Unmanned aerial vehicle edge intelligent analysis terminal based on single Beidou

Abstract

The invention discloses an unmanned aerial vehicle edge intelligent analysis terminal based on a single Beidou, and relates to the technical field of satellite navigation and unmanned aerial vehicles, and the terminal comprises an edge calculation unit which is in communication connection with a high-definition image collection module, a high-precision positioning module, a Beidou short message communication module and a ground control station; the high-definition image acquisition module is used for acquiring high-definition image data through a CMOS (Complementary Metal Oxide Semiconductor) sensor; the high-precision positioning module is used for acquiring centimeter-level positioning data in the flight process of the unmanned aerial vehicle through an RTK technology and an INS system; the edge calculation unit is used for carrying out target recognition through a lightweight convolutional neural network in combination with a YOLOv5 target detection algorithm; the Beidou short message communication module is used for realizing short message communication between the unmanned aerial vehicle and a ground control station by using RDSS service; and the ground control station is used for creating a user interface by using Qt, carrying out data visualization through Tableau and adopting a security protocol. According to the scheme, remote control, remote information transmission and high-precision positioning functions in a complex environment of the unmanned aerial vehicle are realized.

Description

Technical Field

[0001] The present invention relates to the field of satellite navigation and unmanned aerial vehicle (UAV) technology, and in particular to an unmanned aerial vehicle (UAV) edge intelligent analysis terminal based on a single Beidou. Background Art

[0002] With the continuous advancement of drone technology and the strengthening of the intelligent trend, the application of edge intelligent analysis terminals in the drone field is gradually showing its great potential. The drone edge intelligent analysis terminal can collect, process and analyze the data transmitted by the drone in real time, providing key decision support for various application scenarios. In the agricultural field, drones can monitor and analyze crops in real time by carrying equipment such as high-definition cameras and multispectral sensors, helping farmers optimize agricultural management and improve yield and efficiency. In terms of environmental monitoring, drones can collect environmental data in real time, providing strong support for environmental protection and disaster warning. In addition, in the fields of logistics, construction, infrastructure inspection, etc., the application of drone edge intelligent analysis terminals is also increasing, improving work efficiency and safety. In the future, with the continuous development of technologies such as artificial intelligence, big data and the Internet of Things, the performance and functions of drone edge intelligent analysis terminals will be further improved. For example, by integrating more advanced algorithms and models, drones can achieve more accurate analysis and prediction of data, providing more reliable support for decision-making. At the same time, with the popularization of high-speed communication technologies such as 5G, data transmission between drones and edge intelligent analysis terminals will be more efficient and stable, providing better protection for real-time applications. To sum up, with the continuous advancement of technology and the continuous expansion of application scenarios, drone edge intelligent analysis terminals will play a greater role in various fields, bringing more convenience and value to people's production and life.

[0003] As the core technology in this field, UAV remote control not only relies on advanced wireless communication technology, but also requires efficient data processing mechanisms to support it. However, UAV edge intelligent analysis terminals still face challenges in remote control, long-distance data information transmission, and precise positioning in complex environments. To solve these problems, the short message function and high-precision positioning function of the Beidou system are particularly important. With its wide coverage and strong anti-interference ability, the Beidou short message function provides UAVs with a solution for remote control and reliable data transmission without ground network coverage. At the same time, the high-precision positioning function of the Beidou system ensures the accurate positioning of UAVs in complex environments and improves the accuracy and safety of mission execution. Therefore, the application of the Beidou system has injected new vitality into the development of UAV edge intelligent analysis terminals. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a UAV edge intelligent analysis terminal based on a single Beidou, which solves the problems in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a single Beidou-based UAV edge intelligent analysis terminal, including an edge computing unit, wherein the edge computing unit is communicatively connected with a high-definition image acquisition module, a high-precision positioning module, a Beidou short message communication module and a ground control station; The high-definition image acquisition module acquires high-definition image data through a high-resolution image sensor; The high-precision positioning module uses RTK technology in combination with the INS system to collect centimeter-level positioning data during the flight of the drone; The edge computing unit deploys the MobileNetV3-Small model and performs target recognition and data analysis through a lightweight convolutional neural network combined with the YOLOv5 target detection algorithm; The Beidou short message communication module uses the RDSS service of the Beidou-2 satellite system to realize short message communication between the UAV and the ground control station; The ground control station uses Qt to create a user interface for sending flight instructions and monitoring drone image information and positioning information, uses Tableau for data visualization, and adopts safety protocols.

[0006] Furthermore, in the high-definition image acquisition module, the process of acquiring high-definition image data by using a high-resolution image sensor includes: A CMOS high-resolution image sensor is deployed in front of the drone to capture ambient light and generate image data through analog-to-digital conversion; The collected raw image data is preprocessed, and the preprocessing steps include denoising, white balance adjustment, color enhancement, and geometric correction. The lightweight machine learning model TensorFlow Lite is used to extract features from the image, identify and annotate visual features, and transmit the preprocessed and feature-extracted high-definition image data to the edge computing unit for data analysis.

[0007] Furthermore, the process of the high-precision positioning module, through RTK technology and combined with the INS system, collecting centimeter-level positioning data of the UAV during flight includes: A high-precision positioning module is deployed inside the drone near the GNSS antenna. The RTK technology uses differential GNSS measurements to perform real-time processing of carrier phase observations between the ground control station and the mobile station on the drone to achieve high-precision positioning. The INS system uses built-in accelerometers and gyroscopes to monitor the attitude changes and motion status of the drone in real time. The Kalman filter algorithm is used to fuse the data from RTK and INS, eliminate noise interference, optimize position estimation, and parse key positioning parameters from the fused data. The key positioning parameters include latitude, longitude, altitude, speed and attitude angle. The collected high-precision positioning data is then transmitted to the edge computing unit, marked with a timestamp, and synchronized with the data from the high-definition image acquisition module.

[0008] Furthermore, in the edge computing unit, the MobileNetV3-Small model is deployed, and the process of image preprocessing includes: The edge computing unit has a built-in large-capacity memory, is equipped with an NVIDIA Jetson Nano high-performance embedded processor, and deploys the lightweight MobileNetV3-Small as the backbone network to adjust the pre-processed image data to 640x640 pixels.

[0009] Furthermore, in the edge computing unit, the process of extracting features and generating feature maps by using a lightweight convolutional neural network algorithm includes: The preprocessed image data is fed into the lightweight convolution layer of MobileNetV3-Small, and a 3×3 deep convolution kernel is applied independently to each input channel to generate a preliminary feature representation. The feature map generated by the deep convolution is linearly combined through a 1×1 point-by-point convolution kernel to change the number of channels. After each convolution operation, the Hard-Swish activation function is applied to perform a nonlinear transformation on the feature map. The feature map processed by the deep separable convolution and activation function is passed to the next layer to gradually extract deeper features until the final multi-scale feature map is generated for target detection. The generated feature map enters the SE module and performs a squeeze operation. The spatial information of each feature map is aggregated into a single value through global average pooling to form a channel descriptor. The channel descriptor passes through two fully connected layers and the middle ReLU activation function, and then passes through the Sigmoid function to generate weights representing the importance of each channel. The weights are used to multiply the channels of the original feature map to adjust the importance of each channel. The feature map adjusted by the SE module is passed to the multi-scale feature extraction part to participate in further feature extraction. Select the feature map adjusted by the SE module, synchronize the resolution through upsampling and downsampling operations, and perform fusion. Apply a set of convolution kernels with different sizes and convolution operations with different expansion rates on the fused feature map to capture the receptive field information. Perform 1x1, 2x2, and 4x4 maximum pooling operations on the fused feature map and splice them together to form a multi-scale feature representation. Use element-level addition to combine feature maps of different scales. Send the fused multi-scale feature map to the detection head of YOLOv5 for the final target detection task.

[0010] Furthermore, in the edge computing unit, the YOLOv5 target detection algorithm performs target detection based on the generated feature map, and combines the high-precision positioning data to provide accurate location information and classification results, including: The generated feature map is input into YOLOv5. YOLOv5 performs bounding box prediction independently on the feature map of each scale, predicts the x, y coordinates, width and height of multiple bounding boxes for each target center point, and the probability of containing the target object in the bounding box. Each predicted bounding box is classified into categories, and the confidence scores of the targets in the bounding box belonging to different categories are given. Non-maximum suppression technology is used to eliminate redundant bounding boxes, retain the bounding boxes with the highest scores, and remove the bounding boxes with low scores with high overlap. Finally, YOLOv5 summarizes the results of all scales and outputs a set of detection results including bounding boxes and their corresponding category labels. The output results are packaged into a short message format and transmitted back to the ground control station in real time through the Beidou short message communication module.

[0011] Furthermore, in the Beidou short message communication module, the process of using the RDSS service of the Beidou-2 satellite system to realize short message communication between the UAV and the ground control station includes: When the edge computing unit on the drone completes image processing and target detection, it generates target location and category label result data, compresses and encrypts the result data, and converts it into a format suitable for transmission via a satellite link. The encoded result data is sent to the Beidou short message communication module, which then converts the ground signal into a radio signal transmitted via a satellite and sends it to the Beidou satellite through the antenna; Utilizing the two-way communication function supported by RDSS, the module accepts requests from the drone and uploads them to the satellite. The Beidou satellite that receives the uplink signal forwards the request to the ground control station through relay. The ground control station is equipped with special receiving equipment to capture the downlink signal from the satellite, decode it and restore the original data content. The processed data is immediately displayed on the operating interface of the ground control station for the operator to view and make decisions, thereby realizing real-time monitoring of the drone mission execution.

[0012] Furthermore, in the ground control station, the process of using Qt to create a user interface for sending flight instructions and monitoring drone image information and positioning information includes: The user interface is designed using Qt, which is used to display the drone's status, image stream, and positioning information, and provides buttons and slider elements for interaction. It is used by the operator to send flight instructions to the drone. The Qt application is directly connected to the Beidou short message communication module, receives decoded data, and updates the UI in real time.

[0013] Furthermore, in the ground control station, the process of data visualization through Tableau and adopting safety protocols includes: The data sent back by the drone is stored on the local server, and Tableau in the ground control station visualizes it by accessing the data source. The SSL / TLS protocol is used to ensure the security of HTTP communication, and the security of the Beidou short message communication module is ensured by message authentication codes and digital signatures.

[0014] The present invention provides a UAV edge intelligent analysis terminal based on single Beidou. It has the following beneficial effects: First, through the close combination of the high-definition image acquisition module and the edge computing unit, the terminal can collect and process high-definition image data in real time, and use lightweight convolutional neural networks and YOLOv5 target detection algorithms to achieve efficient and accurate target recognition and data analysis, greatly improving the operating efficiency and intelligence level of drones in complex environments. Secondly, the high-precision positioning module uses RTK technology combined with the INS system, which not only supports fast initialization, but also can provide centimeter-level precise positioning data during flight, which is crucial for drones to perform high-precision tasks such as surveying and mapping, agricultural spraying, etc., ensuring the high precision and reliability of operations. In addition, the Beidou short message communication module uses the RDSS service of the Beidou-2 satellite system to achieve stable and reliable short message communication between drones and ground control stations, and can maintain smooth communication even in remote areas without network coverage, enhancing the remote control and data transmission capabilities of drones. Finally, the ground control station uses Qt to create a user interface, which is not only easy to operate, but also uses Tableau for data visualization, making flight command sending, drone status monitoring and data analysis more intuitive and clear. At the same time, the adoption of security protocols further ensures the security of data transmission and storage, providing solid technical support for the widespread application of drones. To sum up, this invention has demonstrated significant advantages in improving the intelligence level, operation accuracy, communication capabilities and data security of drones. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a block diagram of a single Beidou-based drone edge intelligent analysis terminal according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] like Figure 1 As shown, the present invention provides a technical solution: a single Beidou-based UAV edge intelligent analysis terminal, including an edge computing unit, wherein the edge computing unit is communicatively connected with a high-definition image acquisition module, a high-precision positioning module, a Beidou short message communication module and a ground control station; The high-definition image acquisition module acquires high-definition image data through a high-resolution image sensor; The high-precision positioning module uses RTK technology in combination with the INS system to collect centimeter-level positioning data during the flight of the drone; The edge computing unit deploys the MobileNetV3-Small model and performs target recognition and data analysis through a lightweight convolutional neural network combined with the YOLOv5 target detection algorithm; The Beidou short message communication module uses the RDSS service of the Beidou-2 satellite system to realize short message communication between the UAV and the ground control station; The ground control station uses Qt to create a user interface for sending flight instructions and monitoring drone image information and positioning information, uses Tableau for data visualization, and adopts safety protocols.

[0018] In the high-definition image acquisition module, the process of acquiring high-definition image data by using a high-resolution image sensor includes: A CMOS high-resolution image sensor is deployed in front of the drone to capture ambient light and generate image data through analog-to-digital conversion; The collected raw image data is preprocessed, and the preprocessing steps include denoising, white balance adjustment, color enhancement, and geometric correction. The lightweight machine learning model TensorFlow Lite is used to extract features from the image, identify and annotate visual features, and transmit the preprocessed and feature-extracted high-definition image data to the edge computing unit for data analysis.

[0019] The process of the high-precision positioning module collecting centimeter-level positioning data of the UAV during flight by using RTK technology in combination with the INS system includes: A high-precision positioning module is deployed inside the drone near the GNSS antenna. The RTK technology uses differential GNSS measurements to perform real-time processing of carrier phase observations between the ground control station and the mobile station on the drone to achieve high-precision positioning. The INS system uses built-in accelerometers and gyroscopes to monitor the attitude changes and motion status of the drone in real time. The Kalman filter algorithm is used to fuse the data from RTK and INS, eliminate noise interference, optimize position estimation, and parse key positioning parameters from the fused data. The key positioning parameters include latitude, longitude, altitude, speed and attitude angle. The collected high-precision positioning data is then transmitted to the edge computing unit, marked with a timestamp, and synchronized with the data from the high-definition image acquisition module.

[0020] In the edge computing unit, the MobileNetV3-Small model is deployed, and the process of image preprocessing includes: The edge computing unit has a built-in large-capacity memory, is equipped with an NVIDIA Jetson Nano high-performance embedded processor, and deploys the lightweight MobileNetV3-Small as the backbone network to adjust the pre-processed image data to 640x640 pixels.

[0021] In the edge computing unit, the process of extracting features and generating feature maps by using a lightweight convolutional neural network algorithm includes: The preprocessed image data is fed into the lightweight convolution layer of MobileNetV3-Small, and a 3×3 deep convolution kernel is applied independently to each input channel to generate a preliminary feature representation. The feature map generated by the deep convolution is linearly combined through a 1×1 point-by-point convolution kernel to change the number of channels. After each convolution operation, the Hard-Swish activation function is applied to perform a nonlinear transformation on the feature map. The feature map processed by the deep separable convolution and activation function is passed to the next layer to gradually extract deeper features until the final multi-scale feature map is generated for target detection. The generated feature map enters the SE module and performs a squeeze operation. The spatial information of each feature map is aggregated into a single value through global average pooling to form a channel descriptor. The channel descriptor passes through two fully connected layers and the middle ReLU activation function, and then passes through the Sigmoid function to generate weights representing the importance of each channel. The weights are used to multiply the channels of the original feature map to adjust the importance of each channel. The feature map adjusted by the SE module is passed to the multi-scale feature extraction part to participate in further feature extraction. Select the feature map adjusted by the SE module, synchronize the resolution through upsampling and downsampling operations, and perform fusion. Apply a set of convolution kernels with different sizes and convolution operations with different expansion rates on the fused feature map to capture the receptive field information. Perform 1x1, 2x2, and 4x4 maximum pooling operations on the fused feature map and splice them together to form a multi-scale feature representation. Use element-level addition to combine feature maps of different scales. Send the fused multi-scale feature map to the detection head of YOLOv5 for the final target detection task.

[0022] In the edge computing unit, the YOLOv5 target detection algorithm performs target detection based on the generated feature map, and the process of providing accurate location information and classification results in combination with high-precision positioning data includes: The generated feature map is input into YOLOv5. YOLOv5 performs bounding box prediction independently on the feature map of each scale, predicts the x, y coordinates, width and height of multiple bounding boxes for each target center point, and the probability of containing the target object in the bounding box. Each predicted bounding box is classified into categories, and the confidence scores of the targets in the bounding box belonging to different categories are given. Non-maximum suppression technology is used to eliminate redundant bounding boxes, retain the bounding boxes with the highest scores, and remove the bounding boxes with low scores with high overlap. Finally, YOLOv5 summarizes the results of all scales and outputs a set of detection results including bounding boxes and their corresponding category labels. The output results are packaged into a short message format and transmitted back to the ground control station in real time through the Beidou short message communication module.

[0023] In the Beidou short message communication module, the process of using the RDSS service of the Beidou-2 satellite system to realize short message communication between the UAV and the ground control station includes: When the edge computing unit on the drone completes image processing and target detection, it generates target location and category label result data, compresses and encrypts the result data, and converts it into a format suitable for transmission via a satellite link. The encoded result data is sent to the Beidou short message communication module, which then converts the ground signal into a radio signal transmitted via a satellite and sends it to the Beidou satellite through the antenna; Utilizing the two-way communication function supported by RDSS, the module accepts requests from the drone and uploads them to the satellite. The Beidou satellite that receives the uplink signal forwards the request to the ground control station through relay. The ground control station is equipped with special receiving equipment to capture the downlink signal from the satellite, decode it and restore the original data content. The processed data is immediately displayed on the operating interface of the ground control station for the operator to view and make decisions, thereby realizing real-time monitoring of the drone mission execution.

[0024] In the ground control station, the process of using Qt to create a user interface for sending flight instructions and monitoring drone image information and positioning information includes: The user interface is designed using Qt, which is used to display the drone's status, image stream, and positioning information, and provides buttons and slider elements for interaction. It is used by the operator to send flight instructions to the drone. The Qt application is directly connected to the Beidou short message communication module, receives decoded data, and updates the UI in real time.

[0025] The process of data visualization through Tableau and adopting safety protocols in the ground control station includes: The data sent back by the drone is stored on the local server, and Tableau in the ground control station visualizes it by accessing the data source. The SSL / TLS protocol is used to ensure the security of HTTP communication, and the security of the Beidou short message communication module is ensured by message authentication codes and digital signatures.

[0026] First, start the drone and the ground control station. Ensure that the high-definition image acquisition module, high-precision positioning module, Beidou short message communication module, and edge computing unit on the drone are in normal working condition. At the same time, start the ground control station normally and successfully connect to the drone. Secondly, send flight instructions through the ground control station. The operator sets the flight route, altitude, speed and other parameters according to the mission requirements on the user interface of the ground control station, and sends the instructions to the drone. After receiving the instructions, the drone starts to fly according to the preset route. During the flight, the high-definition image acquisition module continuously collects high-definition image data and transmits it to the edge computing unit for processing. The edge computing unit uses the MobileNetV3-Small model and the YOLOv5 target detection algorithm to analyze the image data in real time, identify the target object, and perform corresponding data analysis. At the same time, the high-precision positioning module continuously collects centimeter-level positioning data of the drone to ensure the flight accuracy of the drone. The collected data is transmitted to the ground control station in real time for the operator to monitor and analyze. Finally, when the drone completes the mission and needs to return, the operator sends a return instruction through the ground control station. After receiving the instruction, the drone returns safely according to the preset return route. During the entire flight, the Beidou short message communication module maintains communication with the ground control station to ensure real-time transmission of data and accurate receipt of instructions.

[0027] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions. The sentence "includes an element defined by ... does not exclude the existence of other identical elements in the process, method, article or device including the element".

[0028] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The UAV edge intelligent analysis terminal based on single Beidou is characterized by: It includes an edge computing unit, which is communicatively connected with a high-definition image acquisition module, a high-precision positioning module, a Beidou short message communication module and a ground control station; The high-definition image acquisition module acquires high-definition image data through a high-resolution image sensor; The high-precision positioning module uses RTK technology in combination with the INS system to collect centimeter-level positioning data during the flight of the drone; The edge computing unit deploys the MobileNetV3-Small model and performs target recognition and data analysis through a lightweight convolutional neural network combined with the YOLOv5 target detection algorithm; The Beidou short message communication module uses the RDSS service of the Beidou-2 satellite system to realize short message communication between the UAV and the ground control station; The ground control station uses Qt to create a user interface for sending flight instructions and monitoring drone image information and positioning information, uses Tableau for data visualization, and adopts safety protocols.

2. The single Beidou-based UAV edge intelligent analysis terminal according to claim 1 is characterized in that: In the high-definition image acquisition module, the process of acquiring high-definition image data by using a high-resolution image sensor includes: A CMOS high-resolution image sensor is deployed in front of the drone to capture ambient light and generate image data through analog-to-digital conversion; The collected raw image data is preprocessed, and the preprocessing steps include denoising, white balance adjustment, color enhancement, and geometric correction. The lightweight machine learning model TensorFlow Lite is used to extract features from the image, identify and annotate visual features, and transmit the preprocessed and feature-extracted high-definition image data to the edge computing unit for data analysis.

3. The single Beidou-based UAV edge intelligent analysis terminal according to claim 2 is characterized in that: The process of the high-precision positioning module collecting centimeter-level positioning data of the UAV during flight by using RTK technology in combination with the INS system includes: A high-precision positioning module is deployed inside the drone near the GNSS antenna. The RTK technology uses differential GNSS measurements to perform real-time processing of carrier phase observations between the ground control station and the mobile station on the drone to achieve high-precision positioning. The INS system uses built-in accelerometers and gyroscopes to monitor the attitude changes and motion status of the drone in real time. The Kalman filter algorithm is used to fuse the data from RTK and INS, eliminate noise interference, optimize position estimation, and parse key positioning parameters from the fused data. The key positioning parameters include latitude, longitude, altitude, speed and attitude angle. The collected high-precision positioning data is then transmitted to the edge computing unit, marked with a timestamp, and synchronized with the data from the high-definition image acquisition module.

4. The single Beidou-based UAV edge intelligent analysis terminal according to claim 3 is characterized by: In the edge computing unit, the MobileNetV3-Small model is deployed, and the process of image preprocessing includes: The edge computing unit has a built-in large-capacity memory, is equipped with an NVIDIA Jetson Nano high-performance embedded processor, and deploys the lightweight MobileNetV3-Small as the backbone network to adjust the pre-processed image data to 640x640 pixels.

5. The single Beidou-based UAV edge intelligent analysis terminal according to claim 4 is characterized in that: In the edge computing unit, the process of extracting features and generating feature maps by using a lightweight convolutional neural network algorithm includes: The preprocessed image data is fed into the lightweight convolution layer of MobileNetV3-Small, and a 3×3 deep convolution kernel is applied independently to each input channel to generate a preliminary feature representation. The feature map generated by the deep convolution is linearly combined through a 1×1 point-by-point convolution kernel to change the number of channels. After each convolution operation, the Hard-Swish activation function is applied to perform a nonlinear transformation on the feature map. The feature map processed by the deep separable convolution and activation function is passed to the next layer to gradually extract features until the final multi-scale feature map is generated for target detection. The generated feature map enters the SE module and performs a squeeze operation. The spatial information of each feature map is aggregated into a single value through global average pooling to form a channel descriptor. The channel descriptor passes through two fully connected layers and the middle ReLU activation function, and then passes through the Sigmoid function to generate weights representing the importance of each channel. The weights are used to multiply the channels of the original feature map to adjust the importance of each channel. The feature map adjusted by the SE module is passed to the multi-scale feature extraction part to participate in further feature extraction. Select the feature map adjusted by the SE module, synchronize the resolution through upsampling and downsampling operations, and perform fusion. Apply a set of convolution kernels with different sizes and convolution operations with different expansion rates on the fused feature map to capture the receptive field information. Perform 1x1, 2x2, and 4x4 maximum pooling operations on the fused feature map and splice them together to form a multi-scale feature representation. Use element-level addition to combine feature maps of different scales. Send the fused multi-scale feature map to the detection head of YOLOv5 for the final target detection task.

6. The single Beidou-based UAV edge intelligent analysis terminal according to claim 5 is characterized by: In the edge computing unit, the YOLOv5 target detection algorithm performs target detection based on the generated feature map, and the process of providing accurate location information and classification results in combination with high-precision positioning data includes: The generated feature map is input into YOLOv5. YOLOv5 performs bounding box prediction independently on the feature map of each scale, predicts the x, y coordinates, width and height of multiple bounding boxes for each target center point, and the probability of containing the target object in the bounding box. Each predicted bounding box is classified into categories, and the confidence scores of the targets in the bounding box belonging to different categories are given. Non-maximum suppression technology is used to eliminate redundant bounding boxes, retain the bounding boxes with the highest scores, and remove the bounding boxes with low scores with high overlap. Finally, YOLOv5 summarizes the results of all scales and outputs a set of detection results including bounding boxes and their corresponding category labels. The output results are packaged into a short message format and transmitted back to the ground control station in real time through the Beidou short message communication module.

7. The single Beidou-based UAV edge intelligent analysis terminal according to claim 6 is characterized by: In the Beidou short message communication module, the RDSS service of the Beidou-2 satellite system is used, and the specific process includes: When the edge computing unit on the drone completes image processing and target detection, it generates target location and category label result data, compresses and encrypts the result data, and converts it into a format suitable for transmission via a satellite link. The encoded result data is sent to the Beidou short message communication module, which then converts the ground signal into a radio signal transmitted via a satellite and sends it to the Beidou satellite through the antenna.

8. The single Beidou-based UAV edge intelligent analysis terminal according to claim 7 is characterized in that: In the Beidou short message communication module, the process of realizing short message communication between the UAV and the ground control station includes: Utilizing the two-way communication function supported by RDSS, the module accepts requests from the drone and uploads them to the satellite. The Beidou satellite that receives the uplink signal forwards the request to the ground control station through relay. The ground control station is equipped with special receiving equipment to capture the downlink signal from the satellite, decode it and restore the original data content. The processed data is immediately displayed on the operating interface of the ground control station for the operator to view and make decisions, thereby realizing real-time monitoring of the drone mission execution.

9. The single Beidou-based UAV edge intelligent analysis terminal according to claim 8, characterized in that: In the ground control station, the process of using Qt to create a user interface for sending flight instructions and monitoring drone image information and positioning information includes: The user interface is designed using Qt, which is used to display the drone's status, image stream, and positioning information, and provides buttons and slider elements for interaction. It is used by the operator to send flight instructions to the drone. The Qt application is directly connected to the Beidou short message communication module, receives decoded data, and updates the UI in real time.

10. The single Beidou-based UAV edge intelligent analysis terminal according to claim 9 is characterized in that: The process of data visualization through Tableau and adopting safety protocols in the ground control station includes: The data sent back by the drone is stored on the local server, and Tableau in the ground control station visualizes it by accessing the data source. The SSL / TLS protocol is used to ensure the security of HTTP communication, and the security of the Beidou short message communication module is ensured by message authentication codes and digital signatures.

Images