Search

CN-115424063-B - YOLO-MobileNet-based wearing state detection method for safety protection equipment of electric power operation site

CN115424063BCN 115424063 BCN115424063 BCN 115424063BCN-115424063-B

Abstract

The invention belongs to the technical field of target detection, and particularly discloses a method for detecting the wearing state of safety protection equipment in an electric power operation site based on YOLO-MobileNet. The invention improves the bottleneck module forming MobileNetv < 2 >, utilizes the improved bottleneck module and the original bottleneck module to construct an improved MobileNetv < 2 > feature extraction network, replaces the main feature extraction network CSPDARKNET53 of YOLOv < 4 >, replaces the YOLOv < 4 > part of standard convolution network with a deep separable convolution network, reduces the detection precision of the safety helmet from 97.5% to 95.0%, improves the detection speed from 22 frames/second to 65 frames/second, and can completely meet the requirement of real-time detection.

Inventors

  • Guo Tiebin
  • LI DA
  • SHI GENHUA
  • ZHANG NAN
  • Diao Naixun
  • GUAN XIAOZHUO
  • LU YIXIN
  • ZHANG JIAXING
  • LI WEI
  • DING WEI
  • YANG SHUO
  • GAO YE
  • FU RAO

Assignees

  • 国网吉林省电力有限公司吉林供电公司

Dates

Publication Date
20260505
Application Date
20220831

Claims (6)

  1. 1. A method for detecting the wearing state of safety protection equipment on an electric power operation site based on YOLO-MobileNet is characterized by comprising the following specific steps: Step 1), constructing a data set and labeling images of the data set; step 2) constructing a safety protection equipment wearing state detection network based on YOLO-MobileNet: (1) Constructing MobileNetv bottleneck modules containing a channel attention network: ① MobileNetv2 the bottleneck module consists of three parts: The first part consists of an expansion layer, a batch normalization layer and an activation function layer, wherein the activation function of the activation function layer is a ReLU6 activation function; the second part comprises a depth separable convolution with a convolution kernel of 3, a batch normalization layer and an activation function layer, wherein the activation function of the activation function layer is a ReLU6 activation function; the third part consists of a linear layer and a batch normalization layer; ② Adding a channel attention network between the batch normalization layer and the activation function layer of the second part, thereby constructing a MobileNetv bottleneck module containing the channel attention network; (2) Construction of an improved MobileNetv network: ① In the MobileNetv bottleneck module containing the channel attention network constructed in step 2) of (1), replacing the ReLU6 activation function of the second part in the MobileNetv bottleneck module containing the channel attention network with the hard-Swish function, thereby constructing an improved MobileNetv2 network; ② Alternately using the improved bottleneck module and the original bottleneck module at different stages of the improved MobileNetv network, and adjusting the number of the bottleneck modules according to the actual test result; ③ The number of the bottleneck modules after adjustment is 1,2,3,4,3,2,1 in sequence; (3) Replacing the YOLOv backbone feature extraction network CPSDARKNET with the improved MobileNetv network constructed in (2), thereby constructing a YOLO-MobileNet-based safety equipment wearing state detection network; Step 3) training and testing a network model: and training and testing the network model by using the constructed network model training data set and the network model testing data set.
  2. 2. The method for detecting the wearing state of safety equipment on an electric power operation site based on YOLO-MobileNet according to claim 1, wherein the step 1) of constructing a dataset and labeling the dataset image includes the steps of: (1) Video acquisition is carried out on the electric power operation site by utilizing video acquisition equipment; (2) Screening the collected video images; (3) And forming the screened images into a network model training data set and a network model testing data set, and respectively labeling.
  3. 3. The method for detecting the wearing state of safety protection equipment for an electric power operation site based on YOLO-MobileNet according to claim 2, wherein the method for screening the collected video image is specifically as follows: ① Selecting images of electric power operators in different scenes to be grouped, wherein the images with safety protection equipment are worn as a first group, and the images without safety protection equipment are not worn as a second group; ② Dividing the first group of images into a training group and a testing group, and dividing the second group of images into a training group and a testing group; ③ Forming a training data set of the network model by the training one group of images and the training two groups of images, and forming a testing data set of the network model by the training two groups of images and the testing two groups of images; ④ The division ratio of the first group of images is the same as the division ratio of the second group of images.
  4. 4. The YOLO-MobileNet-based power operation site safety equipment wear state detection method of claim 1, wherein (3) of step 2) represents the constructed modified MobileNetv network instead of YOLOv trunk feature extraction network CPSDARKNET as: Replacing YOLOv main feature extraction network CPSDARKNET with the improved MobileNetv network constructed in step 2), obtaining input features of three detection heads through a path aggregation network, and replacing standard convolution connected with input and output of the spatial pyramid network with depth separable convolution.
  5. 5. The method for detecting the wearing state of safety equipment on an electric power operation site based on YOLO-MobileNet according to claim 4, wherein the input features of the three detection heads are a shallow input feature, a middle input feature and a deep input feature respectively, wherein: In the improved MobileNetv network, the output features of the last layer of bottleneck layer with the size of 1/8 of the input image size are taken as the shallow input features of the YOLOv path aggregation network Path Aggregation Network, PANet; In the improved MobileNetv network, the output characteristics of the last layer of the bottleneck layer with the size of the output characteristic diagram being 1/16 of the size of the input image are taken as the middle-layer input characteristics of the YOLOv path aggregation network; in the modified MobileNetv network, the output features of the last layer of bottleneck layer with the size of 1/32 of the input image size are taken as the input of the YOLOv spatial pyramid pooling network SPATIAL PYRAMID Pooling Networks, SPPNet, and the output of the spatial pyramid pooling network is taken as the deep input features of the YOLOv path aggregation network.
  6. 6. The method for detecting the wearing state of safety equipment on an electric power operation site based on YOLO-MobileNet as claimed in claim 1, wherein the step 3) of training and testing the network model is specifically as follows: ① Network model training In the model training process, the total iteration number is 300, the minimum batch is 15, the initial learning rate is 0.06, the cosine annealing learning rate strategy is adopted, the network parameters are optimized by using a random gradient descent method, the momentum of a random gradient descent optimizer is 0.6, and the weight attenuation is 0.0004; ② Testing And testing the network model which is trained by ① by using a test set, testing whether the network model can meet the requirements, if so, ending the training, and if not, adjusting parameters to train and test the model again.

Description

YOLO-MobileNet-based wearing state detection method for safety protection equipment of electric power operation site Technical Field The invention belongs to the technical field of target detection, and relates to a method for detecting the wearing state of safety protection equipment of an electric power operation site based on YOLO-MobileNet, which is used for detecting the wearing state of the safety protection equipment of an operator of the electric power operation site, is particularly suitable for detecting the wearing state of a safety helmet of the operator of the electric power operation site, and monitors the wearing condition of the safety helmet of the operator. Background Safety protection equipment, particularly safety helmets, are important safety protection measures in electric power operation and are important for guaranteeing personal safety in electric power operation. The safety helmet can protect the head of the power operator and avoid electric shock and injury of the head. In electric power operation, when a worker is impacted by falling objects, the hat shell and the hat lining of the safety helmet decompose impact force to the whole area of the cranium in the moment, and then the elastic deformation, plastic deformation and allowable structural damage of the structures and materials of the hat shell and the hat lining and the arranged buffer structures (sockets, ropes, sutures, buffer cushions and the like) are utilized to absorb most of the impact force, so that the impact force finally applied to the head of the worker is reduced to below 4900N, and the head of the worker is protected from injury or injury. In the electric power operation process, due to the fact that the safety consciousness of part of operators is light and the safety consciousness of the operators is high, the safety protection equipment of part of the electric power operators is often incomplete to wear, electric power operation is particularly easy to carry out under the condition that safety helmets are not worn, and great potential safety hazards are brought to electric power safety production and personal safety of the operators. The traditional method for wearing safety protection equipment of operators, in particular to a method for wearing and supervising safety helmets, requires site management staff to supervise the site through eyes, wastes a great deal of manpower resources, and greatly influences the supervision effect by factors such as responsibility of the supervision staff. In order to improve efficiency of wearing safety protection equipment, particularly safety helmet wearing detection of electric power operators, a detection method based on remote video analysis is adopted at present. According to the method, video acquisition is carried out on an electric power operation site by using video acquisition equipment such as a distribution control ball and the like, the acquired video is transmitted to a remote server, and the server analyzes the video by using an artificial intelligence method, so that detection of safety helmets of electric power operators is realized. The existing method for detecting the safety helmet of the electric power operator mainly comprises a YOLO-based method and an R-CNN-based method, wherein the YOLO-based method is used for real-time detection, the detection speed is relatively high, the detection accuracy is better, but the detection speed is low, and the method is not suitable for real-time detection. Although the detection speed of the detection method based on the YOLO is relatively high, the detection speed still needs to be further improved in the actual application process because of more operation sites, more videos need to be analyzed and limited calculation capacity of a server, so that the detection instantaneity is not good enough, and the real-time requirement of detection is met. Therefore, the complexity of the YOLO method needs to be reduced under the condition of meeting the detection precision requirement, so as to improve the detection speed of the wearing state of the safety protection equipment of the electric power operation site based on artificial intelligence and meet the real-time requirement of detection. Disclosure of Invention The invention aims to overcome the defects of the prior art, and provides the method for detecting the wearing state of the safety protection equipment of the electric power operation site based on the YOLO-MobileNet, which reduces the complexity of the YOLO method, improves the detection speed of the wearing state of the safety protection equipment of the electric power operation site based on artificial intelligence and meets the real-time requirement of detection under the condition of meeting the detection precision requirement. The technical scheme for solving the technical problems is that the method for detecting the wearing state of the safety protection equipment of the electric power operation site ba