Unmanned aerial vehicle full-spectrum autonomous inspection system based on visual perception

Abstract

The invention discloses an unmanned aerial vehicle full-spectrum autonomous inspection system based on visual perception, and belongs to the technical field of unmanned aerial vehicle inspection, and in the unmanned aerial vehicle full-spectrum autonomous inspection system, a self-adaptive environment module can sense changes of the surrounding environment, weather conditions, obstacles, topographic relief and the like in real time and quickly adapt to the changes. And the unmanned aerial vehicle can keep stable flight and accurate data acquisition in a complex and changeable inspection environment. When strong wind or sudden weather change occurs, the self-adaptive environment module can automatically adjust the flight attitude and speed of the unmanned aerial vehicle, safety is ensured, and task continuity is maintained. Besides, by continuously learning environment characteristics, the module can predict potential risks and plan an obstacle avoidance path in advance, so that the autonomy and reliability of the system are improved. The self-adaptive capability remarkably improves the inspection efficiency and safety of the unmanned aerial vehicle, and reduces the risk of task failure or equipment damage caused by environmental factors.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle inspection, and specifically is a full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception. Background Art

[0002] With the continuous development of science and technology and the growing demand for applications, drone technology has been widely used in many fields such as military, commercial and civilian. Especially in industries such as electricity, agriculture, forestry and security, drone full-spectrum autonomous inspection systems are gradually replacing traditional manual inspection methods as an efficient and convenient inspection method. However, existing drone inspection systems still have certain limitations when dealing with complex and changing environments and mission requirements. Most existing drone inspection systems use preset flight paths and fixed parameters for inspection.

[0003] However, this method often fails to adjust the flight strategy in time when faced with sudden environmental changes, resulting in poor inspection results or even mission failure. In addition, traditional drone inspection systems have deficiencies in environmental perception and data processing, making it difficult to achieve real-time analysis and effective use of inspection data.

[0004] Therefore, how to improve the adaptability and intelligence level of UAV inspection systems in complex environments has become an urgent problem to be solved in the current field of UAV technology. Summary of the invention

[0005] The purpose of the present invention is to provide a full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception in order to solve the above-mentioned problems.

[0006] The technical solution adopted by the present invention is as follows: a full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception, the system comprising: an unmanned aerial vehicle platform module, a full-spectrum imaging module, a data processing module, an adaptive learning planning module, a task control and management module and a ground control station module;

[0007] The adaptive learning planning module is internally provided with an adaptive environment module and an inspection path optimization module, and the inspection path optimization module includes a dynamic path planning module and a decision-making module;

[0008] The UAV platform module is connected to the full-spectrum imaging module, the data processing module, the adaptive learning planning module, the mission control and management module, and the ground control station module; the full-spectrum imaging module is directly embedded in the UAV platform module, and the data collected by it is initially analyzed by the data processing module and then transmitted to the adaptive learning planning module for further environmental adaptation and path optimization; the output instructions of the adaptive learning planning module are transmitted to the mission control and management module, which is responsible for executing these instructions and monitoring the entire inspection process, while maintaining communication with the ground control station module for real-time monitoring and data exchange;

[0009] The dynamic path planning module receives environmental information from the adaptive environment module and generates an optimized flight path, while the decision-making module makes specific flight decisions based on these paths and real-time environmental data. These decisions are then fed back to the mission control and management module to adjust the flight status of the drone.

[0010] In a preferred embodiment, the UAV platform module is responsible for carrying all sensors and equipment and performing stable flight in the mission area; it includes the aircraft body, power system, navigation and control system, communication system and safety protection system in emergency situations; the power system is composed of batteries, motors and propellers to ensure that the UAV has sufficient endurance; the navigation and control system uses GPS, GLONASS, and inertial measurement unit sensors for positioning and attitude control to ensure that the UAV can accurately fly along the preset route; the communication system is responsible for data exchange with the ground control station to ensure real-time monitoring and command transmission.

[0011] In a preferred embodiment, the full-spectrum imaging module is responsible for collecting multi-spectral and high-definition image data of the target area; the module includes a series of sensors of different bands, including visible light cameras, infrared cameras, multi-spectral cameras and thermal imaging cameras; these sensors can cover a wide spectral range from ultraviolet to infrared, thereby obtaining detailed information of the target object under different spectra; the imaging module is also equipped with a precise time synchronization device to ensure that all sensors capture data at the same time, so that accurate data fusion can be performed during subsequent processing.

[0012] In a preferred embodiment, the data processing module processes and analyzes the collected full-spectrum data in real time; it includes data preprocessing, image processing, feature extraction and pattern recognition submodules; the data preprocessing stage performs denoising, calibration and registration operations on the original image to improve data quality; the image processing stage uses various algorithms to enhance, segment and detect targets on the image to extract useful information; the feature extraction submodule extracts key features from the processed image, including color, shape, and texture; the pattern recognition submodule uses machine learning or deep learning models to perform target recognition, classification and anomaly detection based on the extracted features.

[0013] In a preferred embodiment, the adaptive environment module selects CNN convolutional neural network as the deep learning model and uses Q learning as the reinforcement learning algorithm to learn and identify different environmental patterns from the inspection data and automatically adjust the flight parameters accordingly, specifically including the following steps:

[0014] S1: The specific process of using CNN model training for the adaptive environment module includes:

[0015] Define the loss function, where the formula is: Among them: θ is the parameter of the CNN model; y i is the true label of the i-th sample; y^ i is the predicted label of the i-th sample; N is the number of training samples;

[0016] Select the Adam optimization algorithm to minimize the loss function:

[0017] S2: Use the trained CNN model to perform environmental pattern recognition on data collected in real time;

[0018] The environmental pattern recognition calculation formula includes:

[0019] Feature extraction:

[0020] Category: y^=g(φ)

[0021] Where g is the classification function, which is a softmax layer;

[0022] S3: Automatically adjust flight parameters based on recognized environmental patterns;

[0023] The flight speed adjustment formula is:

[0024] v new =v base +αv·Δv;

[0025] Among them, v new is the new flight speed, v baseis the basic flight speed, αv is the speed adjustment factor, and Δv is the speed adjustment amount;

[0026] The flight altitude adjustment formula is:

[0027] h new =h base+ αh·Δh;

[0028] Among them, h new is the new flight altitude, h base is the basic flight altitude, αh is the altitude adjustment coefficient, and Δh is the altitude adjustment amount;

[0029] The camera angle adjustment formula is:

[0030] θ new =θ base +αθ·Δθ;

[0031] Among them, θ new is the new flight angle, θ base is the basic flight angle, αθ is the angle adjustment coefficient, and Δθ is the angle adjustment amount;

[0032] S4: Real-time optimization strategy, using Q-learning algorithm to optimize flight paths and parameters in real time.

[0033] In a preferred embodiment, the use of the Q learning algorithm to optimize the flight path and parameters in real time specifically includes:

[0034] State estimation: S t =f(X t ;θ\*);

[0035] Among them, S t is the state estimate at time step t, X t is the observed data at time step t, θ\* is the optimal parameter of the state estimation function;

[0036] Action selection: Among them, A t is the action selected at time step t, and Q is the Q function used to evaluate the t The value of taking action a, θ Q is the parameter of the Q function;

[0037] Reward calculation: R t = r(S t ,A t ), where R t is the reward at time step t, r is the reward function, according to the state S t and action A t Calculate rewards;

[0038] The Q value is updated and the calculation formula is:

[0039] Q(S t ,A t )←Q(S t ,A t )+α·(R t +γ·a′maxQ(S t+1 ,a′)-Q(S t ,A t ));

[0040] Among them: α is the learning rate, which controls the update speed of the Q value; γ is the discount factor used to balance the immediate reward and future reward; a′ is the next state S t+1 The set of all possible actions.

[0041] In a preferred embodiment, the dynamic path planning module specifically includes:

[0042] S1: Environmental perception and status update: The sensors on board perceive environmental changes in real time and update the current status. The status update formula is:

[0043] St = {X t ,M t ,W t};

[0044] Among them, S t is the state of the system at time t, X t is the location information, M t is the environment pattern obtained from the deep learning model, W t It is the external factor of weather;

[0045] S2: Define an objective function to evaluate the quality of the flight path. Factors considered may include path length, safety distance, and data collection quality.

[0046] The objective function formula is:

[0047] Where J(π) is the total cost of path π, d(Pt) is the path length at time t, q(Dt) is the data quality, r(St) is the risk assessment, and a, β, γ are the corresponding weight coefficients;

[0048] S3: Dynamic Path Planning

[0049] Initialization: P0 = {S0}, where P0 is the initial path;

[0050] Select the next state:

[0051] St+1=argS′∈N(St)minJ(Pt∪{S′});

[0052] Among them, N(St) is the set of neighbor states of St;

[0053] Update path:

[0054] P t+1 =Pt∪{S t+1};

[0055] Where P t is the state set at time step t, which includes all states experienced up to time step t; S t+1 is the state at time step t+1, is the state observed or estimated at the current time step;

[0056] P t+1 is the updated state set at time step t+1, which is composed of P t and S t+1 Obtained through the union operation, including all the experienced states and the new state S t+1 .

[0057] In a preferred embodiment, the decision-making module makes a flight decision based on the current state and the environmental perception result. The decision-making formula is: t =π(S t ), where A t is the action at time t, π is the decision strategy;

[0058] When encountering special situations, the decision-making module quickly adjusts the inspection plan and replans the path with the formula: t+1 =Replan(P t ,S t ,W t );

[0059] The formula for adjusting the decision is: t =Adjust(A t ,W t );

[0060] The decision-making module executes decisions and provides feedback based on the execution results to optimize future decisions, where the execution and feedback formula is:

[0061] S {t+1} =f(S t ,A t );

[0062] π=Learn(S t+1 ,A t ,R t );

[0063] Among them, f is the state transition function, R_tRt is the reward at time t, and Learn is the policy learning function.

[0064] In a preferred embodiment, the task control and management module is responsible for planning tasks, scheduling resources and monitoring task execution; it includes a task planning system, a resource management system and a real-time monitoring system; the task planning system generates the optimal flight path and task sequence according to the inspection objectives and environmental conditions; the resource management system is responsible for allocating the power, computing resources and storage resources of the UAV to ensure efficient operation of the system; the real-time monitoring system tracks the flight status, equipment status and task progress of the UAV, and once an abnormality is found, takes immediate measures to adjust or re-plan the task.

[0065] In a preferred embodiment, the ground control station module is the operating interface and data center of the UAV full-spectrum autonomous inspection system, providing operators with a platform for task planning, monitoring and controlling UAVs; it includes two parts: hardware facilities and software systems; the hardware facilities include workstations, communication equipment, display screens and control panels; the software system provides user interfaces, data processing and analysis tools, task management software and communication protocols; operators monitor the flight status and imaging data of the UAV in real time through the ground control station, send flight instructions, adjust mission parameters, and receive inspection results sent back by the UAV; the ground control station also has data storage and backup functions to ensure the security and traceability of all inspection data.

[0066] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0067] 1. In the present invention, the adaptive environment module has produced significant beneficial effects on the entire UAV full-spectrum autonomous inspection system. First, the module can perceive changes in the surrounding environment, such as weather conditions, obstacles, terrain undulations, etc. in real time, and quickly adapt to these changes. This capability enables the UAV to maintain stable flight and accurate data collection in complex and changeable inspection environments. When encountering strong winds or sudden weather changes, the adaptive environment module can automatically adjust the flight attitude and speed of the UAV to ensure safety and maintain mission continuity. In addition, by continuously learning environmental characteristics, the module can predict potential risks and plan obstacle avoidance paths in advance, thereby improving the autonomy and reliability of the system. This adaptive capability significantly improves the efficiency and safety of UAV inspections and reduces the risk of mission failure or equipment damage due to environmental factors.

[0068] 2. In the present invention, the inspection path optimization module brings efficient path planning and decision support to the full-spectrum autonomous inspection system of the UAV. Through dynamic path planning, the module can optimize the flight route of the UAV according to the environmental data and task requirements collected in real time. This optimization not only reduces the flight distance and improves the inspection efficiency, but also ensures that the UAV covers all key detection points without missing any important information. In the inspection of power lines, the inspection path optimization module can plan the best inspection path according to the layout of the line and previous experience data, so as to timely discover potential fault points. At the same time, the module can also quickly adjust the path according to the actual situation during the inspection process. For example, when encountering an emergency or new inspection needs, the path can be quickly replanned to ensure the flexibility and adaptability of the task. These optimization measures greatly improve the operational efficiency and mission success rate of UAV inspections, reduce operating costs, and provide high-quality and comprehensive information support for subsequent data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 is the overall system block diagram of the present invention;

[0070] Figure 2 This is a system block diagram of the adaptive learning planning module in the present invention. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0072] Reference Figure 1-2 ,

[0073] Example:

[0074] A full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception, the system comprising: an unmanned aerial vehicle platform module, a full-spectrum imaging module, a data processing module, an adaptive learning planning module, a mission control and management module and a ground control station module;

[0075] The adaptive learning planning module is internally provided with an adaptive environment module and an inspection path optimization module, and the inspection path optimization module includes a dynamic path planning module and a decision-making module;

[0076] The UAV platform module is connected to the full-spectrum imaging module, data processing module, adaptive learning planning module, mission control and management module, and ground control station module. The full-spectrum imaging module is directly embedded in the UAV platform module. The data collected by it is initially analyzed by the data processing module and then transmitted to the adaptive learning planning module for further environmental adaptation and path optimization. The output instructions of the adaptive learning planning module are passed to the mission control and management module, which is responsible for executing these instructions and monitoring the entire inspection process, while maintaining communication with the ground control station module for real-time monitoring and data exchange.

[0077] The dynamic path planning module receives environmental information from the adaptive environment module and generates an optimized flight path, while the decision-making module makes specific flight decisions based on these paths and real-time environmental data. These decisions are then fed back to the mission control and management module to adjust the flight status of the drone.

[0078] The UAV platform module is the core carrier of the entire full-spectrum autonomous inspection system. It is responsible for carrying all sensors and equipment and performing stable flight in the mission area. This module usually includes the aircraft body, power system, navigation and control system, communication system, and safety protection system in emergency situations. The design of the aircraft body needs to consider lightness and strength to adapt to different flight environments and mission requirements. The power system usually consists of batteries, motors and propellers to ensure that the UAV has sufficient endurance. The navigation and control system uses sensors such as GPS, GLONASS, and inertial measurement unit (IMU) for positioning and attitude control to ensure that the UAV can accurately fly along the preset route. The communication system is responsible for data exchange with the ground control station to ensure real-time monitoring and command transmission. The safety protection system includes functions such as redundant design, automatic return and emergency landing to cope with emergencies.

[0079] The full-spectrum imaging module is the eye of the drone inspection system, responsible for collecting multi-spectral and high-definition image data of the target area. The module usually includes a series of sensors with different bands, such as visible light cameras, infrared cameras, multi-spectral cameras, and thermal imaging cameras. These sensors can cover a wide spectral range from ultraviolet to infrared, thereby obtaining detailed information about the target object under different spectra. The imaging module is also equipped with a precise time synchronization device to ensure that all sensors capture data at the same time so that accurate data fusion can be performed during subsequent processing. In addition, the imaging module also includes a lens cleaning system, a temperature control system, etc. to ensure that high-quality image data can be obtained even in complex environments.

[0080] The data processing module is the brain of the drone inspection system, responsible for real-time processing and analysis of the collected full-spectrum data. This module usually includes sub-modules such as data preprocessing, image processing, feature extraction and pattern recognition. The data preprocessing stage performs operations such as denoising, calibration and registration on the original image to improve data quality. The image processing stage uses various algorithms to enhance, segment and detect targets on the image to extract useful information. The feature extraction submodule extracts key features such as color, shape, texture, etc. from the processed image. The pattern recognition submodule uses machine learning or deep learning models to perform target recognition, classification and anomaly detection based on the extracted features. The data processing module must also have efficient data compression and transmission capabilities to ensure real-time performance and reduce bandwidth requirements.

[0081] The adaptive environment module selects CNN convolutional neural network as the deep learning model and uses Q learning as the reinforcement learning algorithm to learn and identify different environmental patterns from inspection data and automatically adjust flight parameters accordingly. The specific steps include:

[0082] S1: The specific process of using CNN model training for the adaptive environment module includes:

[0083] Define the loss function, where the formula is: Among them: θ is the parameter of the CNN model; y i is the true label of the i-th sample; y^ i is the predicted label of the i-th sample; N is the number of training samples;

[0084] Select the Adam optimization algorithm to minimize the loss function:

[0085] S2: Use the trained CNN model to perform environmental pattern recognition on data collected in real time;

[0086] The environmental pattern recognition calculation formula includes:

[0087] Feature extraction:

[0088] Category: y^=g(φ)

[0089] Where g is the classification function, which is a softmax layer;

[0090] S3: Automatically adjust flight parameters based on recognized environmental patterns;

[0091] The flight speed adjustment formula is:

[0092] v new =v base +αv·Δv;

[0093] Among them, v newis the new flight speed, v base is the basic flight speed, αv is the speed adjustment factor, and Δv is the speed adjustment amount;

[0094] The flight altitude adjustment formula is:

[0095] h new =h base+ αh·Δh;

[0096] Among them, h new is the new flight altitude, h base is the basic flight altitude, αh is the altitude adjustment coefficient, and Δh is the altitude adjustment amount;

[0097] The camera angle adjustment formula is:

[0098] θ new =θ base +αθ·Δθ;

[0099] Among them, θ new is the new flight angle, θ base is the basic flight angle, αθ is the angle adjustment coefficient, and Δθ is the angle adjustment amount;

[0100] S4: Real-time optimization strategy, using Q-learning algorithm to optimize flight paths and parameters in real time.

[0101] Using Q-learning algorithms to optimize flight paths and parameters in real time includes:

[0102] State estimation: S t =f(X t ;θ\*);

[0103] Among them, S t is the state estimate at time step t, X t is the observed data at time step t, θ\* is the optimal parameter of the state estimation function;

[0104] Action selection: Among them, A t is the action selected at time step t, and Q is the Q function used to evaluate the t The value of taking action a, θ Q is the parameter of the Q function;

[0105] Reward calculation: R t = r(S t ,A t ), where R t is the reward at time step t, r is the reward function, according to the state S t and action A t Calculate rewards;

[0106] The Q value is updated and the calculation formula is:

[0107] Q(S t ,A t )←Q(S t ,A t )+α·(R t +γ·a′maxQ(S t+1 ,a′)-Q(S t ,A t ));

[0108] Among them: α is the learning rate, which controls the update speed of the Q value; γ is the discount factor used to balance the immediate reward and future reward; a′ is the next state S t+1 The set of all possible actions.

[0109] The dynamic path planning module specifically includes:

[0110] S1: Environmental perception and status update: The sensors on board perceive environmental changes in real time and update the current status. The status update formula is:

[0111] St = {X t ,M t ,W t};

[0112] Among them, S t is the state of the system at time t, X t is the location information, M t is the environment pattern obtained from the deep learning model, W t It is the external factor of weather;

[0113] S2: Define an objective function to evaluate the quality of the flight path. Factors considered may include path length, safety distance, and data collection quality.

[0114] The objective function formula is:

[0115] Where J(π) is the total cost of path π, d(Pt) is the path length at time t, q(Dt) is the data quality, r(St) is the risk assessment, and a, β, γ are the corresponding weight coefficients;

[0116] S3: Dynamic Path Planning

[0117] Initialization: P0 = {S0}, where P0 is the initial path;

[0118] Select the next state:

[0119] St+1=argS′∈N(St)minJ(Pt∪{S′});

[0120] Among them, N(St) is the set of neighbor states of St;

[0121] Update path:

[0122] P t+1 =Pt∪{S t+1};

[0123] Where P t is the state set at time step t, which includes all states experienced up to time step t; S t+1 is the state at time step t+1, is the state observed or estimated at the current time step;

[0124] P t+1 is the updated state set at time step t+1, which is composed of P t and S t+1 Obtained through the union operation, including all the experienced states and the new state S t+1 .

[0125] The decision-making module makes flight decisions based on the current state and environmental perception results. The decision-making formula is: t =π(S t ), where A t is the action at time t, π is the decision strategy;

[0126] When encountering special situations, the decision-making module quickly adjusts the inspection plan and replans the route. The formula is: t+1 =Replan(P t ,S t ,W t );

[0127] The formula for adjusting the decision is: t =Adjust(A t ,W t );

[0128] The decision-making module executes decisions and provides feedback based on the execution results to optimize future decisions. The execution and feedback formula is:

[0129] S {t+1} =f(S t ,A t );

[0130] π=Learn(S t+1 ,A t ,R t );

[0131] Where f is the state transition function, R_tRt is the reward at time t, and Learn is the policy learning function;

[0132] Through the above process, the drone can dynamically adjust the flight route in real time according to the environmental perception results, and quickly adjust the inspection plan when encountering special situations to ensure the continuity of the mission and data quality. The parameters in the formula need to be defined and adjusted according to the actual application scenario.

[0133] The mission control and management module is the command center of the drone inspection system, responsible for planning tasks, scheduling resources, and monitoring the execution of tasks. This module includes a mission planning system, a resource management system, and a real-time monitoring system. The mission planning system generates the optimal flight path and mission sequence based on the inspection objectives and environmental conditions. The resource management system is responsible for allocating the power, computing resources, and storage resources of the drone to ensure efficient operation of the system. The real-time monitoring system tracks the flight status, equipment status, and mission progress of the drone, and takes immediate measures to adjust or re-plan the mission once an abnormality is found. In addition, this module also has mission recording and playback functions for post-analysis and evaluation.

[0134] The ground control station module is the operating interface and data center of the drone full-spectrum autonomous inspection system, providing operators with a platform for mission planning, monitoring and controlling drones. This module usually includes two parts: hardware facilities and software systems. Hardware facilities include workstations, communication equipment, display screens and control panels. The software system provides user interfaces, data processing and analysis tools, mission management software and communication protocols. Operators can monitor the flight status and imaging data of drones in real time through the ground control station, send flight instructions, adjust mission parameters, and receive inspection results sent back by drones. The ground control station also has data storage and backup functions to ensure the security and traceability of all inspection data.

[0135] From the above, it can be seen that in the present invention, the adaptive environment module has produced significant beneficial effects on the entire UAV full-spectrum autonomous inspection system. First, the module can perceive changes in the surrounding environment, such as weather conditions, obstacles, terrain undulations, etc. in real time, and quickly adapt to these changes. This capability enables the UAV to maintain stable flight and accurate data collection in complex and changeable inspection environments. When encountering strong winds or sudden weather changes, the adaptive environment module can automatically adjust the flight attitude and speed of the UAV to ensure safety and maintain mission continuity. In addition, by continuously learning environmental characteristics, the module can predict potential risks and plan obstacle avoidance paths in advance, thereby improving the autonomy and reliability of the system. This adaptive capability significantly improves the efficiency and safety of UAV inspections and reduces the risk of mission failure or equipment damage due to environmental factors.

[0136] In the present invention, the inspection path optimization module brings efficient path planning and decision support to the full-spectrum autonomous inspection system of the UAV. Through dynamic path planning, the module can optimize the flight route of the UAV according to the environmental data and task requirements collected in real time. This optimization not only reduces the flight distance and improves the inspection efficiency, but also ensures that the UAV covers all key detection points without missing any important information. In the inspection of power lines, the inspection path optimization module can plan the best inspection path according to the layout of the line and previous experience data, so as to timely discover potential fault points. At the same time, the module can also quickly adjust the path according to the actual situation during the inspection process. For example, when encountering an emergency or new inspection requirements, the path can be quickly replanned to ensure the flexibility and adaptability of the task. These optimization measures greatly improve the operational efficiency and mission success rate of UAV inspections, reduce operating costs, and provide high-quality and comprehensive information support for subsequent data analysis.

[0137] 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 term "comprise" or any other variant thereof is 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 elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0138] The above description enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in their embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception, characterized by: The system includes: a UAV platform module, a full-spectrum imaging module, a data processing module, an adaptive learning planning module, a mission control and management module, and a ground control station module; The adaptive learning planning module is internally provided with an adaptive environment module and an inspection path optimization module, and the inspection path optimization module includes a dynamic path planning module and a decision-making module; The UAV platform module is connected to the full-spectrum imaging module, the data processing module, the adaptive learning planning module, the mission control and management module, and the ground control station module; the full-spectrum imaging module is directly embedded in the UAV platform module, and the data collected by it is initially analyzed by the data processing module and then transmitted to the adaptive learning planning module for further environmental adaptation and path optimization; the output instructions of the adaptive learning planning module are transmitted to the mission control and management module, which is responsible for executing these instructions and monitoring the entire inspection process, while maintaining communication with the ground control station module for real-time monitoring and data exchange; The dynamic path planning module receives environmental information from the adaptive environment module and generates an optimized flight path, while the decision-making module makes specific flight decisions based on these paths and real-time environmental data. These decisions are then fed back to the mission control and management module to adjust the flight status of the drone.

2. The full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception as claimed in claim 1, characterized in that: The UAV platform module is responsible for carrying all sensors and equipment and performing stable flight in the mission area; It includes the aircraft body, power system, navigation and control system, communication system and safety protection system for emergency situations; the power system is composed of batteries, motors and propellers to ensure that the UAV has sufficient endurance; the navigation and control system uses GPS, GLONASS and inertial measurement unit sensors for positioning and attitude control to ensure that the UAV can accurately fly along the preset route; the communication system is responsible for data exchange with the ground control station to ensure real-time monitoring and command transmission.

3. The full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception as claimed in claim 1, characterized in that: The full-spectrum imaging module is responsible for collecting multi-spectral and high-definition image data of the target area; the module includes a series of sensors of different bands, including visible light cameras, infrared cameras, multi-spectral cameras and thermal imaging cameras; these sensors can cover a wide spectral range from ultraviolet to infrared, so as to obtain detailed information of the target object under different spectra; the imaging module is also equipped with a precise time synchronization device to ensure that all sensors capture data at the same time, so that accurate data fusion can be performed in subsequent processing.

4. The full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception as claimed in claim 1, characterized in that: The data processing module processes and analyzes the collected full-spectrum data in real time; it includes data preprocessing, image processing, feature extraction and pattern recognition submodules; The data preprocessing stage performs denoising, calibration and registration operations on the original image to improve data quality; the image processing stage uses various algorithms to enhance, segment and detect targets on the image to extract useful information; The feature extraction submodule extracts key features from the processed image, including color, shape, and texture; the pattern recognition submodule uses machine learning or deep learning models to perform target recognition, classification, and anomaly detection based on the extracted features.

5. The full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception as claimed in claim 1, characterized in that: The adaptive environment module adopts CNN convolutional neural network as a deep learning model and uses Q learning as a reinforcement learning algorithm to learn and identify different environmental patterns from inspection data and automatically adjust flight parameters accordingly. Specifically, the following steps are included: S1: The specific process of using CNN model training for the adaptive environment module includes: Define the loss function, where the formula is: Among them: θ is the parameter of the CNN model; y i is the true label of the i-th sample; y^ i is the predicted label of the i-th sample; N is the number of training samples; Select the Adam optimization algorithm to minimize the loss function: S2: Use the trained CNN model to perform environmental pattern recognition on data collected in real time; The environmental pattern recognition calculation formula includes: Feature extraction: Category: y^=g(φ) Where g is the classification function, which is a softmax layer; S3: Automatically adjust flight parameters based on recognized environmental patterns; The flight speed adjustment formula is: v new =v base +αv·Δv; Among them, v new is the new flight speed, v base is the basic flight speed, αv is the speed adjustment factor, and Δv is the speed adjustment amount; The flight altitude adjustment formula is: h new =h base+ αh·Δh; Among them, h new is the new flight altitude, h base is the basic flight altitude, αh is the altitude adjustment coefficient, and Δh is the altitude adjustment amount; The camera angle adjustment formula is: i new =θ base +θ·Δθ; Among them, θ new is the new flight angle, θ base is the basic flight angle, αθ is the angle adjustment coefficient, and Δθ is the angle adjustment amount; S4: Real-time optimization strategy, using Q-learning algorithm to optimize flight paths and parameters in real time.

6. The full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception as claimed in claim 5, characterized in that: The use of the Q learning algorithm to optimize the flight path and parameters in real time specifically includes: State estimation: S t =f(X t ;θ\*); Among them, S t is the state estimate at time step t, X t is the observed data at time step t, θ\* is the optimal parameter of the state estimation function; Action selection: Among them, A t is the action selected at time step t, and Q is the Q function used to evaluate the t The value of taking action a, θ Q is the parameter of the Q function; Reward calculation: R t = r(S t ,A t ), where R t is the reward at time step t, r is the reward function, according to the state S t and action A t Calculate rewards; The Q value is updated and the calculation formula is: Q(S t ,A t )←Q(S t ,A t )+α·(R t +γ·a′maxQ(S t+1 ,a′)-Q(S t ,A t )); Among them: α is the learning rate, which controls the update speed of the Q value; γ is the discount factor used to balance the immediate reward and future reward; a′ is the next state S t+1 The set of all possible actions.

7. The full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception as claimed in claim 1, characterized in that: When the dynamic path planning module actually works, it specifically includes the following steps: S1: Environmental perception and status update: The sensors on board perceive environmental changes in real time and update the current status. The status update formula is: St={X t ,M t ,W t }; Among them, S t is the state of the system at time t, X t is the location information, M t is the environment pattern obtained from the deep learning model, W t It is the external factor of weather; S2: Define an objective function to evaluate the quality of the flight path. Factors considered may include path length, safety distance, and data collection quality. The objective function formula is: Where J(π) is the total cost of path π, d(Pt) is the path length at time t, q(Dt) is the data quality, r(St) is the risk assessment, and a, β, γ are the corresponding weight coefficients; S3: Dynamic Path Planning Initialization: P0 = {S0}, where P0 is the initial path; Select the next state: St+1=argS′∈N(St)minJ(Pt∪{S′}); Among them, N(St) is the set of neighbor states of St; Update path: P t+1 =Pt∪{S t+1 } Where P t is the state set at time step t, which contains all states experienced up to time step t; S t+1 is the state at time step t+1, is the state observed or estimated at the current time step; P t+1 is the updated state set at time step t+1, which is composed of P t and S t+1 Obtained through the union operation, including all the experienced states and the new state S t+1 .

8. The full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception as claimed in claim 1, characterized in that: The decision-making module makes flight decisions based on the current state and environmental perception results. The decision-making formula is: t =π(S t ), where A t is the action at time t, π is the decision strategy; When encountering special situations, the decision-making module quickly adjusts the inspection plan and replans the path with the formula: t+1 =Replan(P t ,S t ,W t ); The formula for adjusting the decision is: t =Adjust(A t ,W t ); The decision-making module executes decisions and provides feedback based on the execution results to optimize future decisions, where the execution and feedback formula is: S {t+1} =f(S t ,A t ); π=Learn(S t+1 ,A t ,R t ); Among them, f is the state transition function, R_tRt is the reward at time t, and Learn is the policy learning function.

9. The full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception as claimed in claim 1, characterized in that: The task control and management module is responsible for planning tasks, scheduling resources and monitoring task execution; It includes a mission planning system, a resource management system and a real-time monitoring system. The mission planning system generates the optimal flight path and mission sequence based on inspection targets and environmental conditions. The resource management system is responsible for allocating the drone's power, computing resources, and storage resources to ensure efficient operation of the system; the real-time monitoring system tracks the drone's flight status, equipment status, and mission progress, and takes immediate measures to adjust or re-plan the mission if any abnormality is detected.

10. The full-spectrum autonomous inspection system for unmanned aerial vehicles based on visual perception as claimed in claim 1, characterized in that: The ground control station module is the operating interface and data center of the UAV full-spectrum autonomous inspection system, providing operators with a platform for mission planning, monitoring and controlling UAVs; It includes two parts: hardware facilities and software systems; the hardware facilities include workstations, communication equipment, display screens and control panels; The software system provides a user interface, data processing and analysis tools, mission management software, and communication protocols. The operator uses the ground control station to monitor the flight status and imaging data of the drone in real time, send flight instructions, adjust mission parameters, and receive inspection results sent back by the drone. The ground control station also has data storage and backup functions to ensure the security and traceability of all inspection data.

Images