All-weather unmanned aerial vehicle forest fire detecting and positioning system based on SAR image and working method

Abstract

The invention discloses an all-weather unmanned aerial vehicle forest fire detection and positioning system based on SAR images and a working method, and belongs to the technical field of unmanned aerial vehicle target detection and positioning. The all-weather unmanned aerial vehicle forest fire detection and positioning system based on the SAR image comprises an SAR image acquisition module used for acquiring image data; the target detection module is used for obtaining identification confidence and coordinates of the forest fire in an image pixel coordinate system; and the binocular vision precise positioning module realizes precise positioning of the forest fire in a world coordinate system. According to the invention, the large-area forest fire detection capability of the unmanned aerial vehicle can be improved, and the detection accuracy and real-time performance can be improved. Large-area forest fire detection is achieved through the unmanned aerial vehicle, meanwhile, the performance and accuracy of forest fire detection of the unmanned aerial vehicle in a large-area forest are improved, and better real-time performance is achieved.

Description

Technical Field

[0001] The present invention relates to a wildfire detection and positioning system and a working method, and particularly to an all-weather unmanned aerial vehicle (UAV) wildfire detection and positioning system and a working method based on SAR images, belonging to the technical field of UAV target detection and positioning. Background Art

[0002] The global forest area covers approximately 40 hectares, accounting for 31% of the total land area. With the continuous enhancement of people's environmental protection awareness, people are paying more and more attention to the protection of forest resources. In recent years, with the global warming, the frequency of wildfires has been increasing continuously, which has become an important reason for destroying forest resources and ecological balance. Therefore, wildfire detection is of great significance for protecting forest resources and the ecological environment.

[0003] With the rapid development of modern industry, science and technology, the further expansion of market demand and the continuous support of policies, UAVs have witnessed rapid development and been widely used in various fields. Currently, UAVs are widely used in agricultural crop detection, traffic detection, disaster rescue, environmental protection and other fields. However, for the application of UAVs in wildfire detection, most of the research mainly focuses on applying UAVs to power line patrols and detecting wildfires on the established patrol routes, rather than using UAVs to detect wildfires in a large area. There is still great potential for the application of UAVs in this aspect.

[0004] The progress of computer vision has brought great changes to the application of UAVs. Using UAVs for wildfire detection can not only reduce labor costs and time, but also detect wildfires in a timely manner, reduce the losses caused by wildfires, and provide information for wildfire rescue. Therefore, UAV detection of wildfires is a better choice in wildfire detection. Since the image size collected by UAVs during high-altitude flight is small, and the wildfire is small in the initial stage, the detection target becomes a small target, which reduces the detection accuracy and makes it difficult to accurately identify wildfires. In addition, the images collected under normal circumstances are easily affected by weather conditions and the acquisition time period, making UAVs face many technical challenges in the process of all-weather wildfire detection. Therefore, how to obtain processable images all-weather and accurately detect small targets is the key factor for effectively realizing wildfire detection and positioning.

[0005] Synthetic Aperture Radar (SAR) can be used for earth observation all day and all weather conditions without being restricted by illumination and climate conditions, ensuring that ground images can be obtained at all times and in various climate conditions. Moreover, SAR can penetrate the earth's surface or vegetation to obtain the information hidden beneath. In addition, through synthetic aperture technology, SAR can achieve high-resolution imaging and maintain good resolution even at long distances. Therefore, SAR is widely used in military, agricultural detection, geological exploration and other fields.

[0006] Traditional UAV detection systems mainly rely on ordinary cameras and target detection algorithms. After obtaining the results output by the target detection algorithm, a positioning algorithm is used to locate the target. These technologies played an important role in the initial UAV detection systems. Image data is collected by the camera, and preprocessing such as normalization and denoising is performed on the image data. Then, the preprocessed image data is input into the target detection algorithm for target positioning in the image. Finally, the actual position of the detected target in the world coordinate system is finally located through the principle of binocular vision imaging. However, traditional target detection systems usually cannot detect small targets well, and the images collected by ordinary cameras are greatly affected by weather, illumination and other conditions, and cannot perform large-area wildfire detection well throughout the day.

[0007] Therefore, developing an all-weather UAV wildfire detection and positioning system based on SAR images and its working method has important technical breakthrough significance and application value. Summary of the Invention

[0008] In order to solve the problem of improving the ability of UAVs to detect large-area wildfires and improving the accuracy and real-time performance of detection, the present invention further proposes an all-weather UAV wildfire detection and positioning system and working method based on SAR images.

[0009] To achieve the above objectives, the present invention adopts the following technical solutions:

[0010] An all-weather UAV wildfire detection and positioning system based on SAR images, the all-weather UAV wildfire detection and positioning system based on SAR images includes:

[0011] An SAR image acquisition module, the SAR image acquisition module includes two synthetic aperture radars for collecting image data;

[0012] A target detection module, the algorithm of the target detection module is an improved YOLOv5 target detection algorithm, which is used to obtain the recognition confidence and the coordinates of the wildfire in the image pixel coordinate system;

[0013] Binocular vision precise positioning module, which converts the coordinates of the wildfire in the image pixel coordinate system into the coordinates of the wildfire in the world coordinate system through binocular vision algorithm, so as to realize the precise positioning of the wildfire in the world coordinate system.

[0014] Furthermore, the two synthetic aperture radars in the SAR image acquisition module are respectively arranged at the left and right viewing angles of the unmanned aerial vehicle.

[0015] Furthermore, the image data collected by the SAR image acquisition module needs to be preprocessed before being input into the target detection module.

[0016] Furthermore, the improved YOLOv5 in the target detection module, compared with YOLOv5, introduces a tiny object detection head and a decoupled head in the head network. The improved YOLOv5 algorithm can be applied to small object detection and decouples the classification head and the detection head.

[0017] Furthermore, the tiny object detection head is a detection head that can be downsampled by 4 times, enabling YOLO to detect objects with 4*4 pixels.

[0018] Furthermore, the decoupled head is a decoupled head constructed by a mixed channel strategy, reducing the number of 3*3 convolutional layers to only one.

[0019] Furthermore, the method for converting the coordinates of the wildfire in the image pixel coordinate system to the coordinates of the wildfire in the world coordinate system in the binocular vision precise positioning module is the transformation of the coordinate system. The transformation of the coordinate system is specifically that the world coordinate system obtains the camera coordinate system through rigid body transformation, the camera coordinate system obtains the ideal image coordinate system through perspective projection, the ideal image coordinate system is corrected for distortion to obtain the actual image coordinate system, and the actual image coordinate system is then transformed to the image pixel coordinate system through affine transformation.

[0020] Furthermore, the conversion of the coordinates of the wildfire in the image pixel coordinate system to the coordinates of the wildfire in the world coordinate system in the binocular vision precise positioning module specifically includes:

[0021] Let the coordinates of the wildfire in the world coordinate system be P(X w , Y w , Z w ), the coordinates of the camera coordinate system be (X c , Y c , Z c ), the coordinates of the SAR imaging coordinate system be (x, y), and the coordinates of the image pixel coordinate system be (u, v). After the transformation of the coordinate system, there are the following formulas:

[0022]

[0023] In formula (1), (u 0 , v 0 ) is the origin of the SAR imaging plane coordinate system represented by the coordinates of the pixel coordinate system. d x and d y respectively represent the lengths of a unit pixel in the X and Y directions of the image pixel coordinate system. The rotation matrix R and the translation vector T are the external parameters of the SAR;

[0024] Obtain the internal and external parameters of the camera, and perform radial distortion correction, and then perform stereo matching of the camera;

[0025] Through binocular vision model calculation, obtain the depth information Z of the object and determine the exact position of the object in space. The expression is as follows:

[0026]

[0027] In formulas (2) to (4), d is the disparity, regarded as pixel units here, S is the distance between the two SAR optical centers, and f is the focal length of the SAR.

[0028] Furthermore, the method for obtaining the internal and external parameters of the camera is to obtain the internal and external parameters of the SAR through Zhang's calibration method for off-line calibration.

[0029] Furthermore, the method for stereo matching of the camera is to fuse the least squares filtering through the SGBM algorithm.

[0030] A working method of an all-weather UAV wildfire detection and positioning system based on SAR images. The system is the all-weather UAV wildfire detection and positioning system based on SAR images. The working method is realized through the following steps:

[0031] S1: Collect wildfire image data through the SAR image acquisition module;

[0032] S2: Perform image preprocessing on the image data obtained in S1;

[0033] S3: Input the image data processed in S2 into the target detection module, and identify the input image based on the improved YOLOv5 target detection algorithm to detect whether there is a wildfire and locate it in the pixel coordinate system;

[0034] S4: Use the binocular vision precise positioning module for the coordinates of the wildfire obtained in S3 in the pixel coordinate system to realize the positioning of the coordinates of the wildfire in the world coordinate system.

[0035] The beneficial effects of the present invention are:

[0036] 1. The present invention proposes an all-weather UAV wildfire detection and positioning system based on SAR images, which realizes large-area wildfire detection by UAVs. ， Meanwhile, it improves the performance and accuracy of UAVs in wildfire detection in large-area forests and has better real-time performance.

[0037] 2. Compared with other object detection tasks, for the task of UAV wildfire detection in large-area forests, the present invention collects image data through the SAR image acquisition module and improves the YOLOv5 algorithm to make it more suitable for small object detection, making this system more suitable for the task of wildfire detection, realizing all-time wildfire detection, and then further realizing the positioning of wildfires through the binocular vision precise positioning module to determine the coordinate position of wildfires in the world coordinate system.

[0038] 3. Starting from the three aspects of the small object characteristics of wildfire detection, the real-time requirements of UAVs, and the system's portability requirements, the ordinary YOLOv5 algorithm cannot accurately detect small objects. Therefore, on the basis of the traditional YOLOv5 algorithm, a tiny object detection head and a more efficient decoupled head are introduced to improve the detection accuracy and real-time performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a schematic structural framework diagram of the improved YOLOv5 algorithm of the present invention;

[0040] Figure 2 is a schematic structural diagram of the decoupled head adopted by the improved YOLOv5 algorithm of the present invention;

[0041] Figure 3 is a schematic diagram of the camera imaging principle of the present invention;

[0042] Figure 4 is a schematic diagram of the coordinate system conversion process of the present invention;

[0043] Figure 5 is a schematic diagram of the coordinate system conversion of the present invention;

[0044] Figure 6 is a schematic diagram of the binocular vision ranging principle of the present invention;

[0045] Figure 7 is a schematic structural diagram of the decoupled head adopted by the improved YOLOv5 algorithm of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0046] In the description of the present invention, it should be noted that all directional indications (such as left, right, etc.) are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0047] Specific Embodiment 1: In combination with Figures 1-7 This embodiment is described. The all-weather UAV wildfire detection and positioning system based on SAR images described in this embodiment includes:

[0048] An SAR image acquisition module, which includes two synthetic aperture radars for collecting image data. Preferably, the two synthetic aperture radars are respectively arranged at the left and right viewing angles of the UAV. This module includes two image acquisition devices (SAR), one for the left viewing angle and one for the right viewing angle, respectively collecting image data. SAR is a technology that uses radar principles for imaging. By transmitting microwave signals and receiving the echoes reflected from the targets, high-resolution ground images can be generated. Its working principle is based on synthetic aperture technology. With the movement of the UAV, signals are transmitted and received at different positions, and then these multi-angle data are used to synthesize high-resolution images. After obtaining the image data, preprocessing of the image data is carried out. The preprocessing is to obtain the preprocessed image data through operations such as image normalization, size adjustment, and denoising. That is, the image data collected by the SAR image acquisition module needs to be preprocessed before being input into the target detection module.

[0049] A target detection module, the algorithm of which is the improved YOLOv5 target detection algorithm, used to obtain the recognition confidence and the coordinates of the wildfire in the image pixel coordinate system. YOLOv5 (You Only Look Once version 5) is one of the YOLO series of algorithms. The ingenuity of the YOLO series of algorithms lies in that it transforms the target detection problem into a regression task, making the algorithm fast in detection speed and high in accuracy. This module has made two improvements based on YOLOv5 to enhance the performance of YOLOv5 for wildfire detection. The improved YOLOv5 in the target detection module, compared with YOLOv5, introduces a small target detection head and a decoupled head in the head network. The improved YOLOv5 algorithm can be applied to small target detection and decouples the classification head and the detection head.

[0050] YOLOv5 can be generally divided into three structures, namely the Backbone, the Neck, and the Head. Among them, the Backbone is responsible for extracting the features and details of the input image and generating high-level feature maps, which contain rich semantic information and provide a basis for subsequent detection tasks. The Neck is responsible for further processing and fusing the features extracted by the Backbone. By fusing feature maps at different levels, more comprehensive image features can be obtained. The Head is responsible for performing the final detection on the multi-scale feature maps, including classification and regression tasks, and generating the final bounding boxes and class confidence scores. Specifically, as Figure 1 shown, the Head in the YOLOv5 structure is improved. The tiny object detection head is a detection head with 4x downsampling, enabling YOLO to detect objects with 4*4 pixels. The original YOLOv5 algorithm performs 8x, 16x, and 32x downsampling on the input image. To improve the performance of small object detection, a detection head specifically for tiny objects is added at the Head. That is, an additional detection head with 4x downsampling is introduced, enabling YOLO to detect objects with 4*4 pixels. This improvement enables YOLOv5 to detect small objects with a resolution of 16. In this way, even when the drone is flying at a high altitude, it can effectively detect the small object of the wildfire.

[0051] As Figure 2 shown, the decoupled head is a decoupled head constructed using a hybrid channel strategy, reducing the number of 3*3 convolutional layers to only one. By introducing the decoupled head into YOLOv5, the detection head of YOLOv5 changes the number of channels of the input feature map through convolution, making its output channels contain target bounding box regression parameters, confidence scores, and class scores. Multiple types of information are integrated in one feature map, and the classification task and the localization task are strongly coupled with shared weights. Specifically, to improve the performance of YOLOv5, a decoupled head is introduced. A hybrid channel (HC) strategy is adopted to construct a more efficient decoupled head, reducing the latency while maintaining the accuracy. The number of 3*3 convolutional layers is reduced to only one, thereby reducing the number of parameters. This improvement further reduces the computational cost, achieves a lower inference latency, and better meets the improvement requirements for the real-time performance of drone object detection. Furthermore, it makes the drone more accurate and fast in performing the wildfire detection task.

[0052] Binocular vision precise positioning module. The binocular vision precise positioning module converts the coordinates of the wildfire in the image pixel coordinate system into the coordinates of the wildfire in the world coordinate system through the binocular vision algorithm, realizing the precise positioning of the wildfire in the world coordinate system.

[0053] As Figures 3-6As shown in the figure, the binocular vision algorithm relies on binocular SAR to achieve the positioning of the detected target and obtain the coordinates of the detected target in the world coordinate system. Through the target detection module, we can obtain the coordinates of the wildfire in the image pixel coordinate system. The conversion between the image pixel coordinate system and the world coordinate system can be achieved through the transformation between coordinate systems. The world coordinate system is transformed into the camera coordinate system through a rigid body transformation, the camera coordinate system is projected onto an ideal image coordinate system through a perspective projection, and the ideal image coordinate system is corrected for distortion to obtain the actual image coordinate system. The actual image coordinate system is then transformed into the image pixel coordinate system through an affine transformation. Based on this, the coordinates of the detected target in the world coordinate system can be obtained through the binocular vision algorithm, completing the detection and positioning of the wildfire.

[0054] The conversion of the coordinates of the wildfire in the image pixel coordinate system to the coordinates of the wildfire in the world coordinate system in the binocular vision precise positioning module specifically includes:

[0055] Let the coordinates of the wildfire in the world coordinate system be P(X w ,Y w ,Z w ), the coordinates of the camera coordinate system be (X c ,Y c ,Z c ), the coordinates of the SAR imaging coordinate system be (x, y), and the coordinates of the image pixel coordinate system be (u, v). After the transformation of the coordinate system, there are the following formulas:

[0056]

[0057] In formula (1), (u 0 ,v 0 ) is the origin of the SAR imaging plane coordinate system represented by the pixel coordinate system, d x and d y respectively represent the lengths of a unit pixel in the X and Y directions of the image pixel coordinate system, and the rotation matrix R and the translation vector T are the external parameters of the SAR;

[0058] In order to obtain the internal and external parameters of the camera, here, through Zhang's calibration method, the internal and external parameters of the SAR are obtained through offline calibration, and radial distortion correction is performed. Then, through the SGBM (Semiglobal Bidirectional Matching) method, combined with the least squares filtering (WLS) method, stereo matching of the camera is performed.

[0059] It can be known from formula (1) that due to the lack of the depth information Z of the object, the exact position of the object in space cannot be determined, while the binocular camera can calculate the depth information based on the information differences obtained by the two cameras. Through the calculation of the binocular vision model, the following formula can be obtained:

[0060]

[0061] In formulas (2) to (4), d is the parallax, regarded as pixel units here, S is the distance between the two SAR optical centers, and f is the focal length of the SAR.

[0062] The all-weather UAV wildfire detection and positioning system based on SAR images can detect wildfires over a large area all-weather, and the detection accuracy is little affected by surface vegetation. It can accurately identify small-scale wildfires and locate them. It can not only effectively reduce the damage to ecological resources caused by wildfires, but also reduce the wildfire detection cost and the economic losses brought by wildfires.

[0063] As Figure 7 shown, a working method of an all-weather UAV wildfire detection and positioning system based on SAR images. The system is the all-weather UAV wildfire detection and positioning system based on SAR images, and the working method is realized through the following steps:

[0064] S1: Collect wildfire image data through the SAR image acquisition module. The SAR image acquisition module mainly includes two synthetic aperture radars, which collect image data from the left and right perspectives respectively.

[0065] S2: Perform image preprocessing on the image data obtained in S1, and preprocess the collected images, such as removing noise and normalizing pixels.

[0066] S3: Input the image data processed in S2 into the target detection module, and identify the input image based on the improved YOLOv5 target detection algorithm to detect whether there is a wildfire and locate it in the pixel coordinate system.

[0067] S4: Use the binocular vision precise positioning module to realize the positioning of the coordinates of the wildfire in the world coordinate system based on the coordinates of the wildfire in the pixel coordinate system obtained in S3, and realize the positioning of the coordinates of the wildfire in the world coordinate system through the coordinates of the wildfire in the pixel coordinate system.

[0068] The images captured by ordinary cameras are greatly affected by conditions such as weather and lighting, and cannot effectively capture images at night or in low visibility weather. Therefore, they cannot be used for all-weather wildfire detection. Compared with ordinary cameras, SAR has unique advantages. SAR is not restricted by lighting and climate changes, and can continuously observe the ground at any time and under various harsh climate conditions, ensuring stable acquisition of ground images. This characteristic makes it very suitable for all-weather wildfire monitoring. In addition, SAR can penetrate the surface layer and vegetation, revealing important information that is obscured, thereby significantly improving the accuracy of wildfire detection. By applying advanced synthetic aperture technology, SAR can also achieve high-resolution long-range imaging, ensuring the clarity of images even when detecting at long distances, which further enhances the accuracy and overall efficiency of wildfire detection.

[0069] Object detection algorithms in the field of computer vision are mainly used to identify objects in images or videos and determine their categories and positions in the images. They can be roughly divided into two types: single-stage and two-stage. The representative algorithm of the single-stage is the YOLO series. The YOLO series algorithms transform the object detection problem into a regression problem and directly predict the bounding boxes and class probabilities through a single forward propagation. Two-stage object detection algorithms mainly perform object detection through two steps: first, generate potential candidate regions, and then classify and accurately locate the bounding boxes of these candidate regions. Two-stage object detection algorithms such as R-CNN, Fast R-CNN, etc. pay more attention to accuracy and have better detection effects in small object detection and complex scenes, but the detection speed is slower. Although the single-stage object detection algorithm has relatively lower detection accuracy compared to the two-stage object detection algorithm, its detection speed is fast, which better meets the real-time improvement requirements of wildfire detection by drones. Among the YOLO series algorithms, the YOLOv5 algorithm shows unique advantages and has great improvement potential in the rapidly developing field of object detection technology due to its excellent detection speed, accuracy, and enduring model architecture, and there is still a vast exploration space in the field of small object detection in specific scenarios. This makes YOLOv5 particularly suitable for optimization in specific scenarios, thus promoting the implementation and deployment in practical applications. Therefore, the present invention improves the YOLOv5 algorithm and enhances its small object detection accuracy and related performance in the context of wildfire detection. Based on the YOLOv5 detection algorithm, a tiny object detection head is introduced, enabling the YOLOv5 algorithm to detect targets with 4*4 pixels, and decoupling the classification head and the detection head to make the tasks of classification and regression of YOLOv5 proceed in parallel, reducing the number of parameters of the algorithm and improving the real-time performance of YOLOv5. Finally, binocular vision is used to locate the detected wildfire target, improving the ability of drones to detect large-area wildfires and enhancing the detection accuracy and real-time performance, which is of great significance for wildfire detection.

[0070] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Although the present invention has been disclosed above with the preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to equivalent embodiments by using the disclosed technical content within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any simple modification, equivalent replacement, and improvement made to the above embodiments within the spirit and principle of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. An all-weather UAV mountain fire detection and positioning system based on SAR images, characterized by: The all-weather UAV mountain fire detection and positioning system based on SAR images includes: A SAR image acquisition module, wherein the SAR image acquisition module includes two synthetic aperture radars for acquiring image data; A target detection module, wherein the algorithm of the target detection module is an improved YOLOv5 target detection algorithm, which is used to obtain the recognition confidence and the coordinates of the wildfire in the image pixel coordinate system; A binocular vision precise positioning module, wherein the binocular vision precise positioning module converts the coordinates of the wildfire in the image pixel coordinate system into the coordinates of the wildfire in the world coordinate system through a binocular vision algorithm, thereby realizing precise positioning of the wildfire in the world coordinate system.

2. The all-weather UAV mountain fire detection and positioning system based on SAR images according to claim 1 is characterized by: The two synthetic aperture radars in the SAR image acquisition module are respectively arranged at the left and right viewing angles of the UAV.

3. The all-weather UAV mountain fire detection and positioning system based on SAR images according to claim 2 is characterized by: The image data collected by the SAR image acquisition module needs to be input into the target detection module after image preprocessing.

4. The all-weather UAV mountain fire detection and positioning system based on SAR images according to claim 1 is characterized by: Compared with YOLOv5, the improved YOLOv5 of the target detection module introduces a small target detection head and a decoupling head in the head network. The improved YOLOv5 algorithm can be suitable for small target detection and decouples the classification head and the detection head.

5. The all-weather UAV mountain fire detection and positioning system based on SAR images according to claim 4 is characterized by: The tiny target detection head is a 4x down-sampling detection head, which enables YOLO to detect targets with 4*4 pixels.

6. The all-weather UAV mountain fire detection and positioning system based on SAR images according to claim 4 is characterized by: The decoupling head is a decoupling head constructed by a hybrid channel strategy, which reduces the number of 3*3 convolutional layers to only one.

7. The all-weather UAV mountain fire detection and positioning system based on SAR images according to claim 1 is characterized by: The method of converting the coordinates of the wildfire in the image pixel coordinate system in the binocular vision precise positioning module into the coordinates of the wildfire in the world coordinate system is the transformation of the coordinate system. The transformation of the coordinate system is specifically to obtain the camera coordinate system through rigid body transformation of the world coordinate system, and the camera coordinate system is obtained by perspective projection. The ideal image coordinate system is corrected for distortion to obtain the actual image coordinate system, and the actual image coordinate system is then converted to the image pixel coordinate system through affine transformation.

8. The all-weather UAV mountain fire detection and positioning system based on SAR images according to claim 7 is characterized by: The conversion of the coordinates of the mountain fire in the image pixel coordinate system in the binocular vision precise positioning module into the coordinates of the mountain fire in the world coordinate system specifically includes: Suppose the coordinates of the wildfire in the world coordinate system are P(X w ,Y w ,Z w ), the coordinates of the camera coordinate system are (X c ,Y c ,Z c ), the coordinates of the SAR imaging coordinate system are (x, y), and the coordinates of the image pixel coordinate system are (u, v). After the coordinate system transformation, the following formula is obtained: In formula (1), (u0, v0) is the origin of the SAR imaging plane coordinate system represented by the pixel coordinate system, d x and d y They represent the length of the unit pixel in the X and Y directions of the image pixel coordinate system, respectively. The rotation matrix R and the translation vector T are the external parameters of SAR. Obtain the camera's intrinsic and extrinsic parameters, correct radial distortion, and then perform stereo matching of the camera; Through the binocular vision model calculation, the depth information Z of the object is obtained to determine the exact position of the object in space. The expression is as follows: In equations (2) to (4), d is the parallax, which is considered as pixel units here, S is the distance between the two SAR optical centers, and f is the focal length of the SAR.

9. The all-weather UAV mountain fire detection and positioning system based on SAR images according to claim 8 is characterized by: The method of obtaining the internal and external parameters of the camera is to obtain the internal and external parameters of the SAR through offline calibration using Zhang's calibration method. The method of stereo matching of the camera is to use the SGBM algorithm and integrate the least squares filter.

10. A working method of an all-weather UAV mountain fire detection and positioning system based on SAR images, characterized in that: The system is an all-weather UAV mountain fire detection and positioning system based on SAR images as described in any one of claims 1 to 9, and the working method is implemented by the following steps: S1: Collect wildfire image data through SAR image acquisition module; S2: perform image preprocessing on the image data obtained in S1; S3: Input the image data processed by S2 into the target detection module, identify the input image based on the improved YOLOv5 target detection algorithm, detect whether there is a wildfire and locate it in the pixel coordinate system; S4: The coordinates of the wildfire obtained in S3 in the pixel coordinate system are positioned in the world coordinate system using the binocular vision precision positioning module.

Images