Lightweight forest fire detection model design method and system based on auxiliary training branch

Abstract

The invention belongs to the technical field of deep learning, and discloses an auxiliary training branch-based lightweight forest fire detection model design method and system, and the method comprises the steps: collecting forest fire image samples through an inspection unmanned plane and a monitoring holder, and constructing a forest fire data set; a Mosaic-4 data enhancement method is used for the collected mountain fire data set; preprocessing the collected complete data set by adopting a histogram equalization algorithm, and marking a preprocessed image sample; dividing the marked image samples into a training set and a test set in proportion; designing a lightweight convolutional neural network main body model structure; and training the lightweight forest fire detection model based on the auxiliary training branches by using the forest fire data set, and finally obtaining a lightweight forest fire detection model. According to the method, a lightweight convolutional network design and auxiliary training branch design method and a Mosaic-4 data enhancement and histogram equalization algorithm are combined, so that the detection speed is greatly improved under the condition of ensuring the detection accuracy.

Description

Technical Field

[0001] The present invention belongs to but is not limited to the field of deep learning technology, and in particular relates to a lightweight wildfire detection model design method and system based on auxiliary training branches. Background Art

[0002] Wildfires are extremely destructive natural disasters that can destroy large tracts of forests, wildlife habitats, and endanger human life, property, and the environment. Edge devices such as monitoring PTZs and inspection drones are the current mainstream edge devices for wildfire monitoring. They can capture the source of fire through visible light and thermal imaging technology and push warning information in a timely manner. With the widespread application of target detection algorithms in the field of wildfire detection, the demand for deploying efficient and accurate wildfire detection models on edge devices is increasing. Currently, most of the more widely used wildfire detection methods rely on deep learning models with complex structures and large numbers of parameters. These models face challenges such as limited computing resources and storage capacity when deployed on edge devices. Therefore, designing a lightweight wildfire detection model has become an urgent need.

[0003] The existing methods for detecting wildfires are as follows: At present, the main methods of wildfire detection include detection based on multi-sensor fusion, detection based on satellite remote sensing, and detection based on computer vision. Among them, the detection range of detection based on multi-sensor fusion is small and the sensor maintenance is difficult. The detection technology based on satellite remote sensing is easily affected by weather conditions and has a high startup cost. Thanks to the rapid development of electronic and image technology, the wildfire detection method based on deep learning has become a reality, providing a low-cost, safe and convenient solution for wildfire monitoring and detection. It can not only be used for ground-based monitoring PTZ, but also for aerial monitoring through drones. The forest fire detection algorithm based on deep learning is trained on a large amount of data to establish a fire detection model. The wildfire detection method based on deep learning obtains real-time forest images through drone inspections, tower video monitoring, etc., and detects forest fires by identifying flames and smoke in the images. It has the characteristics of flexible deployment and wide coverage, and is suitable for use in open and large-space forest environments.

[0004] However, traditional deep learning models require a lot of computing resources when running on edge devices, resulting in slow response speed and low computing efficiency; secondly, large models have a large number of parameters and computational complexity, which exceeds the computing processing power and storage capacity of edge devices, greatly limiting the deployable devices of the model; most importantly, wildfire detection usually requires real-time and rapid response so that timely measures can be taken, and the complexity of traditional models often leads to high detection delays, which cannot meet the requirements of real-time wildfire detection. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a lightweight wildfire detection model design method and system based on auxiliary training branches.

[0006] The present invention is implemented as follows: a lightweight wildfire detection model design method based on auxiliary training branches, the method comprising: S1: Wildfire image samples are collected by inspection drones and monitoring gimbals to build a wildfire dataset.

[0007] S2: The Mosaic-4 data augmentation method is used on the collected wildfire dataset, that is, 4 pictures are used and spliced ​​into one picture by random cropping and random arrangement. At the same time, some Gaussian noise is reasonably introduced to enhance the robustness of the model.

[0008] S3: The collected complete data set is preprocessed using the histogram equalization algorithm, and the preprocessed image samples are labeled.

[0009] S4: Divide the labeled image samples into training set and test set in proportion.

[0010] S5: Design a lightweight convolutional neural network main model structure. The designed lightweight convolutional neural network model compresses the model structure size; the proposed lightweight detection head using shared convolution can use shared convolution to realize the sharing of some convolution parameters, thereby improving the detection efficiency of the model.

[0011] S6: Design an auxiliary training branch structure, which can aggregate the gradient information of all target objects and retain the information of deep features, so as to obtain reliable gradients and alleviate the problem of information loss in deep supervision.

[0012] S7: Use the wildfire dataset to train a lightweight wildfire detection model based on auxiliary training branches, remove the auxiliary training branch structure in the trained model, and finally obtain a lightweight wildfire detection model.

[0013] Further, in said S5, the structure of the designed lightweight aircraft door detection model includes: (1) The real-time target detection network YOLOv8n is used to build the backbone network to improve the real-time detection speed of the model.

[0014] (2) A partial convolution operator is used to construct the feature extraction and fusion module of the neck network to improve the computational efficiency of the model.

[0015] (3) A shared convolution kernel is used to construct the output detection head. The number of parameters of the detection head is reduced by sharing parameters between modules, which improves the real-time detection speed of the model and realizes the identification and location of wildfires.

[0016] Furthermore, in S6, the structure of the auxiliary training branch designed includes: (1) A normalization layer consisting of a convolutional layer, a batch normalization layer, and an activation function is used to normalize the original image and the feature map input through the backbone network, standardize the output of the convolutional layer, speed up the training speed, and improve the model stability.

[0017] (2) A deformable convolution operator is used to construct a feature fusion module. By merging feature maps at different levels and utilizing their complementary information, the model’s ability to recognize targets is enhanced.

[0018] (3) The merged feature map is passed to the output detection head to provide additional auxiliary information for wildfire detection during model training and improve the detection accuracy of the model.

[0019] Furthermore, in S7, the lightweight wildfire detection model based on the auxiliary training branch includes: (1) The main structure consisting of the backbone network, neck network and output detection head is responsible for processing and detecting the input video images to realize the identification and location of wildfires.

[0020] (2) The auxiliary training branch consisting of the normalization layer and the feature fusion module is responsible for providing additional auxiliary information during model training, assisting the main structure in learning the features of image samples and improving the detection accuracy of the model.

[0021] (3) The auxiliary training branches will be removed after the model training is completed. Finally, the model will only be left with the main structure consisting of the backbone network, the neck network and the output detection head. This will reduce the number of model parameters and calculations and improve the real-time detection speed of the model without reducing the detection accuracy of the model.

[0022] Another object of the present invention is to provide a lightweight mountain fire detection model design system based on auxiliary training branches based on the lightweight mountain fire detection model design method based on auxiliary training branches, and the system specifically includes: The dataset construction module uses patrol drones and monitoring gimbals to collect wildfire image samples and construct a wildfire dataset.

[0023] The data enhancement module is connected to the dataset construction module and uses the Mosaic-4 data enhancement method on the collected wildfire dataset. That is, 4 pictures are used and spliced ​​into one picture by random cropping and random arrangement. At the same time, some Gaussian noise is reasonably introduced to enhance the robustness of the model.

[0024] The preprocessing module, connected with the data enhancement module, preprocesses the collected complete data set using the histogram equalization algorithm and labels the preprocessed image samples.

[0025] The partitioning module is connected to the preprocessing module to divide the labeled image samples into training sets and test sets in proportion.

[0026] The structural design module is connected to the division module to design the main model structure of the lightweight convolutional neural network. The designed lightweight convolutional neural network model compresses the model structure size; the proposed lightweight detection head using shared convolution can use shared convolution to realize the sharing of some convolution parameters, thereby improving the detection efficiency of the model.

[0027] The branch structure design module is connected to the structure design module to design an auxiliary training branch structure. The designed auxiliary training branch structure can aggregate the gradient information of all target objects and retain the information of deep features, so as to obtain reliable gradients and alleviate the problem of information loss in deep supervision.

[0028] The model training module is connected to the branch structure design module, and uses the wildfire dataset to train the lightweight wildfire detection model based on the auxiliary training branch, removes the auxiliary training branch structure in the trained model, and finally obtains the lightweight wildfire detection model.

[0029] Another object of the present invention is to provide a computer device, characterized in that the computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for designing a lightweight wildfire detection model based on an auxiliary training branch.

[0030] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the method for designing a lightweight wildfire detection model based on an auxiliary training branch.

[0031] Another object of the present invention is to provide an information data processing terminal, which is used to implement the lightweight wildfire detection model design system based on auxiliary training branches.

[0032] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows: First, the present invention designs a lightweight wildfire detection model design method based on auxiliary training branches. The model combines lightweight model design methods such as lightweight convolutional network design and auxiliary training branch design, and introduces Mosaic-4 data enhancement and histogram equalization algorithms to greatly improve the detection speed while ensuring the detection accuracy; the present invention focuses on the real-time detection of the model, so that wildfire detection can achieve fast and real-time response on edge devices to meet the real-time task requirements of wildfire detection, so that the prevention and control departments can take measures in time; the lightweight model design method and model-assisted training branches adopted by the present invention can greatly reduce the model parameter amount and calculation amount, so that it can be effectively deployed on edge devices, ensuring that wildfire detection tasks can be efficiently performed on wildfire detection edge devices.

[0033] Second, as auxiliary evidence of the inventiveness of the claims of the present invention, it is also reflected in the following important aspects: As wildfires become more frequent due to global climate change, the market demand for efficient and real-time wildfire detection systems continues to increase. This invention can meet this demand and has broad market prospects.

[0034] The present invention can significantly reduce the number of model parameters and computational complexity, enabling it to run efficiently on resource-limited edge devices, which makes it possible to widely deploy wildfire detection systems, especially in remote and resource-limited areas.

[0035] The lightweight design used in the present invention reduces the hardware requirements and energy consumption of the model, and can be deployed using low-cost, low-performance edge devices, thereby reducing the cost of the overall system. This has important economic significance for large-scale deployment and long-term operation.

[0036] At present, the model design for wildfire detection in the domestic and foreign industries mainly adopts traditional models with large parameters. Traditional models require a lot of computing resources when running on edge devices, resulting in slow response speed and low computing efficiency; secondly, the number of parameters and calculations of large models are large, which exceeds the computing processing power and storage capacity of edge devices, greatly limiting the deployable devices of the model; most importantly, the detection of wildfires usually requires real-time and rapid response so that timely measures can be taken, and the complexity of traditional models often leads to high detection delays and cannot meet the requirements of real-time wildfire detection.

[0037] The present invention solves the high demand problem of traditional deep convolutional neural networks in computing resources and memory consumption by significantly reducing the number of model parameters and the amount of computation through a lightweight convolutional network design method, so that the model can run efficiently on resource-constrained edge devices.

[0038] The present invention adopts a training method of an auxiliary training branch, uses a deformable convolution operator with strong feature extraction capability on the auxiliary training branch, and removes the auxiliary training branch after the model training is completed, thereby reducing the number of model parameters and calculations and improving the real-time detection speed of the model without reducing the detection accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of a lightweight wildfire detection model design method based on auxiliary training branches provided in an embodiment of the present invention.

[0040] Figure 2 It is a main structural diagram of the lightweight wildfire detection model provided by an embodiment of the present invention.

[0041] Figure 3 It is a branch structure diagram of auxiliary training provided by an embodiment of the present invention.

[0042] Figure 4 It is an overall structural diagram of a lightweight wildfire detection model based on auxiliary training branches provided in an embodiment of the present invention.

[0043] Figure 5 It is a structural diagram of a lightweight wildfire detection model design system based on auxiliary training branches provided in an embodiment of the present invention.

[0044] Figure 6 This is a smoke detection effect diagram provided by an embodiment of the present invention.

[0045] Figure 7 This is a flame detection effect diagram provided by an embodiment of the present invention.

[0046] In the figure: 1. Data set construction module; 2. Data enhancement module; 3. Preprocessing module; 4. Partitioning module; 5. Structure design module; 6. Branch structure design module; 7. Model training module. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0048] like Figure 1 As shown, an embodiment of the present invention provides a lightweight wildfire detection model design method based on auxiliary training branches, the method comprising: S1: Wildfire image samples are collected by inspection drones and monitoring gimbals to build a wildfire dataset.

[0049] S2: The Mosaic-4 data augmentation method is used on the collected wildfire dataset, that is, 4 pictures are used and spliced ​​into one picture by random cropping and random arrangement. At the same time, some Gaussian noise is reasonably introduced to enhance the robustness of the model.

[0050] S3: The collected complete data set is preprocessed using the histogram equalization algorithm, and the preprocessed image samples are labeled.

[0051] S4: Divide the labeled image samples into training set and test set in proportion.

[0052] S5: Design a lightweight convolutional neural network main model structure. The designed lightweight convolutional neural network model compresses the model structure size; the proposed lightweight detection head using shared convolution can use shared convolution to realize the sharing of some convolution parameters, thereby improving the detection efficiency of the model.

[0053] S6: Design an auxiliary training branch structure, which can aggregate the gradient information of all target objects and retain the information of deep features, so as to obtain reliable gradients and alleviate the problem of information loss in deep supervision.

[0054] S7: Use the wildfire dataset to train a lightweight wildfire detection model based on auxiliary training branches, remove the auxiliary training branch structure in the trained model, and finally obtain a lightweight wildfire detection model.

[0055] The present invention provides a lightweight wildfire detection model design method based on auxiliary training branches, which realizes efficient and accurate wildfire detection through the collaborative work of multiple steps. First, in step S1, wildfire image samples are collected using patrol drones and monitoring gimbals to construct a wildfire dataset covering a variety of scenes and fire states. These sample data are collected and organized to provide sufficient diversity and representativeness, providing a solid data foundation for subsequent model training and detection.

[0056] Next, in step S2, the collected wildfire dataset is subjected to Mosaic-4 data enhancement processing. The Mosaic-4 method randomly crops and splices four images into one image, while introducing Gaussian noise to enhance the robustness of the model. This step effectively increases the diversity and complexity of the data, allowing the model to more stably identify the characteristics of fires when faced with wildfire scenes in real environments, reducing detection errors caused by noise and environmental changes.

[0057] In step S3, the dataset is preprocessed, including using a histogram equalization algorithm to balance the contrast and brightness of the image, thereby improving the clarity and detail of the image. Subsequently, the preprocessed image samples are annotated to ensure that each sample in the dataset has clear target information. The annotated data samples are further divided into a training set and a test set (step S4) to ensure that the training and testing processes of the model are sufficiently representative and feasible.

[0058] In step S5, a lightweight convolutional neural network main model is designed. By compressing the size of the model structure, the amount of calculation is reduced and the detection speed of the model is improved. In particular, the lightweight detection head using shared convolution effectively improves the detection efficiency of the model by sharing some convolution parameters, ensuring that the detection accuracy of the model is maintained while reducing the number of parameters. This design is particularly important in resource-constrained environments, such as in drones or other embedded systems.

[0059] Next, in step S6, an auxiliary training branch structure is designed. This structure is introduced to aggregate the gradient information of all target objects during the model training process, while retaining the information of deep features, to ensure that the model obtains reliable gradient signals during the training process. The auxiliary training branch can alleviate the problem of information loss in deep supervision, thereby improving the training effect and detection accuracy of the model. After training (step S7), the final model removes the auxiliary training branch and retains the lightweight detection model. This design ensures the lightweight and high efficiency of the model, and provides an efficient and accurate solution for wildfire detection in practical applications.

[0060] like Figure 2 As shown, in S5, the structure of the designed lightweight aircraft door detection model includes: (1) The real-time target detection network YOLOv8n is used to build the backbone network to improve the real-time detection speed of the model.

[0061] (2) A partial convolution operator is used to construct the feature extraction and fusion module of the neck network to improve the computational efficiency of the model.

[0062] (3) A shared convolution kernel is used to construct the output detection head. The number of parameters of the detection head is reduced by sharing parameters between modules, which improves the real-time detection speed of the model and realizes the identification and location of wildfires.

[0063] like Figure 3 As shown, in S6, the structure of the auxiliary training branch designed includes: (1) A normalization layer consisting of a convolutional layer, a batch normalization layer, and an activation function is used to normalize the original image and the feature map input through the backbone network, standardize the output of the convolutional layer, speed up the training speed, and improve the model stability.

[0064] (2) A deformable convolution operator is used to construct a feature fusion module. By merging feature maps at different levels and utilizing their complementary information, the model’s ability to recognize targets is enhanced.

[0065] (3) The merged feature map is passed to the output detection head to provide additional auxiliary information for wildfire detection during model training and improve the detection accuracy of the model.

[0066] like Figure 4 As shown, in S7, the lightweight wildfire detection model based on the auxiliary training branch includes: (1) The main structure consisting of the backbone network, neck network and output detection head is responsible for processing and detecting the input video images to realize the identification and location of wildfires.

[0067] (2) The auxiliary training branch consisting of the normalization layer and the feature fusion module is responsible for providing additional auxiliary information during model training, assisting the main structure in learning the features of image samples and improving the detection accuracy of the model.

[0068] (3) The auxiliary training branches will be removed after the model training is completed. Finally, the model will only be left with the main structure consisting of the backbone network, the neck network and the output detection head. This will reduce the number of model parameters and calculations and improve the real-time detection speed of the model without reducing the detection accuracy of the model.

[0069] like Figure 5 As shown, an embodiment of the present invention provides a lightweight mountain fire detection model design system based on auxiliary training branches based on the lightweight mountain fire detection model design method based on auxiliary training branches, and the system specifically includes: In the data set construction module 1, the inspection drone and the monitoring gimbal collect wildfire image samples to construct a wildfire data set.

[0070] Data enhancement module 2 is connected to the data set construction module 1, and uses the Mosaic-4 data enhancement method on the collected wildfire data set, that is, 4 pictures are used and spliced ​​into one picture by random cropping and random arrangement. At the same time, some Gaussian noise is reasonably introduced to enhance the robustness of the model.

[0071] The preprocessing module 3 is connected to the data enhancement module 2, and preprocesses the collected complete data set using a histogram equalization algorithm, and annotates the preprocessed image samples.

[0072] The division module 4 is connected to the preprocessing module 3 and divides the labeled image samples into a training set and a test set in proportion.

[0073] The structural design module 5 is connected to the division module 4, and designs the main model structure of the lightweight convolutional neural network. The designed lightweight convolutional neural network model compresses the model structure size; the proposed lightweight detection head using shared convolution can use shared convolution to realize the sharing of some convolution parameters, thereby improving the detection efficiency of the model.

[0074] The branch structure design module 6 is connected to the structure design module 5 to design an auxiliary training branch structure. The designed auxiliary training branch structure can aggregate the gradient information of all target objects and retain the information of deep features, so as to obtain reliable gradients and alleviate the problem of information loss in deep supervision.

[0075] The model training module 7 is connected to the branch structure design module 6, and uses the wildfire dataset to train the lightweight wildfire detection model based on the auxiliary training branch, removes the auxiliary training branch structure in the trained model, and finally obtains the lightweight wildfire detection model.

[0076] An embodiment of the present invention provides a computer device, characterized in that the computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for designing a lightweight wildfire detection model based on an auxiliary training branch.

[0077] An embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the method for designing a lightweight wildfire detection model based on an auxiliary training branch.

[0078] An embodiment of the present invention provides an information data processing terminal, which is used to implement the lightweight wildfire detection model design system based on auxiliary training branches.

[0079] Relevant evidence of the technical effects achieved by the embodiments of the present invention.

[0080] The performance of the lightweight wildfire detection model based on auxiliary training branches designed by the present invention is as follows: (1) The number of parameters is 2.3M, which is 23% less than the original model (3.0M).

[0081] (2) The computational complexity is 6.3G, which is 22% less than the original model (8.1G).

[0082] (3) The average recognition accuracy is 77%, which is 5% higher than the original model (72%).

[0083] (4) The detection speed is 80 frames, which is 29% higher than the original model (57 frames).

[0084] The detection effect of the model is as follows Figure 6 and Figure 7 As shown, Figure 6 For smoke detection effect, Figure 7 For flame detection effect.

[0085] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0086] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A lightweight wildfire detection model design method based on auxiliary training branches, characterized in that: The method includes: S1: Inspection drones and monitoring gimbals collect wildfire image samples to build a wildfire dataset; S2: The Mosaic-4 data augmentation method is used on the collected wildfire dataset, that is, 4 pictures are taken and spliced ​​into one picture by random cropping and random arrangement, and some Gaussian noise is reasonably introduced; S3: Preprocess the collected complete data set using the histogram equalization algorithm and annotate the preprocessed image samples; S4: Divide the labeled image samples into training set and test set according to the proportion; S5: Design a lightweight convolutional neural network main model structure, wherein the designed lightweight convolutional neural network model compresses the model structure size; a lightweight detection head using shared convolution uses shared convolution to realize the sharing of some convolution parameters; S6: Design an auxiliary training branch structure, wherein the designed auxiliary training branch structure aggregates the gradient information of all target objects and retains the information of deep features, thereby obtaining a reliable gradient and alleviating the information loss problem in deep supervision; S7: Use the wildfire dataset to train a lightweight wildfire detection model based on auxiliary training branches, remove the auxiliary training branch structure in the trained model, and finally obtain a lightweight wildfire detection model.

2. The lightweight wildfire detection model design method based on auxiliary training branches according to claim 1 is characterized in that: In S5, the structure of the designed lightweight aircraft door detection model includes: (1) Use the real-time detection target detection network YOLOv8n to build the backbone network to improve the real-time detection speed of the model; (2) The feature extraction and fusion module of the neck network is constructed using partial convolution operators to improve the computational efficiency of the model; (3) A shared convolution kernel is used to construct the output detection head. The number of parameters of the detection head is reduced by sharing parameters between modules, which improves the real-time detection speed of the model and realizes the identification and location of wildfires.

3. The lightweight wildfire detection model design method based on auxiliary training branches as claimed in claim 1 is characterized in that: In S6, the structure of the auxiliary training branch is designed, including: (1) A normalization layer including a convolutional layer, a batch normalization layer, and an activation function is used to normalize the original image and the feature map input through the backbone network, standardize the output of the convolutional layer, speed up the training speed, and improve the model stability; (2) A feature fusion module is constructed using a deformable convolution operator. By merging feature maps at different levels and using their complementary information, the model’s ability to recognize targets is enhanced. (3) The merged feature map is passed to the output detection head to provide additional auxiliary information for wildfire detection during model training and improve the detection accuracy of the model.

4. The lightweight wildfire detection model design method based on auxiliary training branches as claimed in claim 1 is characterized in that: In S7, the lightweight wildfire detection model based on the auxiliary training branch includes: (1) The main structure consisting of the backbone network, neck network, and output detection head is responsible for processing and detecting the input video image to achieve wildfire identification and location; (2) The auxiliary training branch consisting of the normalization layer and the feature fusion module is responsible for providing additional auxiliary information during model training, assisting the main structure in learning the features of image samples and improving the detection accuracy of the model; (3) The auxiliary training branches will be removed after the model training is completed. Finally, the model will only be left with the main structure consisting of the backbone network, the neck network and the output detection head. This will reduce the number of model parameters and calculations and improve the real-time detection speed of the model without reducing the detection accuracy of the model.

5. A lightweight wildfire detection model design system based on auxiliary training branches based on the lightweight wildfire detection model design method based on auxiliary training branches as described in claims 1-4, characterized in that: The system specifically includes: The dataset construction module collects wildfire image samples using patrol drones and monitoring gimbals to construct wildfire datasets; The data enhancement module is connected to the dataset construction module. It uses the Mosaic-4 data enhancement method on the collected wildfire dataset, that is, it uses 4 pictures and splices them into one picture by randomly cropping and randomly arranging them, and reasonably introduces some Gaussian noise; The preprocessing module is connected to the data enhancement module, which preprocesses the collected complete data set using the histogram equalization algorithm and annotates the preprocessed image samples; The partitioning module is connected to the preprocessing module to divide the labeled image samples into training sets and test sets in proportion; The structure design module is connected with the partitioning module to design a lightweight convolutional neural network main model structure, wherein the designed lightweight convolutional neural network model compresses the model structure size; the lightweight detection head using shared convolution uses shared convolution to realize the sharing of some convolution parameters; A branch structure design module is connected to the structure design module to design an auxiliary training branch structure. The designed auxiliary training branch structure can aggregate gradient information containing all target objects and retain information of deep features, thereby obtaining reliable gradients and alleviating the problem of information loss in deep supervision. The model training module is connected to the branch structure design module, and uses the wildfire dataset to train the lightweight wildfire detection model based on the auxiliary training branch, removes the auxiliary training branch structure in the trained model, and finally obtains the lightweight wildfire detection model.

6. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the lightweight wildfire detection model design method based on auxiliary training branches as described in any one of claims 1-4.

7. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the method for designing a lightweight wildfire detection model based on an auxiliary training branch as described in any one of claims 1 to 4.

8. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the lightweight wildfire detection model design system based on auxiliary training branches as described in claim 5.

Images