Lightweight fire hazard multi-target detection method in electric power environment

Abstract

The invention discloses a lightweight fire hazard multi-target detection method in an electric power environment. The method comprises the following steps: S1, constructing a lightweight fire hazard multi-target detection data set in the electric power environment; s2, a preset RT-DETR model is improved, and an improved RT-DETR model is obtained; s3, training the improved RT-DETR model by using the data set, and optimizing model parameters through an improved loss function to obtain a lightweight fire hazard multi-target detection model in the electric power environment; and S4, performing multi-target detection by using the lightweight fire hazard multi-target detection model in the electric power environment. According to the invention, high-precision disaster and hidden danger multi-target detection is realized.

Description

Technical Field

[0001] The present invention relates to the field of target detection, and in particular to a lightweight fire hazard multi-target detection method in an electric power environment. Background Art

[0002] With the rapid development of the global power industry, overhead power lines play a vital role in the power transmission and distribution network. However, the natural environment around power lines and their surroundings is complex and changeable, and there are many potential fire hazards, such as dead leaves, flammable materials, smoke, and human activities (such as smoking), which can easily cause fires. These hidden dangers not only threaten the stable operation of power lines, but in serious cases may also damage power facilities and even cause large-scale power outages, causing huge social and economic losses. Therefore, timely detection and early warning of fire hazards around power lines are of great significance to ensure the safety of the power grid and reduce the risk of fire.

[0003] Traditional fire detection methods mainly rely on sensors or surveillance cameras to identify fires that have occurred. For example, sensor-based monitoring systems determine the occurrence of fires by detecting physical features such as flames and smoke. However, these methods have obvious lags and can usually only monitor and alarm after a fire occurs, making it difficult to detect potential fire hazards in a timely manner. In addition, the detection effect of surveillance cameras is usually limited by viewing angles, lighting and weather conditions, and real-time and effectiveness are difficult to guarantee.

[0004] In recent years, with the rapid development of deep learning technology, fire detection technology has gradually introduced target detection models based on convolutional neural networks (CNN). These models can automatically extract fire-related features from video surveillance or image data to automatically identify fire areas. However, the generalization ability of CNN models is weak in complex environments, especially in the face of different lighting, weather changes and complex terrain, the detection performance is significantly reduced. In addition, as the complexity of the task increases, the computational cost and complexity of the model are also increasing. Traditional convolutional networks are difficult to deploy in real time on resource-constrained devices or low-power environments.

[0005] In order to further improve the detection speed and accuracy, the field of target detection has recently begun to adopt technologies based on the Transformer architecture. In particular, the detection method based on DETR (Detection Transformer) has demonstrated excellent end-to-end detection capabilities. DETR models global dependencies through the self-attention mechanism, effectively improving the detection accuracy and avoiding the region proposal mechanism of the traditional model. However, DETR still has limitations in small target detection, real-time performance, and computational efficiency: when dealing with small-scale targets, it is prone to missed detection or false detection problems; at the same time, the DETR model takes a long time to train and has high computational overhead, making it difficult to be widely used in real-time detection tasks.

[0006] The Chinese invention application with application number 202311243814.3 discloses "A low-complexity real-time fire and smoke detection method based on deep learning", and its technical solution includes: including collecting video image information through a camera to construct a data set, and calibrating the data set; dividing the obtained data set into a training set and a test set, and performing preprocessing operations on the data set; using the Pytorch framework to build a Focus CNN network model, using the cross entropy loss function in the training stage, and after the parameters are set, the model training is performed in the offline stage; the trained Focus CNN network is tested, evaluated and screened, and the recognition accuracy is used to measure the reasoning accuracy of the model; the optimal model is selected after the indicator evaluation, and deployed to the terminal for online forest fire and smoke detection. Its technical solution has insufficient generalization ability in different environments, especially under different weather conditions or light changes, the detection performance is significantly reduced. Summary of the invention

[0007] In order to solve the technical problems of long training time, low calculation efficiency and poor real-time performance in the prior art fire hazard detection, the present invention provides a lightweight fire hazard multi-target detection method in an electric power environment. The technical solution adopted by the present invention is:

[0008] A first aspect of the present invention provides a lightweight fire hazard multi-target detection method in an electric power environment, the method comprising:

[0009] S1: Construct a lightweight fire hazard multi-target detection dataset in the power environment;

[0010] S2: By improving the preset RT-DETR model, an improved RT-DETR model is obtained;

[0011] S3: using the data set to train the improved RT-DETR model, and optimizing the model parameters through the improved loss function, to obtain a lightweight fire hazard multi-target detection model in the power environment;

[0012] S4: Perform multi-target detection using the lightweight fire hazard multi-target detection model in the power environment.

[0013] As a preferred solution, in step S1, the method for constructing a data set includes:

[0014] A number of power work scene images were generated through the preset large model, combined with common fire hazard elements; at the same time, a number of images related to fire hazards were obtained from the Internet and accurately annotated to obtain the initial data set;

[0015] Expanding and enhancing the initial data set to obtain a lightweight fire hazard multi-target detection data set in an electric power environment;

[0016] The lightweight fire hazard multi-target detection dataset in the power environment is divided into a training set, a test set and a validation set according to a preset ratio.

[0017] As a preferred solution, in step S2, the preset RT-DETR model includes a backbone network, a hybrid encoder and a decoder;

[0018] The backbone network includes a first convolution batch normalization module, a second convolution batch normalization module, a third convolution batch normalization module, a maximization module, a first basic block, a second basic block, a third basic block, and a fourth basic block connected in sequence;

[0019] The hybrid encoder includes a same-scale feature interaction module and a cross-scale feature fusion module;

[0020] By improving the preset RT-DETR model, the method of obtaining the improved RT-DETR model includes:

[0021] Replacing the second basic block, the third basic block, and the fourth basic block with the first partial convolution module, the second partial convolution module, and the third partial convolution module, respectively, to obtain an improved backbone network;

[0022] The SimAM attention mechanism is introduced into the same-scale feature interaction module to obtain an improved hybrid encoder;

[0023] The Inner-GIoU loss function is introduced into the RT-DETR model to obtain an improved RT-DETR model.

[0024] As a preferred solution, the number of floating-point operations per second of the first partial convolution module, the second partial convolution module, and the third partial convolution module is:

[0025]

[0026] Among them, h and w represent the dimensions of the feature map, k represents the convolution kernel size, and c p Indicates the number of channels used by some convolution modules.

[0027] As a preferred solution, the SimAM attention mechanism is used to calculate the energy function of neurons to determine the importance of each feature. The specific calculation formula is:

[0028]

[0029] Among them, the calculation formula of neuron energy E is:

[0030]

[0031]

[0032]

[0033] Among them, E is the energy function of the neuron, μ is the average value of each channel in the input feature map, and σ 2 is the variance, and X is the original feature map.

[0034] As a preferred solution, the Inner-GIoU loss function introduces a multi-scale auxiliary bounding box in the regression process, and calculates the loss by controlling the scale ratio of the auxiliary bounding box, which is defined as follows:

[0035]

[0036] union=(w gt *h gt )*(ratio) 2 +(w*h)*(ratio) 2 -inter

[0037]

[0038]

[0039]

[0040] Among them, b gt and b represent the real box and anchor box respectively, w gt and h gt represents the width and height of the real box, w and h represent the width and height of the anchor box, ratio is the auxiliary factor that controls the size of the auxiliary box, IoU represents the intersection over union ratio, A c It represents the area of ​​the smallest enclosing rectangle containing the predicted box and the true box, and U represents the union area.

[0041] As a preferred solution, in step S3, the improved RT-DETR model is trained using the data set, and the model parameters are optimized by an improved loss function to obtain a lightweight fire hazard multi-target detection model in an electric power environment. The method includes:

[0042] The improved RT-DETR model is trained using the training set, and then the model performance is evaluated using the test set. Then, the model with the best performance is retained, and the model parameters are optimized using the validation set and an improved loss function to obtain a lightweight fire hazard multi-target detection model in an electric power environment.

[0043] As a preferred solution, the common fire hazard elements include at least one of the following:

[0044] Smoking, dead leaves, smoke.

[0045] A second aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned lightweight fire hazard multi-target detection method in an electric power environment.

[0046] The third aspect of the present invention provides a computer device, including a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor. When the computer program is executed by the processor, the steps of the aforementioned lightweight fire hazard multi-target detection method in an electric power environment are implemented.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The present invention introduces some convolution modules into the backbone network of the RT-DETR model to replace some traditional basic block convolution operations, which significantly reduces the computational burden and memory consumption and improves the detection capability of small target fire hazards.

[0049] The present invention introduces the SimAM attention mechanism into the same-scale feature interaction module to form a sim-AIFI module, thereby enhancing the adaptability of the model to complex power environments and reducing missed detections and false detections.

[0050] By combining the Inner-IoU and GIoU loss functions, the present invention designs the Inner-GIoU loss with auxiliary bounding box control, which significantly improves the convergence speed and detection accuracy of the model and is suitable for precise positioning of multi-scale targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A flow chart of a lightweight fire hazard multi-target detection method in an electric power environment provided in this embodiment;

[0052] Figure 2 The RT-DETR model structure diagram provided for this embodiment;

[0053] Figure 3 The structure diagram of the improved RT-DETR model provided in this embodiment;

[0054] Figure 4 The PConv structure schematic diagram provided for this embodiment;

[0055] Figure 5 This is the SimAM structure diagram provided for this embodiment. DETAILED DESCRIPTION

[0056] The accompanying drawings are only used for illustrative purposes and are not to be construed as limiting the present invention;

[0057] It should be clear that the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the embodiments of the present application.

[0058] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0059] When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0060] In addition, in the description of this application, unless otherwise specified, "multiple" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0061] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0062] Example 1

[0063] Please refer to Figure 1 This embodiment provides a lightweight fire hazard multi-target detection method in an electric power environment, the method comprising:

[0064] S1: Construct a lightweight fire hazard multi-target detection dataset in the power environment;

[0065] S2: By improving the preset RT-DETR model, an improved RT-DETR model is obtained;

[0066] S3: using the data set to train the improved RT-DETR model, and optimizing the model parameters through the improved loss function, to obtain a lightweight fire hazard multi-target detection model in the power environment;

[0067] S4: Perform multi-target detection using the lightweight fire hazard multi-target detection model in the power environment.

[0068] In a specific embodiment, in step S1, the method for constructing a data set includes:

[0069] A number of power work scene images were generated through the preset large model, combined with common fire hazard elements; at the same time, a number of images related to fire hazards were obtained from the Internet and accurately annotated to obtain the initial data set;

[0070] Expanding and enhancing the initial data set to obtain a lightweight fire hazard multi-target detection data set in an electric power environment;

[0071] The lightweight fire hazard multi-target detection dataset in the power environment is divided into a training set, a test set and a validation set according to a preset ratio.

[0072] Specifically, the lightweight fire hazard multi-target detection dataset in the power environment covers a variety of power scene elements, including power facilities, staff, smoking, dead leaves and smoke, etc., truly restoring the complexity of the power work site to enhance the generalization and small target detection capabilities of the model.

[0073] Specifically, the method of constructing the dataset includes:

[0074] Scene image generation and collection: 532 power work scene images were generated through the large model, combined with common fire hazard elements such as smoking, dead leaves and smoke. At the same time, 2,000 images related to fire hazards were obtained from the Internet and accurately labeled;

[0075] Data expansion and enhancement: Using rotation, symmetry and other enhancement techniques, the number of images was expanded to 4,539, which increased the scene diversity and improved the robustness of the model under different perspectives and conditions.

[0076] Dataset division: The dataset is divided into a training set of 3177 images, a test set of 681 images, and a validation set of 681 images, which are used for model training, performance evaluation, and parameter optimization, respectively.

[0077] It should be noted that the present invention uses RT-DETR as the basic model. RT-DETR (Real-Time Detection Transformer) is a new real-time end-to-end object detection model. Through innovative architecture design, the model significantly shortens the training time while maintaining high detection accuracy, and shows strong potential in industrial applications. Its architecture is mainly divided into three core parts: backbone network, hybrid encoder and decoder.

[0078] Backbone network: The backbone network of RT-DETR is based on a convolutional neural network (CNN), which is responsible for extracting multi-scale features from the input image. Through feature maps of three different step sizes (8, 16, and 32), the backbone network can effectively capture the global and local features of the target. In order to improve detection efficiency, the neck network of RT-DETR introduces an attention mechanism and uses the same-scale feature interaction module (AIFI) to improve performance while reducing the amount of computation. The cross-scale feature fusion module (CCFM) enhances the model's detection capabilities for targets of various sizes by integrating features of different scales.

[0079] Hybrid encoder: RT-DETR's hybrid encoder processes input features through the Transformer architecture and applies a Transformer-based denoising mechanism, which improves the quality of sample matching and significantly accelerates the training convergence speed. The encoder also dynamically adjusts the IoU threshold, allowing the model to adaptively detect key areas and improve the robustness of detection.

[0080] Decoder: The decoder part continues to use the Transformer-based design and uses the IoU-aware mechanism to obtain the initial target query. Through multiple iterative optimizations, this mechanism improves the prediction accuracy of key areas and improves detection accuracy.

[0081] In a specific embodiment, please refer to Figure 2 , in step S2, the preset RT-DETR model includes a backbone network, a hybrid encoder and a decoder;

[0082] The backbone network includes a first convolution batch normalization module, a second convolution batch normalization module, a third convolution batch normalization module, a maximization module, a first basic block, a second basic block, a third basic block, and a fourth basic block connected in sequence;

[0083] The hybrid encoder includes a same-scale feature interaction module and a cross-scale feature fusion module;

[0084] By improving the preset RT-DETR model, the method of obtaining the improved RT-DETR model includes:

[0085] Please refer to Figure 3 , the second basic block, the third basic block, and the fourth basic block are replaced by the first partial convolution module, the second partial convolution module, and the third partial convolution module, respectively, to obtain an improved backbone network; the improved backbone network significantly reduces the amount of calculation and memory consumption, and enhances the detection capability of small targets.

[0086] It should be noted that the present invention selects the relatively lightweight ResNet-18 as the basic backbone network of the RT-DETR model. In order to solve the problem of computational redundancy in feature extraction of traditional convolutional neural networks and improve the speed and accuracy of multi-target detection, the present invention introduces a partial convolution module PConv-Block in the backbone network. This module aims to reduce the computational burden and improve the detection performance of small targets by introducing PConv convolution. PConv only uses filters on a selected subset of input channels and retains the rest, ultimately resulting in lower floating-point operations per second (FLOPs) compared to standard convolution. This method ensures that the speed of multi-target detection in complex environments is improved without affecting the accuracy of recognition. The structural principle of PConv is as follows Figure 4 Shown

[0087] The SimAM attention mechanism is introduced into the same-scale feature interaction module, and the sim-AIFI module is constructed to obtain an improved hybrid encoder; this module can capture local details and global dependencies in feature maps in complex scenarios, effectively reduce missed detections and false detections, and thus improve the robustness of the model.

[0088] The Inner-GIoU loss function is introduced into the RT-DETR model to obtain an improved RT-DETR model;

[0089] In a specific embodiment, the number of floating-point operations per second of the first partial convolution module, the second partial convolution module, and the third partial convolution module is:

[0090]

[0091] Among them, h and w represent the dimensions of the feature map, k represents the convolution kernel size, and c p Indicates the number of channels used by some convolution modules;

[0092] It should be noted that PConv also has a small amount of memory access, namely:

[0093]

[0094] Because it only convolves a part of the input channels, the rest c~c pChannels do not participate in the calculation. Integrating PConv into the feature extraction network greatly reduces the computation and parameter load, thereby improving the inference speed of the model.

[0095] It should be noted that in order to improve the feature extraction capability of the model under limited computing resources, so that it can focus on key features and ignore irrelevant information, the present invention introduces the SimAM (Simultaneous Attention Mechanism) attention mechanism into the AIFI module of the RT-DETR model. SimAM is a lightweight general attention module. Unlike traditional channel or spatial attention mechanisms, SimAM does not introduce additional network parameters and has plug-and-play characteristics. It can significantly improve the model's ability to focus on important targets without increasing computing costs. Its structure is as follows Figure 5 shown.

[0096] In a specific embodiment, the SimAM attention mechanism is used to calculate the energy function of the neuron to determine the importance of each feature. The specific calculation formula is:

[0097]

[0098] Among them, the calculation formula of neuron energy E is:

[0099]

[0100]

[0101]

[0102] Among them, E is the energy function of the neuron, μ is the average value of each channel in the input feature map, and σ 2 is the variance, X is the original feature map; the energy function E of the neuron compares the difference between the features of the current neuron and the surrounding neurons. When E is lower and the difference with the surrounding neurons is greater, the importance of the neuron is higher. The hyperparameter λ is used to adjust the influence of the target neuron. In order to prevent the E value from being too large, the sigmoid function is used to limit its size. Enhanced feature map It is the result of dot multiplication of the original feature map X and the neuron energy E processed by the sigmoid function.

[0103] It should be noted that in order to further optimize the bounding box regression accuracy of the model and improve the detection performance, the present invention introduces the Inner-GIoU (Inner Generalized Intersection over Union) loss function. Although the traditional IoU loss has a certain effect in accelerating the convergence speed, it performs poorly when dealing with bounding box regression under multiple targets and complex backgrounds, which can easily lead to missed detections and false detections.

[0104] In a specific embodiment, the Inner-GIoU loss function introduces a multi-scale auxiliary bounding box in the regression process, and calculates the loss by controlling the scale ratio of the auxiliary bounding box, which is defined as follows:

[0105]

[0106] union=(w gt *h gt )*(ratio) 2 +(w*h)*(ratio) 2 -inter

[0107]

[0108]

[0109]

[0110] Among them, b gt and b represent the real box and anchor box respectively, w gt and h gt represents the width and height of the real box, w and h represent the width and height of the anchor box, ratio is the auxiliary factor that controls the size of the auxiliary box, IoU represents the intersection over union ratio, A c It represents the area of ​​the smallest enclosing rectangle containing the predicted box and the true box, and U represents the union area.

[0111] The present invention combines Inner-IoU with GIoU and introduces an auxiliary bounding box controlled by a scale factor into GIoU to form Inner-GioU. This improvement enables the model to adaptively detect the target area and significantly improves the convergence speed and detection accuracy.

[0112] In a specific embodiment, in step S3, the improved RT-DETR model is trained using the data set, and the model parameters are optimized by an improved loss function to obtain a lightweight fire hazard multi-target detection model in an electric power environment. The method includes:

[0113] The improved RT-DETR model is trained using the training set, and then the model performance is evaluated using the test set. Then, the model with the best performance is retained, and the model parameters are optimized using the validation set and an improved loss function to obtain a lightweight fire hazard multi-target detection model in an electric power environment.

[0114] In a specific embodiment, the common fire hazard elements include at least one of the following:

[0115] Smoking, dead leaves, smoke.

[0116] Example 2

[0117] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of a lightweight fire hazard multi-target detection method in an electric power environment described in Embodiment 1 are implemented.

[0118] Example 3

[0119] This embodiment provides a computer device, including a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor. When the computer program is executed by the processor, the steps of a lightweight fire hazard multi-target detection method in an electric power environment described in Example 1 are implemented.

[0120] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A lightweight multi-target detection method for fire hazards in an electric power environment, characterized in that: The method comprises: S1: Construct a lightweight fire hazard multi-target detection dataset in the power environment; S2: By improving the preset RT-DETR model, an improved RT-DETR model is obtained; S3: using the data set to train the improved RT-DETR model, and optimizing the model parameters through the improved loss function, to obtain a lightweight fire hazard multi-target detection model in the power environment; S4: Perform multi-target detection using the lightweight fire hazard multi-target detection model in the power environment.

2. According to the method for multi-target detection of lightweight fire hazards in an electric power environment in claim 1, it is characterized in that: In step S1, the method for constructing a data set includes: A number of power work scene images were generated through the preset large model, combined with common fire hazard elements; at the same time, a number of images related to fire hazards were obtained from the Internet and accurately annotated to obtain the initial data set; Expanding and enhancing the initial data set to obtain a lightweight fire hazard multi-target detection data set in an electric power environment; The lightweight fire hazard multi-target detection dataset in the power environment is divided into a training set, a test set and a validation set according to a preset ratio.

3. According to the method for multi-target detection of lightweight fire hazards in an electric power environment in claim 1, it is characterized in that: In step S2, the preset RT-DETR model includes a backbone network, a hybrid encoder and a decoder; The backbone network includes a first convolution batch normalization module, a second convolution batch normalization module, a third convolution batch normalization module, a maximization module, a first basic block, a second basic block, a third basic block, and a fourth basic block connected in sequence; The hybrid encoder includes a same-scale feature interaction module and a cross-scale feature fusion module; By improving the preset RT-DETR model, the method of obtaining the improved RT-DETR model includes: Replacing the second basic block, the third basic block, and the fourth basic block with the first partial convolution module, the second partial convolution module, and the third partial convolution module, respectively, to obtain an improved backbone network; The SimAM attention mechanism is introduced into the same-scale feature interaction module to obtain an improved hybrid encoder; The Inner-GIoU loss function is introduced into the RT-DETR model to obtain an improved RT-DETR model.

4. The lightweight fire hazard multi-target detection method in an electric power environment according to claim 3 is characterized in that: The number of floating point operations per second of the first partial convolution module, the second partial convolution module, and the third partial convolution module is: Among them, h and w represent the dimensions of the feature map, k represents the convolution kernel size, and c p Indicates the number of channels used by some convolution modules.

5. The lightweight fire hazard multi-target detection method in an electric power environment according to claim 3 is characterized in that: The SimAM attention mechanism is used to calculate the energy function of neurons to determine the importance of each feature. The specific calculation formula is: Among them, the calculation formula of neuron energy E is: Among them, E is the energy function of the neuron, μ is the average value of each channel in the input feature map, and σ 2 is the variance, and X is the original feature map.

6. The lightweight fire hazard multi-target detection method in an electric power environment according to claim 3 is characterized in that: The Inner-GIoU loss function introduces a multi-scale auxiliary bounding box in the regression process and calculates the loss by controlling the scale ratio of the auxiliary bounding box. It is defined as follows: union=(w gt *h gt )*(ratio) 2 +(w*h)*(ratio) 2 -inter Among them, b gt and b represent the real box and anchor box respectively, w gt and h gt represents the width and height of the real box, w and h represent the width and height of the anchor box, ratio is the auxiliary factor that controls the size of the auxiliary box, IoU represents the intersection over union ratio, A c It represents the area of ​​the smallest enclosing rectangle containing the predicted box and the true box, and U represents the union area.

7. The method for detecting lightweight fire hazards in a power environment according to claim 2, characterized in that: In step S3, the method of using the data set to train the improved RT-DETR model and optimizing the model parameters by an improved loss function to obtain a lightweight fire hazard multi-target detection model in an electric power environment includes: The improved RT-DETR model is trained using the training set, and then the model performance is evaluated using the test set. Then, the model with the best performance is retained, and the model parameters are optimized using the validation set and an improved loss function to obtain a lightweight fire hazard multi-target detection model in an electric power environment.

8. The lightweight fire hazard multi-target detection method in an electric power environment according to claim 2 is characterized in that: The common fire hazard elements include at least one of the following: Smoking, dead leaves, smoke.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a lightweight fire hazard multi-target detection method in an electric power environment as described in any one of claims 1 to 8 are implemented.

10. A computer device, characterized in that: It includes a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor, and when the computer program is executed by the processor, the steps of a lightweight fire hazard multi-target detection method in an electric power environment as described in any one of claims 1 to 8 are implemented.

Images