A power pipe gallery safe operation detection method and system

Abstract

The application discloses a power pipe gallery safe operation detection method and system, and relates to the field of power pipe gallery safe operation detection. The method comprises the following steps: acquiring a current safety helmet image under a power pipe gallery; detecting the current safety helmet image by using a trained safety helmet detection model to determine a detection result; constructing a lightweight convolution module composed of a first branch and a second branch; the first branch comprises a convolution block; the second branch comprises an expansion block, a depth separable convolution block, an MIsh function and a convolution block connected in sequence; a first output end of the first branch is connected with the expansion block; the expansion block comprises three depth separable convolution blocks and an MIsh function connected in sequence; the main network and the feature fusion module are improved by using the lightweight convolution module to determine an improved YOLO_v3 model; and the trained safety helmet detection model is determined by using a safety helmet data set. The application can improve the detection efficiency under the premise of ensuring the detection accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power pipe gallery safe operation detection, in particular to a power pipe gallery safe operation detection method and system. BACKGROUND

[0002] In the construction process of the power pipe gallery, due to the weak safety awareness of the construction personnel and the lucky attitude towards wearing safety helmets, high-altitude falling accidents occur frequently, which may lead to discipline rectification, production suspension, or even personnel casualties. However, in the power pipe gallery, the construction site is wide and the number of people is large, and it is difficult to supervise the construction personnel wearing safety helmets only by manual supervision. Therefore, it is of practical significance to detect the wearing of safety helmets of construction personnel by using computer vision technology.

[0003] The safety helmet detection in the power pipe gallery has problems such as complex detection scene, dim light, and similar color of safety helmet and building materials. Many scholars at home and abroad have studied these problems. In the Chinese patent "Safety helmet detection method based on target tracking" (CN202110885467.9), Lu Jiahui et al. proposed a high-efficiency single-stage detection algorithm, which regards the worker's head and safety helmet as a single target, classifies the head with or without safety helmet, avoids the complex procedure of detecting the head first and then detecting the safety helmet, and reduces the computational complexity of the model. In the Chinese patent "Safety helmet wearing detection method and system based on improved YOLOV3" (Application No. CN202110669574.8), Li Xinhai et al. proposed an improved algorithm based on YOLO_v3, which uses a unique feature fusion strategy to detect safety helmets from four scales, effectively improving the detection accuracy. In the Chinese patent "Transformer substation worker safety helmet detection method and system based on Faster R-CNN" (CN201811237280.2), Li Haitao et al. proposed an improved algorithm based on Faster-RCNN, which adds a skip connection to extract features in the RPN layer and simultaneously punishes the gap interval to improve the detection efficiency. The above methods have good performance in detection accuracy, but as the network structure becomes more complex, the computational load and parameter quantity also increase, and it is difficult to deploy the algorithm in embedded devices with limited computing power in the power pipe gallery. Therefore, it is necessary to design a lightweight model suitable for safety helmet detection tasks in the power pipe gallery, thereby improving the detection efficiency while ensuring the detection accuracy. SUMMARY

[0004] The purpose of the present application is to provide a power pipe gallery safe operation detection method and system, which can improve the detection efficiency while ensuring the detection accuracy.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] A power pipe gallery safety operation detection method, the method comprises:

[0007] Obtain a current safety helmet image under a power pipe gallery;

[0008] Detect the current safety helmet image by using a trained safety helmet detection model, and determine a detection result;

[0009] The determination process of the trained safety helmet detection model is:

[0010] A lightweight convolution module composed of a first branch and a second branch is constructed; the first branch includes a convolution block; the second branch includes an expansion block, a depth separable convolution block, an MIsh function and a convolution block connected in sequence; the first output end of the first branch is connected with the expansion block; the output of the lightweight convolution module is the splicing of the output features of the second output end of the first branch and the output features of the second branch; the expansion block includes three depth separable convolution blocks and an MIsh function connected in sequence;

[0011] The backbone network and the feature fusion module of the YOLO_v3 model are improved by using the lightweight convolution module, and an improved YOLO_v3 model is determined;

[0012] The improved YOLO_v3 model is trained by using a safety helmet data set to determine a trained safety helmet detection model.

[0013] Optionally, the size of the convolution block in the first branch and the size of the convolution block in the second branch are both 1x1.

[0014] Optionally, the backbone network and the feature fusion module of the YOLO_v3 model are improved by using the lightweight convolution module, and an improved YOLO_v3 model is determined, specifically including:

[0015] The backbone network is determined by using a 3x3 convolution block, a stacked lightweight convolution module and a corresponding pooling layer;

[0016] The convolution block in the feature fusion module is replaced by a lightweight convolution module.

[0017] Optionally, the improved YOLO_v3 model is trained by using a safety helmet data set to determine a trained safety helmet detection model, and further includes:

[0018] A safety helmet data set is constructed.

[0019] Optionally, the safety helmet data set is constructed, specifically including:

[0020] Obtain a safety helmet image sample set;

[0021] The safety helmet image sample set is supplemented by using a synthesis method;

[0022] The supplemented safety helmet image sample set is preprocessed, and the preprocessing includes denoising and deblurring.

[0023] The safety helmet image sample set after preprocessing is labeled by using labelimg software to determine the safety helmet data set.

[0024] A power pipe gallery safety operation detection system, the system comprises:

[0025] A current safety helmet image acquisition module is configured to acquire a current safety helmet image under a power pipe gallery.

[0026] A safety helmet image detection module is configured to detect the current safety helmet image by using a trained safety helmet detection model to determine a detection result.

[0027] The determination process of the trained safety helmet detection model is as follows:

[0028] A lightweight convolution module composed of a first branch and a second branch is constructed, the first branch includes a convolution block, the second branch includes an expansion block, a depth separable convolution block, an MIsh function and a convolution block connected in sequence, a first output end of the first branch is connected with the expansion block, and the output of the lightweight convolution module is the splicing of the output features of a second output end of the first branch and the output features of the second branch, the expansion block includes three depth separable convolution blocks and an MIsh function connected in sequence.

[0029] The backbone network and the feature fusion module of the YOLO_v3 model are improved by using the lightweight convolution module to determine an improved YOLO_v3 model.

[0030] The improved YOLO_v3 model is trained by using the safety helmet data set to determine the trained safety helmet detection model.

[0031] A power pipe gallery safety operation detection system comprises at least one processor, at least one memory and computer program instructions stored in the memory, and when the computer program instructions are executed by the processor, the method is implemented.

[0032] According to the specific embodiments of the present application, the following technical effects are provided:

[0033] The power pipe gallery safe operation detection method and system provided by the application, construct a lightweight convolution module composed of a first branch and a second branch, and use the lightweight convolution module to improve the backbone network and feature fusion module of the YOLO_v3 model, replace the original convolution block, so that the parameter quantity becomes smaller, so that the running effect is faster, and the model can achieve real-time operation results. Further, while greatly reducing the model operation amount, the requirements of the task on the safety helmet detection accuracy are met. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0035] Figure 1 The power pipe gallery safe operation detection method provided by the present application is a flowchart;

[0036] Figure 2 The YOLO_v3 overall structure diagram;

[0037] Figure 3 The lightweight convolution module structure diagram;

[0038] Figure 4 The improved backbone network LW_bone structure diagram;

[0039] Figure 5 The improved feature fusion module LW_FPN structure diagram. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0041] The purpose of the present application is to provide a power pipe gallery safe operation detection method and system, which can improve the detection efficiency while ensuring the detection accuracy.

[0042] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0043] Figure 1A power pipe gallery safe operation detection method provided by the application has a flowchart as shown in Figure 1 The power pipe gallery safe operation detection method provided by the application comprises the following steps:

[0044] S101, acquiring a current safety helmet image under a power pipe gallery;

[0045] S102, detecting the current safety helmet image by using a trained safety helmet detection model to determine a detection result; the detection result is whether a construction worker on a work surface wears a safety helmet in compliance.

[0046] The determination process of the trained safety helmet detection model is as follows:

[0047] S201, constructing a lightweight convolution module (LW_Conv) composed of a first branch and a second branch, as shown in Figure 3 The first branch comprises a convolution block (Conv); the second branch comprises an expansion block (ExpansionBlock), a depth separable convolution block (DWConv), a MIsh function and a convolution block (Conv) connected in sequence; a first output end of the first branch is connected with the expansion block; the output of the lightweight convolution module is the splicing of the output features of the second output end of the first branch and the output features of the second branch; the expansion block comprises three depth separable convolution blocks and a MIsh function connected in sequence, wherein a super parameter t in the expansion block represents a feature expansion ratio, which is used to adjust the expansion number of a feature map; the size of the convolution block in the first branch and the size of the convolution block in the second branch are both 1x1. The expansion block uses the input feature map and the output feature map of the jump branch to generate an intermediate candidate feature map through depth convolution, plays the advantages of depth convolution, fully considers all available features, greatly reduces redundant calculation, and thus significantly reduces FLOPs.

[0048] As shown in Figure 3 , wherein Input refers to an input feature map, Output refers to an output result, DWConv refers to a depth separable convolution, which is a special convolution method, Conv-1x1 refers to 1x1 size convolution of data, and Mish refers to an activation function.

[0049] S202, improving the backbone network and the feature fusion module of the YOLO_v3 model by using the lightweight convolution module to determine an improved YOLO_v3 model;

[0050] As shown in Figure 2As shown, the YOLO_v3 model mainly consists of the following parts: a backbone network for extracting features from the input data, a feature fusion module (FPN), a detection head (Yolo_Head) that performs regression on the feature maps using a loss function, and the loss function YOLO_loss. The backbone network used by the YOLO_v3 algorithm is DarkNet53, and the computational load is mainly concentrated in the feature extraction and fusion stages.

[0051] S202 specifically includes:

[0052] like Figure 4 As shown, the backbone LW_bone is determined using 3×3 convolutional blocks, stacked lightweight convolutional modules, and corresponding pooling layers; Figure 4 As shown, after the first 3×3 convolution, 28 LW_Convs are stacked to extract feature maps of three scales, which significantly reduces FLOPs.

[0053] like Figure 5 As shown, a lightweight convolutional module is used to replace the convolutional blocks in the feature fusion module. The layer expansion stage expands the output size, enhancing the continuity between scales. The repetition stage uses hyperparameters to represent the number of times the blocks are stacked; blue arrows indicate downsampling, and orange arrows indicate upsampling. Research shows that large-scale feature maps contain more detailed features, such as edges and textures; while small-scale feature maps have stronger semantic information. Upsampling small-scale feature maps and then fusing them with large-scale feature maps helps detect targets of different sizes.

[0054] S203, Use the safety helmet dataset to train the improved YOLO_v3 model to determine the trained safety helmet detection model.

[0055] S203 also includes:

[0056] Construct a safety helmet dataset.

[0057] The construction of the safety helmet dataset specifically includes:

[0058] Obtain a sample set of safety helmet images;

[0059] A synthetic method was used to supplement the safety helmet image sample set;

[0060] The supplemented set of safety helmet image samples is preprocessed; the preprocessing includes denoising and deblurring.

[0061] The labelimg software was used to annotate the preprocessed safety helmet image sample set to determine the safety helmet dataset.

[0062] For the above method, the application also provides a power pipe gallery safe operation detection system, the system comprises:

[0063] A current safety helmet image acquisition module is configured to acquire a current safety helmet image under the power pipe gallery.

[0064] A safety helmet image detection module is configured to detect the current safety helmet image by using the trained safety helmet detection model to determine a detection result.

[0065] The determination process of the trained safety helmet detection model is as follows:

[0066] A lightweight convolution module composed of a first branch and a second branch is constructed, the first branch comprises a convolution block, the second branch comprises an expansion block, a depth separable convolution block, an MIsh function and a convolution block connected in sequence, a first output end of the first branch is connected with the expansion block, and the output of the lightweight convolution module is the splicing of the output features of a second output end of the first branch and the output features of the second branch; the expansion block comprises three depth separable convolution blocks and an MIsh function connected in sequence.

[0067] The main network and the feature fusion module of the YOLO_v3 model are improved by using the lightweight convolution module to determine an improved YOLO_v3 model.

[0068] The improved YOLO_v3 model is trained by using a safety helmet data set to determine the trained safety helmet detection model.

[0069] In order to perform the method corresponding to the above-mentioned embodiment one and realize the corresponding functions and technical effects, the application further provides a power pipe gallery safe operation detection system, which comprises at least one processor, at least one memory and computer program instructions stored in the memory, and the computer program instructions are executed by the processor to realize the method.

[0070] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments, and the same or similar parts of each embodiment can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0071] The principles and implementation modes of the application are described by using specific examples in this paper, and the above embodiment description is only used to help understand the method and core idea of the application; at the same time, for those skilled in the art, according to the idea of the application, the specific implementation mode and application range will be changed. In view of the above, the content of the specification should not be understood as a limitation of the application.

Claims

1. A power tube gallery safe operation detection method, characterized in that, The power pipe gallery safe operation detection method comprises: acquiring a current safety helmet image under a power pipe gallery; detecting the current safety helmet image by using a trained safety helmet detection model to determine a detection result; the determination process of the trained safety helmet detection model is as follows: a lightweight convolution module composed of a first branch and a second branch is constructed; the first branch comprises a convolution block; the second branch comprises an expansion block, a depth separable convolution block, a MIsh function and a convolution block connected in sequence; a first output end of the first branch is connected with the expansion block; the output of the lightweight convolution module is the splicing of the output features of a second output end of the first branch and the output features of the second branch; the expansion block comprises three depth separable convolution blocks and a MIsh function connected in sequence; a main network and a feature fusion module of a YOLO_v3 model are improved by using the lightweight convolution module to determine an improved YOLO_v3 model; the improved YOLO_v3 model is trained by using a safety helmet dataset to determine the trained safety helmet detection model; the sizes of the convolution blocks in the first branch and the second branch are both 1*1; the improved YOLO_v3 model is determined by improving the main network and the feature fusion module of the YOLO_v3 model by using the lightweight convolution module, and specifically comprises: a main network is determined by using a 3*3 convolution block, a stacked lightweight convolution module and a corresponding pooling layer; a convolution block in the feature fusion module is replaced by a lightweight convolution module.

2. The power tube gallery safe operation detection method according to claim 1, characterized in that, Before the improved YOLO_v3 model is trained by using the safety helmet dataset to determine the trained safety helmet detection model, the safety helmet dataset is constructed. The safety helmet dataset is constructed, and specifically comprises:

3. The power tube gallery safe operation detection method according to claim 2, characterized in that, a safety helmet image sample set is acquired; the safety helmet image sample set is supplemented by using a synthesis method; the supplemented safety helmet image sample set is preprocessed; the preprocessing comprises denoising and deblurring; the safety helmet image sample set after preprocessing is labeled by using labelimg software to determine the safety helmet dataset. The power pipe gallery safe operation detection system comprises:

4. A power tunnel safe operation detection system for implementing the power tunnel safe operation detection method according to any one of claims 1-3, characterized in that, a current safety helmet image acquisition module for acquiring a current safety helmet image under a power pipe gallery; a safety helmet image detection module for detecting the current safety helmet image by using a trained safety helmet detection model to determine a detection result; the determination process of the trained safety helmet detection model is as follows: a lightweight convolution module composed of a first branch and a second branch is constructed; the first branch comprises a convolution block; the second branch comprises an expansion block, a depth separable convolution block, a MIsh function and a convolution block connected in sequence; a first output end of the first branch is connected with the expansion block; the output of the lightweight convolution module is the splicing of the output features of a second output end of the first branch and the output features of the second branch; the expansion block comprises three depth separable convolution blocks and a MIsh function connected in sequence; a main network and a feature fusion module of a YOLO_v3 model are improved by using the lightweight convolution module to determine an improved YOLO_v3 model; ​ The improved YOLO_v3 model is trained by using a safety helmet data set to determine a trained safety helmet detection model.

5. A power utility tunnel safe operation detection system, characterized in that, Comprising: at least one processor, at least one memory, and computer program instructions stored in the memory that, when executed by the processor, implement the method of any one of claims 1-3.

Images