Unmanned aerial vehicle riverway inspection system based on deep learning

Abstract

The invention provides an unmanned aerial vehicle riverway inspection system based on deep learning, and the system comprises a region division module which is used for dividing a to-be-inspected riverway inspection region into a plurality of sub-regions; the task allocation module is used for allocating a corresponding inspection task to the unmanned aerial vehicle in each divided sub-region; the scheduling algorithm module is used for calling a task scheduling algorithm to dynamically adjust distribution of the inspection tasks; the data processing and feedback module is used for calling a deep learning algorithm to carry out real-time processing on multi-source data collected when the unmanned aerial vehicle executes the inspection task, and feeding back the execution state of the inspection task; and the task execution module is used for guiding the unmanned aerial vehicle to perform flight operation according to the inspection task allocated by the unmanned aerial vehicle so as to execute the inspection task. The unmanned aerial vehicle inspection system is used for overcoming the defects of the unmanned aerial vehicle inspection system in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and particularly to an unmanned aerial vehicle river patrol system based on deep learning. Background Art

[0002] River patrol is an important part of water conservancy management. Traditional manual patrol methods have problems such as low efficiency, high cost, and being greatly affected by weather and other conditions. With the rapid development of unmanned aerial vehicle technology, unmanned aerial vehicles have become an efficient river patrol tool, capable of quickly covering large river areas and obtaining high-definition images and environmental data in real time. However, relying solely on manual or traditional image processing techniques, it is impossible to effectively achieve efficient analysis and anomaly recognition of large-scale river data. With the growing demand for river management and water environment protection, unmanned aerial vehicle technology has gradually become an important tool for river patrol due to its advantages such as high efficiency, flexibility, and low cost. By carrying devices such as high-definition cameras and infrared sensors, unmanned aerial vehicles can take real-time pictures of river areas, helping managers quickly obtain dynamic information of the river and promptly discover potential dangers and problems.

[0003] However, although unmanned aerial vehicles have shown great potential in river patrol, the existing unmanned aerial vehicle patrol systems still have some significant defects. Firstly, the patrol range is still limited by the flight ability and battery life of the unmanned aerial vehicle; secondly, it is unable to efficiently collect a large amount of multi-source data, and the accuracy of automated data analysis is still limited, relying on manual judgment and post-analysis; thirdly, it is overly dependent on preset flight paths and simple automated functions and cannot handle complex and dynamic patrol tasks. For example, when encountering complex river intersections, bridges, dams and other obstacles, the existing automatic flight algorithms are difficult to handle, which may lead to deviation of the flight route or failure to cover all areas that need to be inspected. Summary of the Invention

[0004] The present invention provides an unmanned aerial vehicle river patrol system based on deep learning to overcome the defects existing in the existing unmanned aerial vehicle patrol systems.

[0005] The present invention provides an unmanned aerial vehicle river patrol system based on deep learning, including: an area division module, a task assignment module, a scheduling algorithm module, a data processing and feedback module, and a task execution module. The area division module is used to divide the river patrol area to be inspected into multiple sub-areas; The task assignment module is used to assign corresponding patrol tasks to the unmanned aerial vehicle in each divided sub-area; The scheduling algorithm module is used to call a task scheduling algorithm to dynamically adjust the assignment of patrol tasks; The data processing and feedback module is used to call a deep learning algorithm to perform real-time processing on multi-source data collected when the drone executes an inspection task, and feedback the execution status of the inspection task; The task execution module is used to guide the drone to perform flight operations according to the inspection task assigned to the drone, so as to execute the inspection task.

[0006] In some embodiments, the task execution module is further used to convert the inspection task assignment scheme adjusted by the scheduling algorithm module into specific operations of the drone.

[0007] Further, the specific operations of the drone include: path planning, flight control, and task feedback; The process of path planning includes: planning the best inspection path of the drone according to the inspection task assignment scheme; The process of flight control includes: converting the instructions of path planning into flight control instructions to control the drone to fly according to the best inspection path; The process of task feedback includes: real-time monitoring of the execution status of the inspection task, and feedback of multi-source data and flight data collected when the drone executes the inspection task.

[0008] In some embodiments, in the data processing and feedback module, the process of performing real-time processing on multi-source data collected when the drone executes an inspection task includes: a data preprocessing process, an anomaly detection and recognition process, and a dynamic modeling and analysis process; In the data preprocessing process, the multi-source data collected when the drone executes the inspection task is subjected to data cleaning to obtain target data suitable for analysis by the deep learning algorithm; In the anomaly detection and recognition process, a deep learning algorithm is called to analyze the target data to obtain the anomaly detection result of the river channel environment; In the dynamic modeling and analysis process, a dynamic model of the river channel environment is constructed for the target data.

[0009] Further, the calling of the deep learning algorithm to analyze the target data to obtain the anomaly detection result of the river channel environment includes: Calling a deep learning algorithm to perform target detection on the target data to obtain the anomaly detection result of the river channel environment, where the deep learning algorithm includes at least one of the following: Image processing algorithm; target detection algorithm; The anomaly detection result of the river channel environment includes at least one of the following: River pollution sources, river obstacles, floods, illegal dumping of garbage.

[0010] In some embodiments, the process of real-time processing of multi-source data collected when the drone performs inspection tasks further includes: Using the Kalman filter data fusion algorithm, fuse the multi-source data collected when the drone performs inspection tasks, and generate a real-time river inspection report based on the fused data.

[0011] In some embodiments, the system further includes: a security mechanism module, and the security mechanism module is used to perform the following operations: Fault detection and self-recovery operations; emergency scheduling operations; data redundancy and backup operations.

[0012] Furthermore, the fault detection and self-recovery operations include: Real-time monitor the status of the system and perform fault detection; When a fault is detected, automatically adjust the current inspection task allocation plan and re-allocate inspection tasks.

[0013] Furthermore, the emergency scheduling operations include: When a sudden situation is detected, turn on the emergency scheduling mode of the system; Adjust the current inspection task allocation plan in the emergency scheduling mode to cope with sudden situations.

[0014] Furthermore, the data redundancy and backup operations include: When allocating inspection tasks and feedback data, according to the preset redundancy mechanism, back up the inspection tasks, multi-source data for feedback, and flight data.

[0015] The drone river inspection system based on deep learning provided by the present invention has the following beneficial effects: (1) Through the area division module and the task allocation module, the river inspection area to be inspected is divided into multiple sub-areas to allocate inspection tasks, overcoming the defect that the inspection range of the existing technology drone is still limited by the flight ability and battery life of the drone.

[0016] (2) Through the data processing and feedback module, call the deep learning algorithm to perform real-time processing on the multi-source data collected when the drone performs inspection tasks, and feedback the execution status of the inspection tasks. Overcoming the defects that the existing technology drones cannot efficiently collect a large amount of multi-source data, and the accuracy of automated data analysis is still limited, and manual judgment and post-analysis are required.

[0017] (3) The scheduling algorithm module calls the task scheduling algorithm to dynamically adjust the allocation of inspection tasks, overcoming the defect that the existing technology drones rely too much on preset flight paths and simple automation functions and cannot cope with complex and dynamic inspection tasks. Description of the Drawings

[0018] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce one by one the attached drawings required in the description of the embodiments or the prior art. Obviously, the attached drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other attached drawings can also be obtained based on these attached drawings.

[0019] Figure 1 It is one of the structural schematic diagrams of the UAV river channel inspection system based on deep learning provided by the present invention.

[0020] Figure 2 It is another structural schematic diagram of the UAV river channel inspection system based on deep learning provided by the present invention. Detailed implementation manners

[0021] To make the objectives, technical solutions and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the attached drawings in the present invention. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0022] The following describes the UAV river channel inspection system based on deep learning of the present invention in conjunction with the attached drawings. This system can be deployed or applied in a UAV control center and can perform real-time communication, data transmission and synchronization with the UAV controlled by the control center.

[0023] Figure 1 It is one of the structural schematic diagrams of the method of the UAV river channel inspection system based on deep learning provided by the present invention. As Figure 1 shown, the UAV river channel inspection system based on deep learning (hereinafter referred to as the system) includes: a region division module, a task assignment module, a scheduling algorithm module, a data processing and feedback module, and a task execution module. The following will explain them one by one.

[0024] The region division module is used to divide the river channel inspection region to be inspected into multiple sub-regions, so as to ensure the comprehensive coverage of the inspection tasks in each region. The result of the region division will be sent to the task assignment module.

[0025] The task assignment module is used to assign corresponding inspection tasks to the UAVs in each divided sub-region. Here, according to the result of the region division, combined with the flight ability and battery life of the UAVs, the inspection tasks are reasonably assigned and the assigned inspection tasks are sent to the task execution module to ensure that the UAVs can efficiently complete the inspection tasks of the entire river channel.

[0026] Thus, through the area division module and the task allocation module, the river inspection area to be inspected is divided into multiple sub-areas, and inspection tasks are allocated in the sub-areas, overcoming the defect that the inspection range of the UAV is still limited by the flight ability and battery life of the UAV.

[0027] The scheduling algorithm module is used to call the task scheduling algorithm to dynamically adjust the allocation of inspection tasks. By performing real-time scheduling and real-time adjustment of the inspection tasks of the UAV, and sending the allocation plan of the inspection tasks to the task allocation module for implementation, it can ensure the efficient and rapid completion of the inspection tasks. And the task scheduling algorithm therein can be trained according to the actual task scheduling data, so as to achieve performance optimization and improve the scheduling efficiency. Thus, it overcomes the defect that the UAV relies too much on the preset flight path and simple automation function and cannot cope with complex and dynamic inspection tasks.

[0028] The data processing and feedback module is used to call the deep learning algorithm to perform real-time processing on the multi-source data collected by the UAV during the inspection task and feedback the execution status of the inspection task. The UAV is equipped with multi-source sensors, including high-definition cameras, infrared sensors, lidar, temperature sensors, humidity sensors, etc. This enables the UAV to collect multi-source data of the river during the inspection task, including images, videos, temperature, and humidity information. Then the UAV will feedback these multi-source data to the data processing and feedback module, and the data processing and feedback module calls the deep learning algorithm to perform real-time processing on the multi-source data and feedback the execution status of the inspection task, such as feedback to the UAV control center. Thus, it can provide decision support for river environmental governance based on the processing results and execution status of the multi-source data, overcoming the defects that the UAV cannot efficiently collect a large amount of multi-source data, and the accuracy of automated data analysis is still limited, and it relies on manual judgment and post-analysis.

[0029] The task execution module is used to guide the UAV to perform flight operations according to the inspection tasks assigned to the UAV to execute the inspection tasks. After the corresponding inspection tasks are assigned by the task allocation module, the task execution module guides the UAV to perform flight operations in the corresponding river sub-area according to the inspection tasks assigned to the UAV to execute the inspection tasks, and collects multi-source data of the river through the multi-source sensors carried during the execution of the inspection tasks and feeds them back. Thus, it ensures the accuracy and safety of task execution.

[0030] In some embodiments, the system is designed with a flexible system architecture to ensure seamless cooperation between various modules, real-time data transmission and communication, and achieve a smooth connection between the drone and the control center. In response to the real-time and precision requirements of the task, in the scheduling algorithm module, the performance of the task scheduling algorithm is optimized to improve the computing efficiency, shorten the task completion time, and enhance the scheduling response speed. For the inspection task, high-performance drones and multi-source sensors suitable for river inspection tasks are selected, such as high-definition cameras, infrared thermal imagers, lidar, etc., to ensure the stability and accuracy of data collection. In the scheduling algorithm module and the data processing and feedback module, parallel computing and distributed processing technologies are used to execute the task scheduling algorithm and the deep learning algorithm, accelerating the task scheduling and data processing speed, and ensuring the real-time and efficiency of the system during large-scale inspections.

[0031] In some embodiments, in the task execution module of the system, it is also used to convert the inspection task allocation plan adjusted by the scheduling algorithm module into specific operations of the drone, that is, according to the allocated inspection task, determine the specific operations that the drone needs to perform during the inspection. The specific operations can be determining the flight route of the drone, determining flight data such as the flight speed and flight direction of the drone, collecting multi-source data, feeding back multi-source data and flight data, and so on. Thus, the mapping of the virtual inspection task allocation to the actual inspection task execution is realized, ensuring the smooth completion of the inspection task.

[0032] Furthermore, the specific operations of the drone include: path planning, flight control, and task feedback. Specifically, the process of path planning includes: according to the inspection task allocation plan, planning the best inspection path of the drone, avoiding repeated inspections or missed areas by the drone, and optimizing the flight time and energy consumption.

[0033] The process of flight control includes: converting the instructions of path planning into flight control instructions to control the drone to fly along the best inspection path. In this way, the flight speed and flight direction of the drone can be accurately controlled according to the flight control instructions, ensuring the stability and accuracy of the inspection process.

[0034] The process of task feedback includes: real-time monitoring of the execution status of the inspection task, and feeding back the multi-source data and flight data collected by the drone when performing the inspection task, supporting the optimization and adjustment of subsequent inspection tasks and scheduling. Among them, the multi-source data is the data collected by the multi-source sensors carried on the drone, and the flight data refers to the flight speed and flight direction, etc.

[0035] In some embodiments, in the data processing and feedback module, the process of real-time processing of the multi-source data collected by the drone when performing the inspection task is achieved by calling the deep learning algorithm. This process generally includes a data preprocessing process, an anomaly detection and recognition process, and a dynamic modeling and analysis process.

[0036] Specifically, first in the data preprocessing process, the multi-source data collected by the drone during the inspection task is cleaned to obtain the target data suitable for analysis by deep learning algorithms. The data preprocessing process includes denoising and normalization, which can improve the data quality and make the data suitable for analysis and processing by deep learning algorithms.

[0037] After the data preprocessing process, the anomaly detection and recognition process can be executed. In the anomaly detection and recognition process, deep learning algorithms are called to analyze the target data to obtain the anomaly detection results of the river channel environment.

[0038] The target data is multi-source data, including river channel images, river channel videos, etc. Analyzing these data can be to perform image recognition or image target detection. Therefore, deep learning algorithms can be called for image recognition or target detection on these image and video data, and possible anomalies in the river channel environment can be predicted, providing decision-making support for the environmental control and management of the river channel.

[0039] After the data preprocessing process, the dynamic modeling and analysis process can also be executed. In the dynamic modeling and analysis process, a dynamic model of the river channel environment is constructed for the target data, mapping the real river channel environment scene to a dynamic model in a virtual network, such as a twin model, etc., to achieve all-round monitoring of the river channel environment.

[0040] Furthermore, the process of real-time processing of the multi-source data collected by the drone during the inspection task also includes: using the Kalman filter data fusion algorithm to fuse the multi-source data collected by the drone during the inspection task, and generating a real-time river channel inspection report based on the fused data for feedback. The multi-source data collected includes images, videos, and temperature and humidity information collected by temperature and humidity sensors, and these information can be fused or superimposed. In addition, factors such as water flow, weather, and pollutant diffusion can be combined to generate a real-time river channel inspection report to predict and estimate the information of the future environmental state of the river channel, providing comprehensive information support for river channel inspection and management.

[0041] In some embodiments, a deep learning algorithm is invoked to analyze target data for anomaly detection in the river channel environment, including: invoking the deep learning algorithm to perform target detection on the target data to determine the anomalies in the river channel environment. The target data involves images or videos, which can be used to identify the specific environmental conditions of the river channel. For image data, the deep learning algorithm is directly invoked for target detection, and for videos, some key frames can be extracted for target detection. Target detection requires data processing on river channel images or videos, and the deep learning algorithm includes at least one of the following: image processing algorithm; target detection algorithm. Specifically, it can be a convolutional neural network model, a YOLO target detection model, or other models for target detection of images.

[0042] The anomaly detection results in the river channel environment include at least one of the following: river channel pollution sources, river channel obstacles, floods, and illegal waste dumping. Of course, in some other embodiments, there are also other abnormal situations, not limited to the above four abnormal situations. These abnormal situations can all be detected through the images and videos taken by drones and can be automatically identified. For example, the water quality condition of the river channel can be identified or whether a flood has occurred can be determined. For river channel pollution sources and illegal waste dumping, the accurate positions of the pollution sources and the waste can be predicted. For river channel obstacles, the type of the obstacle (caused by humans or nature) can be predicted.

[0043] According to actual requirements, finally, an analysis report can be generated based on the recognition results. For example, when a river channel pollution source is recognized, a corresponding pollution degree report can be generated, and when illegal waste dumping is recognized, a corresponding river channel sanitation report can be generated and fed back to the control center.

[0044] It should be noted that the deep learning algorithm has high adaptability and can adjust parameters according to different river channel environments, monitoring requirements, and task scales, supporting large-scale and multi-dimensional river channel inspection tasks. Even in complex weather, at night, or in low-light environments, the deep learning algorithm of the system can still operate efficiently, detect and recognize the images and videos taken by drones, and provide solutions for river channel monitoring in various complex environments.

[0045] In the embodiments of the present invention, a deep learning algorithm is invoked to perform target detection on target data to determine the anomalies in the river channel environment, which can automatically analyze and recognize the image and video data collected by drones, automatically detect pollution sources, floating objects, obstacles, illegal waste dumping, etc. in the river channel, and generate an analysis report, providing a scientific decision-making basis for river channel management and water resource protection, assisting in formulating prevention and control strategies and emergency response plans, and greatly improving the intelligent level of river channel inspection.

[0046] In some embodiments, such as Figure 2As shown, the deep learning-based UAV river channel inspection system further includes a security mechanism module. The security mechanism module is connected to any other module in the system and conducts data communication to give early warnings about security issues during the operation of the system and make timely adjustments to inspection tasks. The security mechanism module is used to perform the following operations: fault detection and self-recovery operations; emergency scheduling operations; data redundancy and backup operations. These operations are emergency measures taken for the system when sudden situations or faults occur in the system or the UAV, so as to ensure the security and reliability of the system operation.

[0047] In some embodiments, the fault detection and self-recovery operations of the security mechanism module include: real-time monitoring of the system status and conducting fault detection. When a fault is detected, automatically adjust the current inspection task allocation plan and re-allocate inspection tasks.

[0048] By real-time monitoring of the operation status of each module in the system during operation, quickly detect and handle possible faults. And through the self-recovery function, when a fault is detected, automatically adjust the current inspection task allocation plan in the scheduling algorithm module and re-allocate inspection tasks. This can ensure the normal execution of inspection tasks and prevent the inspection tasks from being affected by system faults.

[0049] In some embodiments, the emergency scheduling operations of the security mechanism module include: when a sudden situation is detected, turn on the emergency scheduling mode of the system. Adjust the current inspection task allocation plan in the emergency scheduling mode to cope with the sudden situation.

[0050] Through real-time monitoring of the river channel environment, when a sudden situation is detected, such as a sudden river channel accident or disaster, which will affect the inspection task of the UAV, at this time, turn on the emergency scheduling mode of the system and adjust the current inspection task allocation plan of the task allocation module in the emergency scheduling mode to cope with the sudden situation. For example, change the sub-inspection areas divided in the area division module, change the path planning of the inspection in the task execution module, etc. This can reduce or even eliminate the impact of sudden situations on the UAV inspection task and avoid delays in inspection tasks.

[0051] In some embodiments, the data redundancy and backup operations of the security mechanism module include: when allocating inspection tasks and giving feedback data, according to the preset redundancy mechanism, back up the inspection tasks, multi-source data for feedback, and flight data.

[0052] Since there is data communication among the drone, the system (each module), and the control center. For example, the inspection tasks assigned by the system to the drone, the task execution status, multi-source data, and flight data fed back by the drone to the system, and the river inspection report and analysis report fed back by the system to the control center. All these processes involve real-time data transmission. When the system fails or unexpected situations occur during the drone inspection, it may affect the real-time data transmission, resulting in intermittent data transmission or data loss.

[0053] Based on this, in the embodiment of the present invention, the security mechanism module adds a redundancy mechanism during the data transmission process. When allocating inspection tasks and feeding back data, according to the preset redundancy mechanism, the inspection tasks, multi-source data, and flight data for feedback are backed up. When the system fails or unexpected situations occur during the drone inspection, the backup data is used to continue data transmission and feedback, thereby avoiding intermittent data transmission or data loss, ensuring real-time data transmission, and ensuring the integrity and security of the data when the system fails or unexpected situations occur during the drone inspection.

[0054] For example, in the data processing and feedback module, due to a system failure, it is impossible to obtain the multi-source data collected by the drone and impossible to feed back the task execution status to the control center. At this time, according to the preset redundancy mechanism, when the multi-source data is transmitted and the task execution status is fed back, the multi-source data and the task execution status data are immediately backed up. When the transmission and feedback process is interrupted due to a failure, the backup multi-source data and task execution status data can still be used to continue the transmission and feedback, ensuring the integrity and security of the data.

[0055] In some embodiments, since the functions of each module, the deep learning algorithms, and the task scheduling algorithms invoked in the drone river inspection system based on deep learning are highly versatile. Therefore, the system is not only applicable to river inspection but can also be extended to environmental monitoring tasks in other water areas (such as lakes, wetlands, coastal areas, etc.). According to different monitoring scenarios and task requirements, the algorithms and functions are optimized, and by integrating with the existing monitoring system and control center, the performance and application value of the overall system are enhanced, supporting the optimization of multi-domain monitoring tasks.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A UAV river inspection system based on deep learning, comprising: The area division module, the task allocation module, the scheduling algorithm module, the data processing and feedback module, and the task execution module are characterized in that the area division module is used to divide the river inspection area to be inspected into multiple sub-areas; The task allocation module is used to allocate corresponding inspection tasks to the drone in each divided sub-area; The scheduling algorithm module is used to call the task scheduling algorithm to dynamically adjust the distribution of inspection tasks; The data processing and feedback module is used to call the deep learning algorithm to process the multi-source data collected when the drone performs the inspection task in real time, and to feedback the execution status of the inspection task; The task execution module is used to guide the drone to perform flight operations according to the inspection tasks assigned to the drone to execute the inspection tasks.

2. The deep learning-based UAV river inspection system according to claim 1 is characterized in that: The task execution module is also used to convert the inspection task allocation plan adjusted by the scheduling algorithm module into specific operations of the UAV.

3. The deep learning-based UAV river inspection system according to claim 2 is characterized in that: The specific operations of the UAV include: path planning, flight control and mission feedback; The path planning process includes: planning the best inspection path for the UAV according to the inspection task allocation plan; The flight control process includes: converting the path planning instructions into flight control instructions to control the UAV to fly according to the optimal inspection path; The task feedback process includes: real-time monitoring of the execution status of the inspection task, and feedback of multi-source data and flight data collected when the UAV performs the inspection task.

4. The deep learning-based UAV river inspection system according to claim 1 is characterized in that: In the data processing and feedback module, the process of real-time processing of multi-source data collected by the drone when performing inspection tasks includes: data preprocessing process, anomaly detection and identification process, and dynamic modeling and analysis process; In the data preprocessing process, the multi-source data collected by the drone when performing the inspection task is cleaned to obtain target data suitable for deep learning algorithm analysis; In the anomaly detection and identification process, a deep learning algorithm is called to analyze the target data to obtain an anomaly detection result of the river environment; In the dynamic modeling and analysis process, a dynamic model of the river environment is constructed based on the target data.

5. The deep learning-based UAV river inspection system according to claim 4 is characterized in that: The calling of the deep learning algorithm to analyze the target data to obtain an abnormality detection result of the river environment includes: Calling a deep learning algorithm to perform target detection on the target data to obtain an abnormality detection result of the river environment, wherein the deep learning algorithm includes at least one of the following: Image processing algorithm; object detection algorithm; The abnormal detection result of the river environment includes at least one of the following: Sources of river pollution, river obstructions, flooding, illegal dumping of garbage.

6. The deep learning-based UAV river inspection system according to claim 4 is characterized in that: The process of real-time processing of multi-source data collected when the drone performs inspection tasks also includes: The Kalman filter data fusion algorithm is used to fuse the multi-source data collected by the UAV when performing inspection tasks, and a real-time river inspection report is generated based on the fused data.

7. The deep learning-based UAV river inspection system according to claim 1 is characterized in that: The system further comprises: a safety mechanism module, wherein the safety mechanism module is used to perform the following operations: fault detection and self-recovery operation; emergency dispatch operation; data redundancy and backup operation.

8. The deep learning-based UAV river inspection system according to claim 7 is characterized in that: The fault detection and self-recovery operations include: Monitor the system status in real time and perform fault detection; When a fault is detected, the current inspection task allocation plan is automatically adjusted and the inspection tasks are reallocated.

9. The deep learning-based UAV river inspection system according to claim 7 is characterized in that: The emergency dispatch operation includes: When an emergency is detected, the system's emergency dispatch mode is turned on; Adjust the current inspection task allocation plan in emergency dispatch mode to deal with emergencies.

10. The deep learning-based UAV river inspection system according to claim 7 is characterized in that: The data redundancy and backup operations include: When allocating inspection tasks and feeding back data, the inspection tasks and the multi-source data and flight data for feedback are backed up according to the preset redundancy mechanism.

Images