of linear transformations, ReLU6 is a nonlinear activation function, ⊙ represents element-by-element product, and the star operation maps the input features to an implicit high-dimensional nonlinear space, ultimately generating a dim dimension. output It is expressed as:

[0057]

[0058] Where d is the number of input channels.

[0059] Specifically, Figure 2The main architecture diagram of a method for identifying the wearing of a safety belt for electric aerial work based on YOLOv10n provided in the first aspect of the embodiment of the present application. The StarNet feature extraction module is mainly composed of a convolution layer and four star-operation operations and the original SPPF and PSA modules of yolov10. The number of channels of the input image is increased from 3 (RGB) to 16 by initializing the convolution layer (ConvBN), providing more feature space for subsequent feature extraction. The purpose is to use Batch Normalization (BN) to stabilize the training process and accelerate convergence.

[0060] In the first stage, Star-Operation extracts preliminary features through depthwise separable convolution and fully connected layers. Depthwise separable convolution (ConvBN, 7x7, stride=1, groups=16) is applied to extract spatial features. Features are further processed through two fully connected layers (linear transformations W1 and W2). Element-wise multiplication operation (star operation) is performed between the two fully connected layers to enhance the nonlinear expression ability of features.

[0061] in, Figure 3 The star-operation phased architecture diagram in the StarNet module in the YOLOv10n-based power aerial work safety belt wearing identification method provided in the first aspect of the embodiment of the present application, each stage consists of a convolutional downsampling module and a number of stacked star blocks (Star Blocks). The depth of each stage of the main part of StarNet is [1,1,3,1], and the number of channels gradually doubles as the stage deepens. Each star block is the core unit of feature extraction, which completes the nonlinear high-dimensional feature mapping through the following formula:

[0062]

[0063] in and are weight matrices of two sets of linear transformations, ReLU6 is a nonlinear activation function, and ⊙ represents element-by-element product (i.e., star operation). The star operation maps the input features to an implicit high-dimensional nonlinear space, enhancing the feature expression capability, and the final generated dimension is:

[0064]

[0065] Where d is the number of input channels.

[0066] The second stage Star-Operation further extracts features based on the first stage and doubles the number of channels to 32. Repeat the operations of depth-wise separable convolution and fully connected layers, but double the number of channels in this stage.

[0067] The third stage Star-Operation further enhances the feature extraction capability and increases the number of channels to 64. This stage contains 3 blocks, each of which repeats the operations of depthwise separable convolution and fully connected layers. Through these blocks, the feature expression capability is gradually enhanced.

[0068] The fourth stage Star-Operation further improves the depth of feature extraction based on the third stage and increases the number of channels to 128. One Block is applied to perform depth-separable convolution and fully connected layer operations. Through this stage of operation, deep features are further extracted.

[0069] The SPPF module (Spatial Pyramid Pooling-Fast) enhances the network's detection capability for objects of different sizes by fusing multi-scale features. It performs pooling operations on feature maps at different scales to generate multi-scale feature representations.

[0070] The formula can be expressed as:

[0071]

[0072] Among them, k represents the pooling window size.

[0073] The PSA module (Partial Self-Attention) improves the feature representation capability through the partial self-attention mechanism while reducing the computational cost. The application of the partial self-attention mechanism enables the network to focus on the more important parts of the image while reducing the computational complexity.

[0074] Among them, the optimized feature expression of some self-attention mechanisms is:

[0075] x psa =Softmax(Q·K T )·V

[0076] Q, K, and V represent the query, key, and value matrices generated from the input features, respectively.

[0077] Finally, after being processed by the StarNet module, a multi-scale feature map is generated:

[0078]

[0079] In an optional embodiment of this embodiment, the multi-scale feature map is input into the neck network of the seat belt wearing recognition model for feature fusion to obtain a fused multi-scale feature map, including:

[0080] The initial multi-scale feature map is sequentially transferred to the BiFPN module in the neck network for bidirectional fusion to obtain a preliminary fused multi-scale feature map;

[0081] The preliminary fused multi-scale feature map is input into the C2F module in the neck network for feature enhancement to obtain the fused multi-scale feature map.

[0082] Specifically, the Neck part receives the multi-scale feature maps P2, P3, P4, and P5 output from the StarNet module. These feature maps will be passed to the BiFPN module in turn for bidirectional fusion. The BiFPN module completes the top-down and bottom-up feature fusion in the bidirectional path in turn. The top-down fusion path gradually transfers information upward from low-resolution feature maps (such as P5) to high-resolution feature maps (such as P2). Each high-resolution feature is calculated by fusion of the current layer and the next layer:

[0083]

[0084] in It is represented as the output feature of the i-th layer in the top-down path, and Upsample represents the low-resolution feature map Perform bilinear interpolation upsampling, Fusion represents dynamic weighted feature fusion, and the calculation is as follows:

[0085]

[0086] Where ω1, ω2 are the learned weight parameters.

[0087] The fused features are further enhanced by the C2F module. The core operation of C2F is channel division and splicing, which can be expressed as:

[0088]

[0089] Among them, y 1 ,y 2 is the output of the initial convolution, y bottleNeck is the feature extracted by recursive BottleNeck.

[0090] The bottom-up fusion path gradually transfers information downward from high-resolution feature maps (such as P2) to low-resolution feature maps (such as P5). Each low-resolution feature is calculated by fusion of the current layer and the previous layer:

[0091]

[0092] in It is represented as the output feature of the i-th layer in the bottom-up path, and Downsample represents the high-resolution feature map Perform 3×3 convolution downsampling, and further optimize the fused features through the C2F module. During the feature fusion process, BiFPN uses a dynamic weighting mechanism to learn the weights of different input features. The formula is as follows:

[0093]

[0094] where ω i Represents the weight of each input feature, ∈ takes the value 1e-4 to avoid numerical instability.

[0095] After being processed by the bottom-up and bottom-up paths and the C2F module, the final multi-scale feature map generated by the Neck part is represented as:

[0096]

[0097] In an optional embodiment of this embodiment, the fused multi-scale feature map is input into the head network of the seat belt wearing recognition model for target recognition, and the seat belt wearing recognition result of the operator is obtained, including:

[0098] The fused multi-scale feature map is input into the shared convolution module in the head network for feature extraction to obtain the feature extraction result;

[0099] The feature extraction results are passed to the two parallel bounding box regression branches and classification prediction branches in the head network for target recognition, and the seat belt wearing recognition results of the operator are obtained.

[0100] Specifically, Figure 4 A structural diagram of LSCD in a method for identifying the wearing of a safety belt for high-altitude power work based on YOLOv10n provided in the first aspect of an embodiment of the present application. The LSCD (Lightweight Shared Convolutional Detection) detection head adopts a lightweight shared convolution module, combined with a dual-branch design of bounding box regression and target classification, to complete the target detection task. The detection head receives the multi-scale feature map from the above step three. Each input feature map is first subjected to feature extraction by a shared convolution module, and the output features are passed to two parallel bounding box regression branches and classification prediction branches. After the detection head performs bounding box and classification prediction on the features of P2′, P3′, P4′, and P5′ respectively, it outputs multiple levels of detection results in a multi-scale fusion manner.

[0101] During the inference process, the detection head dynamically generates anchors based on the input features to guide the prediction and decoding of bounding boxes. The anchor size is dynamically adjusted according to the resolution and step size of each layer of feature maps. The predicted bounding box parameters are decoded to generate the actual bounding box coordinates, and the classification scores are normalized to represent the confidence of the target category. The detection results are filtered by non-maximum suppression (NMS) to remove low-confidence and overlapping bounding boxes, and finally output the detection results containing the target location, category and confidence.

[0102] in, Figure 5 This is an example diagram of the recognition effect of a method for identifying the wearing of a safety belt for high-altitude power work based on YOLOv10n provided in the first aspect of an embodiment of the present application.

[0103] The test results of the present invention are compared with deep learning models such as yolov8 and yolov10. The test results are shown in Table 2. It can be seen that the parameters and model size are highly lightweight. In Table 2, the mAP of this method is improved by 0.02 compared with the original yolov10. It has a smaller model size and higher detection accuracy. Experiments show that the model of the present invention is more suitable for the identification of wearing safety belts for high-altitude work in power. The evaluation index used in the present invention is the target detection evaluation standard mAP50-95, and the calculation formula is:

[0104]

[0105] Among them, mAP50-95 represents the average AP under multiple IoU thresholds (from 0.50 to 0.95, with a step size of 0.05. The meaning of the formula is to calculate the AP separately under each IoU threshold, and then take the average value under all IoU thresholds.

[0106] Table 2

[0107]

[0108] As can be seen from the above, the embodiment of the present application first obtains images with power high-altitude workers wearing safety belts, and divides the images into training set images and test set images according to a preset ratio, and then introduces the StarNet module, the bidirectional feature pyramid BiFPN module and the LSCD detection head based on the YOLOv10n model to construct a safety belt wearing recognition model, and extracts the input image features by using the specially adjusted StarNet module as the feature extractor of the model to obtain a multi-layer feature map, and then introduces the BiFPN module to improve the Neck part for feature fusion to obtain a fused multi-scale feature map, and then processes the multi-scale feature map through the innovative LSCD detection head to obtain the final detection result of the high-altitude workers wearing safety belts. The present invention combines the improved StarNet with the Neck part and the innovative lightweight detection head LSCD, and the constructed target detection model is more targeted for the wearing method, and the detection result is more accurate and lighter than the original target detection network.

[0109] It should be understood that the size of the serial number of each step in this embodiment does not mean the order of execution of the steps. The execution order of each step should be determined by its function and internal logic, and should not constitute a sole limitation on the implementation process of the embodiment of this application.

[0110] In summary, Figure 6 A detailed flow chart of a method for identifying the wearing of a safety belt for electric power aerial work based on YOLOv10n is provided in an embodiment of the present application, specifically:

[0111] Step 601: Obtain a training sample set and a test sample set;

[0112] Step 602: Based on the YOLOv10n model, the StarNet module, the bidirectional feature pyramid BiFPN module and the LSCD detection head are introduced to construct an original seat belt wearing recognition model;

[0113] Step 603: training the original seat belt wearing recognition model based on the training sample set, and adjusting the network parameters in the original seat belt wearing recognition model according to the training results and the preset loss function to obtain an adjusted seat belt wearing recognition model;

[0114] Step 604: randomly select power aerial work scene image samples from the test sample set and input them into the adjusted safety belt wearing recognition model for target recognition to obtain test results;

[0115] Step 605: If the test passes, the adjusted seat belt wearing recognition model is used as the trained seat belt wearing recognition model; if the test fails, the adjusted seat belt wearing recognition model continues to be trained;

[0116] Step 606: Acquire an image of an actual power aerial work scene;

[0117] Step 607: Input the actual power aerial work scene image into the backbone network in the trained seat belt wearing recognition model to extract features, and obtain an initial multi-scale feature map;

[0118] Step 608: Input the initial multi-scale feature map into the neck network in the trained seat belt wearing recognition model for feature fusion to obtain a fused multi-scale feature map;

[0119] Step 609: input the fused multi-scale feature map into the head network in the trained seat belt wearing recognition model for target recognition, and obtain the seat belt wearing recognition result of the operator.

[0120] For a more detailed process of each step in steps 601 to 609, please refer to the description of the relevant parts shown in the previous text, and the embodiment of the present application will not be repeated here.

[0121] See also Figure 7 , Figure 7 A schematic diagram of a program module of a device for identifying the wearing of a safety belt for electric power aerial work provided in the second aspect of the embodiment of the present application. The device can be used to implement the method for identifying the wearing of a safety belt for electric power aerial work involved in the embodiment of the present application. The device for identifying the wearing of a safety belt for electric power aerial work mainly includes:

[0122] The acquisition module 701 is used to acquire the actual electric power aerial work scene image;

[0123] The recognition module 702 is used to extract and recognize features of actual power high-altitude operation scene images based on the trained seat belt wearing recognition model to obtain the seat belt wearing recognition results of the operators; wherein the seat belt wearing recognition model is constructed based on the YOLOv10n model by introducing the StarNet module, the bidirectional feature pyramid BiFPN module and the LSCD detection head.

[0124] In some implementations of the present embodiment, before the recognition module 702 executes the step of extracting and identifying features of actual electric power high-altitude working scene images based on the trained seat belt wearing recognition model, it also includes a training module, which is used to: obtain a training sample set and a test sample set; wherein the training sample set and the test sample set include multiple electric power high-altitude working scene image samples; randomly select electric power high-altitude working scene image samples from the training sample set and input them into the original seat belt wearing recognition model for target recognition to obtain a training result; based on the training result and a preset loss function, the network parameters in the original seat belt wearing recognition model are adjusted to obtain an adjusted seat belt wearing recognition model; wherein the preset loss function is a multiple loss function including category loss, bounding box loss and distributed regression loss; randomly select electric power high-altitude working scene image samples from the test sample set and input them into the adjusted seat belt wearing recognition model for target recognition to obtain a test result; if the test passes, the adjusted seat belt wearing recognition model is used as the trained seat belt wearing recognition model; if the test fails, the adjusted seat belt wearing recognition model continues to be trained.

[0125] Furthermore, in some implementations of the present embodiment, when the recognition module 702 performs the step of extracting and recognizing features of actual high-altitude power operation scene images based on the trained seat belt wearing recognition model to obtain the seat belt wearing recognition results of the operators, it is specifically used to: input the actual high-altitude power operation scene images into the backbone network of the seat belt wearing recognition model for feature extraction to obtain an initial multi-scale feature map; input the initial multi-scale feature map into the neck network of the seat belt wearing recognition model for feature fusion to obtain a fused multi-scale feature map; input the fused multi-scale feature map into the head network of the seat belt wearing recognition model for target recognition to obtain the seat belt wearing recognition results of the operators.

[0126] In some implementations of this embodiment, the backbone network in the seat belt wearing recognition model in the recognition module 702 adopts a StarNet module, and the StarNet module includes:

[0127] The first operation module is configured to include an initialized convolutional layer and increase the number of channels of the input image from 3 to 16;

[0128] A second operation module, configured to include a first-stage star operation of a block, each block including a depthwise separable convolutional layer and two fully connected layers;

[0129] The third operation module is configured as a second-stage star operation including one block, each block includes a depthwise separable convolutional layer and two fully connected layers, and the number of channels is increased to 32;

[0130] The fourth operation module is configured as a third-stage star operation containing three blocks, each block includes a depthwise separable convolutional layer and two fully connected layers, and the number of channels is increased to 64;

[0131] a fifth operation module, configured as a fourth-stage star operation including one block, each block including a depthwise separable convolutional layer and two fully connected layers, and increasing the number of channels to 128;

[0132] A sixth operation module, configured as an SPPF module in YOLOv8, for performing multi-scale feature fusion;

[0133] The seventh operation module is configured as the PSA module in YOLOv10, which is used to improve the feature representation capability.

[0134] Furthermore, in some implementations of the present embodiment, when the recognition module 702 performs the step of inputting the multi-scale feature map into the neck network of the seat belt wearing recognition model for feature fusion to obtain a fused multi-scale feature map, it is specifically used to: sequentially pass the initial multi-scale feature map to the BiFPN module in the neck network for bidirectional fusion to obtain a preliminary fused multi-scale feature map; input the preliminary fused multi-scale feature map into the C2F module in the neck network for feature enhancement to obtain a fused multi-scale feature map.

[0135] In some implementations of this embodiment, each stage of the StarNet operation in the recognition module 702 is composed of a convolution downsampling module and stacked Star blocks, and the formula for each Star block to complete the nonlinear high-dimensional feature mapping is expressed as:

[0136]

[0137] in and are weight matrices of two sets of linear transformations, ReLU6 is a nonlinear activation function, ⊙ represents element-by-element product, and the star operation maps the input features to an implicit high-dimensional nonlinear space, ultimately generating a dim dimension. output It is expressed as:

[0138]

[0139] Where d is the number of input channels.

[0140] Furthermore, in some implementations of the present embodiment, when the recognition module 702 executes the step of inputting the fused multi-scale feature map into the head network of the seat belt wearing recognition model for target recognition to obtain the seat belt wearing recognition result of the operator, it is specifically used to input the fused multi-scale feature map into the shared convolution module in the head network for feature extraction to obtain the feature extraction result; the feature extraction result is passed to the two parallel bounding box regression branches and classification prediction branches in the head network for target recognition to obtain the seat belt wearing recognition result of the operator.

[0141] According to the electric power aerial work safety belt wearing recognition device provided by this embodiment, firstly, images of electric power aerial workers wearing safety belts are obtained, and the images are divided into training set images and test set images according to a preset ratio. Then, based on the YOLOv10n model, the StarNet module, the bidirectional feature pyramid BiFPN module and the LSCD detection head are introduced to construct a safety belt wearing recognition model. The input image features are extracted by using the specially adjusted StarNet module as the feature extractor of the model to obtain a multi-layer feature map. Then, the BiFPN module is introduced to improve the Neck part for feature fusion to obtain a fused multi-scale feature map. Then, the multi-scale feature map is processed by the innovative LSCD detection head to obtain the final detection result of the aerial workers wearing safety belts. This device combines the improved StarNet with the Neck part and the innovative lightweight detection head LSCD to build a target detection model that is more targeted to the wearing method, and is more accurate and lighter than the original target detection network detection result.

[0142] See also Figure 8 , Figure 8 A module block diagram of an electronic device provided in an embodiment of the present application.

[0143] like Figure 8 As shown, an embodiment of the present application further provides an electronic device, which can be used to implement the method for wearing a safety belt for high-altitude power work in the aforementioned embodiment, and includes a memory 801 and at least one processor 802; wherein the memory 801 is used to store at least one program, and when the at least one program is executed by the at least one processor 802, the at least one processor 802 executes the method for wearing a safety belt for high-altitude power work provided in the embodiment of the present application.

[0144] See also Fig. 9 , Fig. 9 A module block diagram of a computer-readable storage medium provided in an embodiment of the present application.

[0145] like Fig. 9As shown, the embodiment of the present application further provides a computer-readable storage medium 900, on which executable instructions 910 are stored. When the executable instructions 910 are executed, the method for wearing a safety belt for high-altitude power work provided in the embodiment of the present application is executed.

[0146] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0147] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in this application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integration. The available medium can be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk), etc.

[0148] It should be noted that each embodiment in the present application is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For product-type embodiments, since they are similar to method-type embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method-type embodiments.

[0149] It should also be noted that, in the present application, 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 any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or equipment. In the absence of further restrictions, the elements defined by the statement "comprise one..." do not exclude the existence of other identical elements in the process, method, article or equipment including the elements.

[0150] The above description of the disclosed embodiments enables professionals and technicians in the field to implement or use the present application. Various modifications to these embodiments will be apparent to professionals and technicians in the field, and the general principles defined in the present application can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown in the present application, but will conform to the widest range consistent with the principles and novel features disclosed in the present application.

Claims

1. A method for identifying the wearing of a safety belt for electric power aerial work based on YOLOv10n, characterized in that: include: Acquire images of actual power aerial work scenes; Based on the trained seat belt wearing recognition model, feature extraction and recognition are performed on the actual power high-altitude operation scene image to obtain the seat belt wearing recognition result of the operator; wherein, the seat belt wearing recognition model is constructed based on the YOLOv10n model by introducing the StarNet module, the bidirectional feature pyramid BiFPN module and the LSCD detection head.

2. The method for identifying the wearing of a safety belt for electric power aerial work according to claim 1, characterized in that: Before extracting and identifying features of the actual electric power aerial work scene image based on the trained safety belt wearing recognition model, the method further includes: Acquire a training sample set and a test sample set; wherein the training sample set and the test sample set include a plurality of electric power aerial work scene image samples; Randomly select the power aerial work scene image sample from the training sample set and input it into the original safety belt wearing recognition model for target recognition to obtain a training result; Based on the training result and the preset loss function, the network parameters in the original seat belt wearing recognition model are adjusted to obtain an adjusted seat belt wearing recognition model; wherein the preset loss function is a multiple loss function including category loss, bounding box loss and distributed regression loss; Randomly select the power aerial work scene image sample from the test sample set and input it into the adjusted seat belt wearing recognition model for target recognition to obtain a test result; If the test passes, the adjusted seat belt wearing recognition model is used as the trained seat belt wearing recognition model; if the test fails, the adjusted seat belt wearing recognition model continues to be trained.

3. The method for identifying the wearing of a safety belt for electric power aerial work according to claim 2, characterized in that: The feature extraction and recognition of the actual electric power high-altitude operation scene image based on the trained seat belt wearing recognition model is performed to obtain the seat belt wearing recognition result of the operator, including: Inputting the actual power aerial work scene image into the backbone network of the safety belt wearing recognition model for feature extraction to obtain an initial multi-scale feature map; Inputting the initial multi-scale feature map into the neck network of the seat belt wearing recognition model for feature fusion to obtain a fused multi-scale feature map; The fused multi-scale feature map is input into the head network of the seat belt wearing recognition model for target recognition, so as to obtain the seat belt wearing recognition result of the operator.

4. The method for identifying the wearing of a safety belt for electric power aerial work according to claim 3, characterized in that: The backbone network in the seat belt wearing recognition model adopts the StarNet module, and the StarNet module includes: The first operation module is configured to include an initialized convolutional layer and increase the number of channels of the input image from 3 to 16; A second operation module, configured to include a first-stage star operation of a block, each block including a depthwise separable convolutional layer and two fully connected layers; A third operation module is configured to include a second stage star operation of a block, each block includes a depthwise separable convolutional layer and two fully connected layers, and the number of channels is increased to 32; a fourth operation module, configured as a third-stage star operation including three blocks, each block including a depthwise separable convolutional layer and two fully connected layers, and increasing the number of channels to 64; a fifth operation module configured to include a fourth stage star operation of a block, each block including a depthwise separable convolutional layer and two fully connected layers, and increasing the number of channels to 128; A sixth operation module, configured as a SPPF module in YOLOv8; The seventh operation module is configured as a PSA module in YOLOv10.

5. The method for identifying the wearing of a safety belt for electric power aerial work according to claim 4, characterized in that: The inputting the multi-scale feature map into the neck network of the seat belt wearing recognition model to perform feature fusion to obtain the fused multi-scale feature map comprises: The initial multi-scale feature map is sequentially transmitted to the BiFPN module in the neck network for bidirectional fusion to obtain a preliminary fused multi-scale feature map; The initially fused multi-scale feature map is input into the C2F module in the neck network for feature enhancement to obtain a fused multi-scale feature map.

6. The method for identifying the wearing of a safety belt for electric power aerial work according to claim 4, characterized in that: Each stage of the star operation in the StarNet module consists of a convolutional downsampling module and stacked star blocks. The formula for each star block to complete the nonlinear high-dimensional feature mapping is expressed as: in and are weight matrices of two sets of linear transformations, ReLU6 is a nonlinear activation function, ⊙ represents element-by-element product, and the star operation maps the input features to an implicit high-dimensional nonlinear space, ultimately generating a dim dimension. output It is expressed as: Where d is the number of input channels.

7. The method for identifying the wearing of a safety belt for electric power aerial work according to claim 5, characterized in that: The step of inputting the fused multi-scale feature map into the head network of the seat belt wearing recognition model for target recognition to obtain the seat belt wearing recognition result of the operator includes: Inputting the fused multi-scale feature map into the shared convolution module in the head network for feature extraction to obtain a feature extraction result; The feature extraction result is transferred to two parallel bounding box regression branches and classification prediction branches in the head network for target recognition, so as to obtain the safety belt wearing recognition result of the operator.

8. A YOLOv10n-based power aerial work safety belt wearing recognition system, characterized in that: include: An acquisition module is used to acquire images of actual power aerial work scenes; The recognition module is used to extract and recognize features of the actual power high-altitude operation scene image based on the trained seat belt wearing recognition model to obtain the seat belt wearing recognition results of the operators; wherein the seat belt wearing recognition model is constructed based on the YOLOv10n model by introducing the StarNet module, the bidirectional feature pyramid BiFPN module and the LSCD detection head.

9. An electronic device, characterized in that: The device comprises a memory and a processor, wherein: The processor is used to execute the computer program stored in the memory; When the processor executes the computer program, the steps in the method for identifying the wearing of a safety belt for electric power aerial work described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps in the method for identifying the wearing of a safety belt for electric power aerial work described in any one of claims 1 to 7 are implemented.

Images