CN120107744A - A method and product for identifying the wearing of safety belts for power aerial work based on YOLOv10n - Google Patents
A method and product for identifying the wearing of safety belts for power aerial work based on YOLOv10n Download PDFInfo
- Publication number
- CN120107744A CN120107744A CN202510047791.1A CN202510047791A CN120107744A CN 120107744 A CN120107744 A CN 120107744A CN 202510047791 A CN202510047791 A CN 202510047791A CN 120107744 A CN120107744 A CN 120107744A
- Authority
- CN
- China
- Prior art keywords
- safety belt
- belt wearing
- module
- model
- recognition
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Image Analysis (AREA)
Abstract
The application provides a YOLOv n-based power aerial work safety belt wearing identification method and a product, the method comprises the steps of obtaining an actual power aerial work scene image, carrying out feature extraction and identification on the actual power aerial work scene image based on a trained safety belt wearing identification model to obtain a safety belt wearing identification result of an operator, wherein the safety belt wearing identification model is obtained by constructing a StarNet module, a bidirectional feature pyramid BiFPN module and an LSCD detection head based on a YOLOv n model. According to the application, starNet modules are introduced into the head network, biFPN modules are introduced into the neck network and combined with the LSCD detection head, and the built safety belt wearing recognition model has more pertinence to the wearing mode, and is more accurate and lighter than the detection result of the original target detection network.
Description
Technical Field
The application relates to the technical field of computer vision and the field of electric power safety, in particular to a YOLOv n-based wearing identification method and a YOLOv n-based wearing identification product for an electric power high-altitude operation safety belt.
Background
In the electric power construction operation, the safety problem of the high-altitude operators is always an important issue which needs attention. According to the related statistical data, the fact that the safety belt is not worn is one of important factors causing electric power construction accidents, and the accidents not only cause casualties, but also bring huge economic losses to enterprises. Studies have shown that a substantial proportion of accidents due to the lack of safety belts account for the total number of electric construction accidents, further confirming the necessity of enhanced safety management in high risk environments. The safety belt serves as basic personal protective equipment, and the correct wearing of the safety belt can significantly reduce the risk of falling in high-altitude operations. Although relevant regulations and safety standards have clearly required workers to wear safety belts while working aloft, in actual construction, workers fail to comply with such regulations for various reasons, resulting in an increase in safety hazards. Therefore, how to effectively monitor and ensure that a worker wears the safety belt correctly becomes one of the key measures to raise the level of construction safety management.
The wearing detection of the safety belt is an important link for guaranteeing the safety of workers in the power construction operation. Traditionally, this detection has relied primarily on manual monitoring and observation. Although the method has high flexibility and can discover problems in time, the defects of strong subjectivity and low efficiency make full coverage difficult in a large-scale construction site, and the risk of missed detection exists. Along with the development of technology, the traditional image processing technology is gradually applied to wearing detection of safety belts, real-time monitoring is realized through a camera and an algorithm, interference of human factors can be reduced, and the objectivity of detection is improved. However, existing models often face large volume and computational resource consumption, which limits their popularization in practical applications, especially on resource-constrained construction sites. In addition, the challenges of adaptability of image processing technology in complex environments, data privacy problem, and high equipment cost are also to be solved. Therefore, future research needs to pay attention to how to optimize the model to reduce the consumption of computing resources, and meanwhile, combine artificial intelligence and the internet of things technology to realize more accurate and intelligent safety belt detection, so that the safety management level of the power construction operation is further improved.
In order to solve the above challenges, the present invention focuses on developing a belt target detection model based on YOLOv a 10. The model aims at providing an innovative and efficient solution for the belt wearing detection of the electric power construction site by combining a lightweight network architecture, a scene-specific optimization strategy and an efficient training method. The scheme not only can meet the actual application requirements, but also can promote the technical progress of the electric power system in the aspects of safety and reliability.
Disclosure of Invention
The application aims to provide a YOLOv n-based ultra-lightweight power aloft work safety belt wearing identification method and a product, which at least can solve the problems of insufficient accuracy of a detection model and high resource consumption in the related technology.
In order to solve the above technical problems, a first aspect of the present application provides a method for identifying wearing of a safety belt for power aerial work based on YOLOv n, including:
acquiring an actual power aerial work scene image;
And carrying out feature extraction and recognition on the actual power aerial work scene image based on the trained safety belt wearing recognition model to obtain a safety belt wearing recognition result of an operator, wherein the safety belt wearing recognition model is obtained by introducing a StarNet module, a bidirectional feature pyramid BiFPN module and an LSCD detection head based on a YOLOv n model.
A second aspect of the embodiment of the present application provides a power aerial work safety belt wearing recognition system based on YOLOv n, including:
the acquisition module is used for acquiring an actual power aerial work scene image;
The safety belt wearing recognition module is used for carrying out feature extraction and recognition on the actual power aerial operation scene image based on a safety belt wearing recognition model which is completed through training to obtain a safety belt wearing recognition result of an operator, wherein the safety belt wearing recognition model is obtained by constructing a StarNet module, a bidirectional feature pyramid BiFPN module and an LSCD detection head based on a YOLOv n model.
The third aspect of the application provides an electronic device, which comprises a memory and a processor, wherein the processor is used for executing a computer program stored on the memory, and when the processor executes the computer program, the steps in the method for identifying wearing of the safety belt for the electric power aloft work according to the first aspect of the application are realized.
A fourth aspect of the present application provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the method for identifying belt wear for electric power aloft work according to the first aspect of the present application.
As can be seen from the above, in the embodiment of the application, an actual power aerial work scene image is firstly obtained, and then, feature extraction and recognition are performed on the actual power aerial work scene image based on a trained safety belt wearing recognition model to obtain a safety belt wearing recognition result of an operator, wherein the safety belt wearing recognition model is obtained by constructing a StarNet module, a bidirectional feature pyramid BiFPN module and an LSCD detection head based on a YOLOv n model, and a StarNet module, a bidirectional feature pyramid BiFPN module and an LSCD detection head based on a YOLOv n model.
It should be understood that the description in this section is not intended to identify key or critical features of the application or to delineate the scope of the application. Other features of the present application will become apparent from the description that follows.
Drawings
In order to more clearly illustrate the technical solutions of the related art or embodiments of the present application, the drawings that are needed in the description of the related art or embodiments of the present application will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present application, but not all embodiments, and that other drawings may be obtained according to these drawings without inventive effort to a person of ordinary skill in the art.
Fig. 1 is a schematic flow chart of a method for identifying wearing of a safety belt for power aloft work based on YOLOv n according to a first aspect of the present application;
Fig. 2 is a main structure diagram of a method for identifying wearing of a safety belt for power aloft work based on YOLOv n according to a first aspect of the present application;
FIG. 3 is a schematic diagram of a stage architecture of star-opreation in StarNet module in a method for identifying belt wear for power aloft work based on YOLOv n according to a first aspect of the present application;
Fig. 4 is a block diagram of an LSCD in a method for identifying wearing of a safety belt for power aloft work based on YOLOv n according to a first aspect of the present application;
fig. 5 is an exemplary diagram of an identification effect of a method for identifying wearing of a safety belt for power aloft work based on YOLOv n according to a first aspect of the present application;
fig. 6 is a detailed flow chart of a method for identifying wearing of a safety belt for power aerial operation based on YOLOv n according to a first aspect of the present application;
Fig. 7 is a schematic program module diagram of a belt wearing recognition device for power aerial work according to a second aspect of the present application;
FIG. 8 is a block diagram of an electronic device according to a third aspect of an embodiment of the present application;
Fig. 9 is a block diagram of a computer readable storage medium according to a fourth aspect of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more obvious and understandable, the present application will be clearly and completely described in the following description with reference to the embodiments of the present application and the drawings thereof, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. It should be understood that the following embodiments of the present application are only for explaining the present application and are not intended to limit the present application, that is, all other embodiments obtained by persons skilled in the art without making any inventive effort based on the embodiments of the present application are within the scope of protection of the present application. Furthermore, the technical features referred to in the embodiments of the present application described below may be combined with each other as long as they do not make a conflict with each other.
Referring to fig. 1, fig. 1 is a schematic flow chart of a method for identifying the wearing of a safety belt for power aerial operation based on YOLOv n according to a first aspect of the present application, the method for identifying the wearing of the safety belt for power aerial operation includes the following steps.
And 101, acquiring an actual power aerial work scene image.
And 102, carrying out feature extraction and recognition on an actual power aerial work scene image based on a trained safety belt wearing recognition model to obtain a safety belt wearing recognition result of an operator, wherein the safety belt wearing recognition model is obtained by constructing a StarNet module, a bidirectional feature pyramid BiFPN module and an LSCD detection head based on a YOLOv n model.
In the embodiment of the application, the characteristic extractor of the model is taken as a specially-adjusted StarNet module to extract the input image characteristic, then the improved Neck part of the BiFPN module is introduced to perform characteristic fusion to obtain a fused multi-scale characteristic image, finally the multi-scale characteristic image is processed through an innovative LSCD detection head to obtain the final type of the safety belt worn by the aerial worker.
In an optional embodiment of this embodiment, before the feature extraction and the recognition of the actual power aerial work scene image based on the trained belt wearing recognition model, the method further includes:
acquiring a training sample set and a test sample set, wherein the training sample set and the test sample set comprise a plurality of power aerial work scene image samples;
randomly selecting an image sample of the power aerial work scene from the training sample set, inputting the image sample into an original safety belt wearing recognition model for target recognition, and obtaining a training result;
adjusting network parameters in the original safety belt wearing recognition model based on a training result and a preset loss function to obtain an adjusted safety belt wearing recognition model, wherein the preset loss function is a multiple loss function comprising category loss, boundary box loss and distributed regression loss;
Randomly selecting an image sample of the power aerial work scene from the test sample set, inputting the image sample into the adjusted safety belt wearing recognition model for target recognition, and obtaining a test result;
if the test is passed, the adjusted safety belt wearing recognition model is used as a safety belt wearing recognition model after training is completed; if the test is not passed, the adjusted safety belt wearing recognition model is continuously trained.
In the embodiment of the application, an image acquisition system is utilized to manufacture a special data set for a model of an image of a safety belt worn by an aerial worker, which is cooperatively acquired by an electric power system company, 1800 images are acquired, the resolution is 1936 x 1296, and the image quantity ratio of a training set to a testing set is 4:1. It can be understood that the image in the dataset needs to be preprocessed before training, including recording the position of the wearer in the image of the dataset and the wearing type thereof, and comparing the position with the output result to calculate the accuracy of target detection, as shown in table 1, the number of various wearing types in the dataset.
TABLE 1
Data in the training set are input into the model for training according to batches, training data are loaded from dataset/data.yaml, the total training round number is 300, 32 pictures are trained in each batch, and the image size is 640 x 640. The SGD optimizer was used in training, with an initial learning rate of 0.01, and was adjusted stepwise after warmup _ epochs =3. The model optimizes the model's performance in classification and localization using multiple loss functions including class loss (cls=0.5), bounding box loss (box=7.5), and distributed regression loss (dfl =1.5).
In an optional embodiment of the present embodiment, feature extraction and identification are performed on an actual power aerial work scene image based on a trained belt wearing identification model, so as to obtain a belt wearing identification result of an operator, including:
Inputting an actual power aerial work scene image into a backbone network of a safety belt wearing recognition model to perform feature extraction, and obtaining an initial multi-scale feature map;
inputting the initial multi-scale feature map into a neck network of a safety belt wearing recognition model for feature fusion to obtain a fused multi-scale feature map;
And inputting the fused multi-scale characteristic diagram into a head network of the safety belt wearing recognition model to perform target recognition, and obtaining a safety belt wearing recognition result of an operator.
Specifically, an actual power aerial working scene image is input into a backbone network in a safety belt wearing recognition model for feature extraction to obtain an initial multi-scale feature image, wherein the backbone network is the backbone network with the improved lead-in StarNet module, the specific model is described in a subsequent embodiment, then the initial multi-scale feature image is input into a neck network with the improved lead-in BiFPN module for feature fusion to obtain a fused multi-scale feature image, finally the fused multi-scale feature image is input into a head network with an LSCD detection head for target recognition, and the LSCD detection head in the model carries out final regression of the position of a safety belt wearing person and classification of the wearing type to finally obtain the safety belt wearing recognition result of an operator.
In an alternative embodiment of the present embodiment, the backbone network in the belt wearing recognition model employs a StarNet module, and the StarNet module includes:
A first operation module configured to include an initialized convolution layer and increase the number of channels of the input image from 3 to 16;
A second operation module configured to perform a first phase star operation comprising a block, each block comprising a depth separable convolutional layer and two fully-connected layers;
a third operation module configured to perform a second star operation including a block, each block including a depth separable convolution layer and two full connection layers, and increasing the number of channels to 32;
A fourth operation module configured to perform a third-stage star operation including three blocks, each block including one depth separable convolution layer and two full connection layers, and increasing the number of channels to 64;
A fifth operation module configured to perform a fourth phase star operation including a block, each block including a depth separable convolution layer and two full connection layers, and increasing the number of channels to 128;
A sixth operating module configured as an SPPF module in YOLOv;
a seventh operational module configured as the PSA module in YOLOv.
In an alternative embodiment of the present embodiment, each stage of star operation in the StarNet module is composed of a convolutional downsampling module and stacked star blocks, and each star block performs a nonlinear high-dimensional feature mapping with the formula:
Wherein the method comprises the steps of AndThe weight matrix of two sets of linear transformations, reLU6 is a nonlinear activation function, as well as the product of elements, the star operation maps the input features to the implicit high-dimensional nonlinear space, and finally the generated dimension dim output is expressed as:
Where d is the number of input channels.
Specifically, fig. 2 is a main architecture diagram of a power aerial working safety belt wearing recognition method based on YOLOv n provided in the first aspect of the present application, and the StarNet feature extraction module mainly includes a convolution layer, four star-operation operations, and an original SPPF and PSA module of yolov10, and increases the number of channels of an input image from 3 (RGB) to 16 by initializing the convolution layer (ConvBN), so as to provide more feature space for subsequent feature extraction. The goal is to use Batch Normalization (BN) to stabilize the training process and accelerate convergence.
Star-Operation in the first stage extracts preliminary features by depth separable convolution and full connected layers. Depth separable convolution (ConvBN, 7x7, stride=1, groups=16) is applied to extract spatial features. The features are further processed by two fully connected layers (linear transformations W1 and W2). Element-wise multiplication operations (star operations) are performed between two fully connected layers, enhancing the nonlinear expressive power of the features.
Fig. 3 is a schematic diagram of a Star-opreation stage architecture in a StarNet module in a power aerial working safety belt wearing recognition method based on YOLOv n according to a first aspect of the present application, where each stage is composed of a convolutional downsampling module and a plurality of stacked Star Blocks (Star Blocks). StarNet the body portion has a depth of [1, 3,1] per stage, and the number of channels is multiplied stepwise as the stages are deepened. Each star block is a core unit of feature extraction that performs nonlinear high-dimensional feature mapping by the following formula:
Wherein the method comprises the steps of AndThe weight matrices of two sets of linear transformations, reLU6, which is a nonlinear activation function, represent the element-wise product (i.e., star operation), respectively. The star operation maps the input features to an implicit high-dimensional nonlinear space, enhances the feature expression capability and finally generates dimensions as follows:
Where d is the number of input channels.
Star-Operation in the second stage further extracts features on the basis of the first stage and doubles the number of channels to 32. The repeat depth can separate the operations of rolling and fully connected layers, but at this stage doubles the number of channels.
Star-Operation in the third stage further enhances feature extraction capability and increases the number of channels to 64. This stage contains 3 blocks, each Block repeating operations of depth separable convolution and full join layer. The expressive power of the features is gradually enhanced by these blocks.
Star-Operation in the fourth stage further improves the depth of feature extraction based on the third stage and increases the number of channels to 128. The depth separable rolling and full connection layer operations were performed using 1 Block. Through this stage of operation, deep features are extracted still further.
SPPF module (SPATIAL PYRAMID Pooling-Fast) is to enhance the detection capability of network to different size targets through multi-scale feature fusion. And pooling the feature graphs on different scales to generate a multi-scale feature representation.
Wherein the formula can be expressed as:
Where k represents the pooling window size.
The PSA module (PARTIAL SELF-Attention) improves feature representation capability through a partial self-Attention mechanism while reducing computational costs. The application of a partial self-attention mechanism enables the network to focus on more important parts of the image while reducing computational complexity.
Wherein, the partial self-attention mechanism optimization feature expression is:
x psa＝Softmax(Q·KT)·V
Q, K, V represent the query, key, value matrix generated by the input features, respectively.
Finally, after StarNet module processing, a multi-scale feature map is generated:
In an optional embodiment of this embodiment, feature fusion is performed on a neck network of a belt wearing recognition model for inputting a multi-scale feature map to obtain a fused multi-scale feature map, where the method includes:
Sequentially transmitting the initial multi-scale feature images to BiFPN modules in a neck network for bidirectional fusion to obtain the preliminarily fused multi-scale feature images;
and inputting the preliminarily fused multi-scale feature map to a C2F module in the neck network for feature enhancement to obtain the fused multi-scale feature map.
Specifically, the Neck part receives the multi-scale feature maps P2, P3, P4 and P5 output by the StarNet module, and the feature maps are sequentially transferred to the BiFPN module for bidirectional fusion, wherein the BiFPN module sequentially completes feature fusion from top to bottom and from bottom to top in a bidirectional path. The top-down fusion path gradually transmits information upwards from a low-resolution feature map (such as P5) to a high-resolution feature map (such as P2), and each high-resolution feature is obtained by fusion calculation of the current layer and the next layer of features:
Wherein the method comprises the steps of Represented as i-th layer output features in a top-down path, upsample represents a low resolution feature mapBilinear interpolation upsampling is performed, fusion represents dynamic weighting feature Fusion, and the calculation is as follows:
where ω1, ω2 are the learned weight parameters.
The fused features are further enhanced by a C2F module, and the core operation of the C2F is channel division and splicing, and the method is expressed as follows:
Where y 1、y2 is the output of the initial convolution and y bottleNeck is the feature extracted by recursion BottleNeck.
The bottom-up fusion path gradually transmits information downwards from a high-resolution feature map (such as P2) to a low-resolution feature map (such as P5), and each low-resolution feature is obtained by fusion calculation of the current layer and the previous layer of features:
Wherein the method comprises the steps of Represented as i-th layer output features in bottom-up path, downsample represents a graph of features for high resolutionAnd 3X 3 convolution downsampling is carried out, the fused features are further optimized through a C2F module, and in the feature fusion process, biFPN learns weights of different input features by using a dynamic weighting mechanism, wherein the expression formula is as follows:
Where ω i represents the weight of each input feature, e takes the value 1e-4 to avoid numerical instability.
After bottom-up and bottom-up paths and C2F module processing, the final multi-scale feature map generated in part Neck is expressed as:
in an alternative embodiment of the present embodiment, inputting the fused multi-scale feature map to a head network of a belt wearing recognition model for target recognition, to obtain a belt wearing recognition result of an operator, including:
inputting the fused multi-scale feature map to a shared convolution module in a head network to perform feature extraction to obtain a feature extraction result;
and transmitting the feature extraction result to two parallel boundary box regression branches and classification prediction branches in the head network to perform target recognition, so as to obtain a safety belt wearing recognition result of the operator.
Specifically, fig. 4 is a block diagram of LSCD in a method for identifying wearing of a safety belt for power aloft work based on YOLOv n according to the first aspect of the present application. The LSCD (LIGHTWEIGHT SHARED Convolutional Detection) detection head adopts a lightweight shared convolution module, and combines the double-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 third step, each input feature map firstly carries out feature extraction through the shared convolution module, and the output features are transmitted to two parallel boundary box regression branches and a classification prediction branch. The detection head respectively carries out bounding box and classification prediction on the characteristics of P2', P3', P4', P5', and then outputs detection results of multiple layers in a multi-scale fusion mode.
In the reasoning process, the detection head dynamically generates an anchor point based on the input characteristics and is used for guiding the prediction and decoding of the boundary box. The anchor point size is dynamically adjusted according to the resolution and the step size of each layer of feature map. 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 class. And (3) screening the detection result by non-maximum suppression (NMS), removing the bounding boxes with low confidence and overlarge overlap, and finally outputting the detection result containing the target position, the category and the confidence.
Fig. 5 is an exemplary diagram of an identification effect of a method for identifying wearing of a safety belt for power aloft operation based on YOLOv n according to a first aspect of the present application.
Comparing the test results of the present invention with the deep learning models yolov, yolov, etc., the test results are shown in table 2, and it can be seen that there is a high degree of weight reduction in the parameter amounts and model sizes. In table 2, compared with original yolov, the mAP of the method is improved by 0.02, has smaller model size and higher detection precision, and experiments show that the model of the invention is more suitable for wearing and identifying the safety belt for power aloft work. The evaluation index used in the invention is an evaluation standard mAP50-95 for target detection, and the calculation formula is as follows:
Where mAP50-95 represents the average AP at a number IoU of thresholds (from 0.50 to 0.95, step size 0.05, meaning that the AP is calculated separately at each IoU threshold, and then averaged over all IoU thresholds.
TABLE 2
From the above, it can be seen that, in the embodiment of the application, firstly, an image of a safety belt worn by a power aerial worker is obtained, the image is divided into a training set image and a test set image according to a preset proportion, then, based on a YOLOv10n model, a StarNet module, a bidirectional feature pyramid BiFPN module and an LSCD detection head are introduced to construct a safety belt wearing recognition model, a specially-adjusted StarNet module is used as a feature extractor of the model to extract input image features, a multi-layer feature map is obtained, then, a BiFPN module is introduced to improve Neck part to perform feature fusion, a fused multi-scale feature map is obtained, and then, the innovative LSCD detection head is used for processing the multi-scale feature map to obtain a final detection result of the safety belt worn by the aerial worker.
It should be understood that, the sequence number of each step in this embodiment does not mean the order of execution of the steps, and the execution order of each step should be determined by its functions and internal logic, and should not be construed as a unique limitation on the implementation process of the embodiment of the present application.
To sum up, fig. 6 is a detailed flow chart of a method for identifying wearing of a safety belt for power aerial operation based on YOLOv n according to an embodiment of the present application, specifically:
Step 601, acquiring a training sample set and a test sample set;
step 602, based on YOLOv n model, introducing StarNet module, bidirectional feature pyramid BiFPN module and LSCD detection head to construct original safety belt wearing identification model;
Step 603, training an original safety belt wearing recognition model based on a training sample set, and adjusting network parameters in the original safety belt wearing recognition model according to a training result and a preset loss function to obtain an adjusted safety belt wearing recognition model;
step 604, randomly selecting an electric power aerial work scene image sample from a test sample set, inputting the electric power aerial work scene image sample into an adjusted safety belt wearing recognition model for target recognition, and obtaining a test result;
Step 605, if the test is passed, the adjusted safety belt wearing recognition model is used as the safety belt wearing recognition model after training, if the test is not passed, the training is continued on the adjusted safety belt wearing recognition model;
Step 606, acquiring an actual power aerial work scene image;
step 607, inputting an actual power aerial work scene image into a backbone network in a trained safety belt wearing recognition model for feature extraction, and obtaining an initial multi-scale feature map;
Step 608, inputting the initial multi-scale feature map to a neck network in the trained safety belt wearing recognition model for feature fusion to obtain a fused multi-scale feature map;
step 609, inputting the fused multi-scale feature map to a head network in the trained safety belt wearing recognition model to perform target recognition, and obtaining a safety belt wearing recognition result of an operator.
The detailed flow of each step 601 to 609 is only required to be described in the related parts shown in the foregoing, and the embodiments of the present application are not described herein.
Referring to fig. 7, fig. 7 is a schematic program module diagram of a belt wearing recognition device for power aerial operation according to a second aspect of the present application. The device can be used for realizing the wearing recognition method of the electric power high-altitude operation safety belt. This electric power aloft work safety belt wears recognition device mainly includes:
an acquisition module 701, configured to acquire an actual power aerial work scene image;
The identification module 702 is configured to perform feature extraction and identification on an actual power aerial work scene image based on a trained safety belt wearing identification model to obtain a safety belt wearing identification result of an operator, where the safety belt wearing identification model is obtained by constructing a StarNet module, a bidirectional feature pyramid BiFPN module and an LSCD detection head based on a YOLOv n model.
In some implementations of the present embodiment, before the step of performing feature extraction and identification on the actual power aerial work scene image based on the trained belt wearing identification model by the identification module 702, the method further includes a training module for acquiring a training sample set and a test sample set, where the training sample set and the test sample set include a plurality of power aerial work scene image samples, randomly selecting the power aerial work scene image samples from the training sample set, inputting the power aerial work scene image samples to the original belt wearing identification model to perform target identification to obtain a training result, adjusting network parameters in the original belt wearing identification model based on the training result and a preset loss function to obtain an adjusted belt wearing identification model, wherein the preset loss function is a multiple loss function including category loss, bounding box loss and distributed regression loss, randomly selecting the power aerial work scene image samples from the test sample set, inputting the adjusted belt wearing identification model to perform target identification to obtain a test result, if the test is passed, taking the adjusted belt wearing identification model as the trained belt wearing identification model, and if the test is not passed, continuing training the adjusted belt wearing identification model.
Further, in some implementations of the present embodiment, when performing the step of performing feature extraction and identification on the actual power aerial work scene image based on the trained belt wearing identification model to obtain the belt wearing identification result of the operator, the identification module 702 is specifically configured to input the actual power aerial work scene image to the backbone network of the belt wearing identification model to perform feature extraction to obtain an initial multi-scale feature map, input the initial multi-scale feature map to the neck network of the belt wearing identification model to perform feature fusion to obtain a fused multi-scale feature map, and input the fused multi-scale feature map to the head network of the belt wearing identification model to perform target identification to obtain the belt wearing identification result of the operator.
In some implementations of this embodiment, the backbone network in the seat belt wear identification model in the identification module 702 employs a StarNet module, and the StarNet module includes:
A first operation module configured to include an initialized convolution layer and increase the number of channels of the input image from 3 to 16;
A second operation module configured to perform a first phase star operation comprising a block, each block comprising a depth separable convolutional layer and two fully-connected layers;
a third operation module configured to perform a second star operation including a block, each block including a depth separable convolution layer and two full connection layers, and increasing the number of channels to 32;
A fourth operation module configured to perform a third-stage star operation including three blocks, each block including one depth separable convolution layer and two full connection layers, and increasing the number of channels to 64;
A fifth operation module configured to perform a fourth phase star operation including a block, each block including a depth separable convolution layer and two full connection layers, and increasing the number of channels to 128;
a sixth operation module configured as an SPPF module in YOLOv for performing multi-scale feature fusion;
A seventh operational module, configured as the PSA module in YOLOv, for improved feature representation capabilities.
Further, in some implementations of the present embodiment, when the step of performing feature fusion on the neck network of the belt wearing recognition model to obtain a fused multi-scale feature map by using the recognition module 702 is specifically used for sequentially transmitting the initial multi-scale feature map to the BiFPN module in the neck network to perform bidirectional fusion to obtain a primarily fused multi-scale feature map, and inputting the primarily fused multi-scale feature map to the C2F module in the neck network to perform feature enhancement to obtain the fused multi-scale feature map.
In some implementations of the present embodiment, each stage of star operation in StarNet module in identification module 702 consists of a convolutional downsampling module and stacked star blocks, each of which performs a non-linear high-dimensional feature mapping formulated as:
Wherein the method comprises the steps of AndThe weight matrix of two sets of linear transformations, reLU6 is a nonlinear activation function, as well as the product of elements, the star operation maps the input features to the implicit high-dimensional nonlinear space, and finally the generated dimension dim output is expressed as:
Where d is the number of input channels.
Further, in some implementations of the present embodiment, when the step of inputting the fused multi-scale feature map to the head network of the belt wearing recognition model to perform target recognition and obtain the belt wearing recognition result of the operator is performed, the recognition module 702 is specifically configured to input the fused multi-scale feature map to the shared convolution module in the head network to perform feature extraction, obtain the feature extraction result, and transmit the feature extraction result to two parallel bounding box regression branches and classification prediction branches in the head network to perform target recognition, thereby obtaining the belt wearing recognition result of the operator.
According to the electric power high-altitude operation safety belt wearing recognition device provided by the embodiment, firstly, an image with a safety belt worn by an electric power high-altitude operation personnel is obtained, the image is divided into a training set image and a testing set image according to a preset proportion, then a safety belt wearing recognition model is built based on a YOLOv n model, a StarNet module, a bidirectional feature pyramid BiFPN module and an LSCD detection head, input image features are extracted through a feature extractor taking a specially-adjusted StarNet module as a model, a multilayer feature map is obtained, then an improved Neck part of a BiFPN module is introduced for feature fusion, a fused multi-scale feature map is obtained, and then the multi-scale feature map is processed through an innovative LSCD detection head to obtain a final detection result of the safety belt worn by the high-altitude operation personnel.
Referring to fig. 8, fig. 8 is a block diagram of an electronic device according to an embodiment of the application.
As shown in fig. 8, the embodiment of the present application further provides an electronic device, which may be used to implement the method for wearing a power aerial work safety belt in the foregoing embodiment, and includes a memory 801 and at least one processor 802, where the memory 801 is configured 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 is caused to execute the method for wearing a power aerial work safety belt provided in the embodiment of the present application.
Referring to fig. 9, fig. 9 is a block diagram of a computer readable storage medium according to an embodiment of the application.
As shown in fig. 9, an embodiment of the present application further provides a computer readable storage medium 900, where executable instructions 910 are stored on the computer readable storage medium 900, and when the executable instructions 910 are executed, the method for wearing the electric power aloft work safety belt provided by the embodiment of the present application is executed.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software modules may be disposed in Random Access Memory (RAM), memory, read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, may 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 processes or functions in accordance with the present application are produced in whole or in part. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may 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 may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by a wired (e.g., coaxial cable, fiber optic, digital subscriber line), or wireless (e.g., infrared, wireless, microwave, etc.). Computer readable storage media can be any available media that can be accessed by a computer or data storage devices, such as servers, data centers, etc., that contain an integration of one or more available media. Usable media may be magnetic media (e.g., floppy disk, hard disk, magnetic tape), optical media (e.g., DVD), or semiconductor media (e.g., solid state disk Solid STATE DISK), etc.
It should be noted that, in the present disclosure, each embodiment is described in a progressive manner, and each embodiment is mainly described as different from other embodiments, and identical and similar parts between the embodiments are all enough to be referred to each other. For product class embodiments, the description is relatively simple as it is similar to method class embodiments, as relevant points are found in the partial description of method class embodiments.
It should also be noted that in the present disclosure, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises an element.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present disclosure is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. The power high-altitude operation safety belt wearing identification method based on YOLOv n is characterized by comprising the following steps of:
acquiring an actual power aerial work scene image;
And carrying out feature extraction and recognition on the actual power aerial work scene image based on the trained safety belt wearing recognition model to obtain a safety belt wearing recognition result of an operator, wherein the safety belt wearing recognition model is obtained by introducing a StarNet module, a bidirectional feature pyramid BiFPN module and an LSCD detection head based on a YOLOv n model.
2. The method according to claim 1, wherein before the training-based belt wearing recognition model performs feature extraction and recognition on the actual power overhead operation scene image, the method further comprises:
Acquiring a training sample set and a test sample set, wherein the training sample set and the test sample set comprise a plurality of power aerial work scene image samples;
Randomly selecting the power aerial work scene image sample from the training sample set, inputting the power aerial work scene image sample into an original safety belt wearing recognition model for target recognition, and obtaining a training result;
Adjusting network parameters in the original safety belt wearing recognition model based on the training result and a preset loss function to obtain an adjusted safety belt wearing recognition model, wherein the preset loss function is a multiple loss function comprising category loss, boundary frame loss and distributed regression loss;
randomly selecting the power aerial work scene image sample from the test sample set, inputting the power aerial work scene image sample into the adjusted safety belt wearing recognition model for target recognition, and obtaining a test result;
if the test is passed, the adjusted safety belt wearing recognition model is used as a safety belt wearing recognition model after training is completed; if the test does not pass, continuing to train the adjusted safety belt wearing recognition model.
3. The method for identifying the wearing of the safety belt for the electric power aloft work according to claim 2, wherein the feature extraction and the identification of the actual electric power aloft work scene image based on the trained safety belt wearing identification model are performed to obtain the safety belt wearing identification result of the operator, and the method comprises the following steps:
Inputting the actual power aerial work scene image into a 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 to a neck network of the safety belt wearing recognition model for feature fusion to obtain a fused multi-scale feature map;
And inputting the fused multi-scale characteristic diagram to a head network of the safety belt wearing recognition model to perform target recognition, so as to obtain a safety belt wearing recognition result of an operator.
4. A method of power aloft work safety belt wear identification as claimed in claim 3 wherein the backbone network in the safety belt wear identification model employs StarNet modules, the StarNet modules comprising:
A first operation module configured to include an initialized convolution layer and increase the number of channels of the input image from 3 to 16;
A second operation module configured to perform a first phase star operation comprising a block, each block comprising a depth separable convolutional layer and two fully-connected layers;
A third operation module configured to include a second star operation of blocks, each block including a depth separable convolutional layer and two full join layers, and increasing the number of channels to 32;
A fourth operation module configured to perform a third phase star operation comprising three blocks, each block comprising a depth separable convolutional layer and two fully-connected layers, and increasing the number of channels to 64;
A fifth operation module configured to perform a fourth phase star operation including a block, each block including a depth separable convolution layer and two full connection layers, and increasing the number of channels to 128;
A sixth operating module configured as an SPPF module in YOLOv;
a seventh operational module configured as the PSA module in YOLOv.
5. The method for identifying the wearing of the safety belt for the electric power aloft work according to claim 4, wherein the step of performing feature fusion on the neck network of the safety belt wearing identification model by inputting the multi-scale feature map to obtain a fused multi-scale feature map comprises the steps of:
sequentially transmitting the initial multi-scale feature map to BiFPN modules in the neck network for bidirectional fusion to obtain a preliminarily fused multi-scale feature map;
And inputting the preliminarily fused multi-scale feature map to a C2F module in the neck network for feature enhancement to obtain a fused multi-scale feature map.
6. The method of claim 4, wherein each stage of star operation in the StarNet module consists of a convolutional downsampling module and stacked star blocks, and each of the star blocks performs a nonlinear high-dimensional feature mapping with a formula expressed as:
Wherein the method comprises the steps of AndThe weight matrix of two sets of linear transformations, reLU6 is a nonlinear activation function, and as a result, the element-wise product is represented by the following, the star operation maps the input features to the implicit high-dimensional nonlinear space, and finally the dimension dim output is represented as:
Where d is the number of input channels.
7. The method for identifying the wearing of the safety belt for the electric power aloft work according to claim 5, wherein the step of inputting the fused multi-scale feature map to the head network of the safety belt wearing identification model to perform target identification, and obtaining the safety belt wearing identification result of the operator comprises the following steps:
inputting the fused multi-scale feature map to a shared convolution module in the head network for feature extraction to obtain a feature extraction result;
And transmitting the feature extraction result to two parallel boundary box regression branches and classification prediction branches in the head network to perform target recognition, so as to obtain a safety belt wearing recognition result of an operator.
8. YOLOv10 n-based power aerial work safety belt wearing recognition system is characterized by comprising:
the acquisition module is used for acquiring an actual power aerial work scene image;
The safety belt wearing recognition module is used for carrying out feature extraction and recognition on the actual power aerial operation scene image based on a safety belt wearing recognition model which is completed through training to obtain a safety belt wearing recognition result of an operator, wherein the safety belt wearing recognition model is obtained by constructing a StarNet module, a bidirectional feature pyramid BiFPN module and an LSCD detection head based on a YOLOv n model.
9. An electronic device comprising a memory and a processor, wherein:
the processor is used for executing the computer program stored on the memory;
the processor, when executing the computer program, implements the steps in the power aloft work seat belt wearing recognition method according to any one of claims 1 to 7.
10. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed by a processor, implements the steps in the electric power aloft work seat belt wearing recognition method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510047791.1A CN120107744A (en) | 2025-01-10 | 2025-01-10 | A method and product for identifying the wearing of safety belts for power aerial work based on YOLOv10n |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510047791.1A CN120107744A (en) | 2025-01-10 | 2025-01-10 | A method and product for identifying the wearing of safety belts for power aerial work based on YOLOv10n |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN120107744A true CN120107744A (en) | 2025-06-06 |
Family
ID=95882006
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202510047791.1A Pending CN120107744A (en) | 2025-01-10 | 2025-01-10 | A method and product for identifying the wearing of safety belts for power aerial work based on YOLOv10n |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN120107744A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118644761A (en) * | 2024-07-24 | 2024-09-13 | 山东滨州城建集团有限公司 | A construction site safety helmet detection method, computer equipment and storage medium |
| CN119048496A (en) * | 2024-10-30 | 2024-11-29 | 南昌工程学院 | PCB defect detection method, system, equipment and storage medium |
| CN119152463A (en) * | 2024-11-13 | 2024-12-17 | 华南理工大学 | Method, device, equipment and medium for detecting small traffic target of high-speed running vehicle |
| CN119169305A (en) * | 2024-08-28 | 2024-12-20 | 合肥工业大学 | A gas leak detection method based on infrared imaging of YOLOV8 |
| CN119206349A (en) * | 2024-09-26 | 2024-12-27 | 浙江工业大学 | A real-time detection method for laparoscopic instruments |
| CN119229476A (en) * | 2024-11-28 | 2024-12-31 | 无锡学院 | Insect detection method and system in complex background based on deep learning feature enhancement |
-
2025
- 2025-01-10 CN CN202510047791.1A patent/CN120107744A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118644761A (en) * | 2024-07-24 | 2024-09-13 | 山东滨州城建集团有限公司 | A construction site safety helmet detection method, computer equipment and storage medium |
| CN119169305A (en) * | 2024-08-28 | 2024-12-20 | 合肥工业大学 | A gas leak detection method based on infrared imaging of YOLOV8 |
| CN119206349A (en) * | 2024-09-26 | 2024-12-27 | 浙江工业大学 | A real-time detection method for laparoscopic instruments |
| CN119048496A (en) * | 2024-10-30 | 2024-11-29 | 南昌工程学院 | PCB defect detection method, system, equipment and storage medium |
| CN119152463A (en) * | 2024-11-13 | 2024-12-17 | 华南理工大学 | Method, device, equipment and medium for detecting small traffic target of high-speed running vehicle |
| CN119229476A (en) * | 2024-11-28 | 2024-12-31 | 无锡学院 | Insect detection method and system in complex background based on deep learning feature enhancement |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11487995B2 (en) | Method and apparatus for determining image quality | |
| EP3716198B1 (en) | Image reconstruction method and device | |
| WO2024040828A1 (en) | Method and device for fusion and classification of remote sensing hyperspectral image and laser radar image | |
| EP4075374A1 (en) | Image processing method and apparatus, and image processing system | |
| WO2021082480A1 (en) | Image classification method and related device | |
| CN114419413A (en) | Method for constructing sensing field self-adaptive transformer substation insulator defect detection neural network | |
| CN116664930A (en) | Personalized federated learning image classification method and system based on self-supervised contrastive learning | |
| CN108052881A (en) | The method and apparatus of multiclass entity object in a kind of real-time detection construction site image | |
| CN116012653A (en) | A hyperspectral image classification method and system based on an attention residual unit neural network | |
| CN116580184B (en) | YOLOv 7-based lightweight model | |
| CN115512226B (en) | LiDAR point cloud filtering method integrated with attention mechanism multi-scale CNN | |
| CN114581789A (en) | Hyperspectral image classification method and system | |
| CN115761268B (en) | Pole tower key part defect identification method based on local texture enhancement network | |
| CN119599946A (en) | A road crack detection method, device and electronic equipment | |
| CN118334532A (en) | Lightweight remote sensing change detection method and system based on dual-temporal remote sensing images | |
| CN118552792A (en) | Equipment detection method, system, computer storage medium and electronic equipment | |
| CN114169506A (en) | Deep learning edge computing system framework based on industrial Internet of things platform | |
| CN117237599A (en) | Image target detection method and device | |
| CN116912774A (en) | A method, electronic device and storage medium for infrared image target recognition of power transmission and transformation equipment based on edge computing | |
| CN120431068A (en) | Blood cell analysis system and method based on computer vision | |
| CN120259847A (en) | A graspable object detection method for low-quality images based on EMSDH-YOLO network | |
| CN120580200A (en) | Wire rope defect detection method and device based on improved YOLOv8n model | |
| Gong et al. | Research on mobile traffic data augmentation methods based on SA-ACGAN-GN | |
| CN120107744A (en) | A method and product for identifying the wearing of safety belts for power aerial work based on YOLOv10n | |
| CN114756425B (en) | Intelligent monitoring method, device, electronic device and computer readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |