Unmanned aerial vehicle inspection method and system based on artificial intelligence

Abstract

The invention belongs to the technical field of unmanned aerial vehicle inspection, and discloses an unmanned aerial vehicle inspection method and system based on artificial intelligence. The method comprises the following steps: a cloud data center constructs an unmanned aerial vehicle path planning model, an inspection image analysis model and an inspection scheme optimization model; the edge computing gateway performs unmanned aerial vehicle path planning by using an unmanned aerial vehicle path planning model according to a preset inspection scheme; the unmanned aerial vehicle performs routing inspection according to the real-time unmanned aerial vehicle routing inspection path and collects real-time routing inspection image data; the edge computing gateway is used for carrying out inspection image analysis by using an inspection image analysis model according to the real-time inspection image data; and the edge computing gateway optimizes a preset inspection scheme by using the inspection scheme optimization model according to the real-time inspection image analysis result. The problems of low inspection efficiency, poor inspection effect and lack of dynamic adjustment in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of UAV inspection, and particularly relates to a UAV inspection method and system based on artificial intelligence. Background Art

[0002] With the rapid development of UAV technology, UAVs are increasingly widely used in military, civilian, and commercial fields. Using UAVs for UAV inspection can adapt to harsh field environments, avoid safety accidents, improve inspection efficiency, and reduce labor cost input.

[0003] The existing technologies mainly have the following problems:

[0004] 1) Low inspection efficiency: The inspection paths of UAVs in many existing technologies are pre-set and cannot be dynamically adjusted according to real-time environmental changes, resulting in low efficiency;

[0005] 2) Poor inspection effect: Under long-distance or high-speed flight conditions, the image resolution collected by UAVs may be low, and existing image processing technologies are difficult to accurately identify targets, resulting in poor inspection effects;

[0006] 3) Lack of dynamic adjustment: Once the existing inspection plan is formulated, it is difficult to dynamically adjust according to the actual inspection situation, resulting in the inability to respond to emergencies in a timely manner. Summary of the Invention

[0007] In order to solve the problems of low inspection efficiency, poor inspection effect, and lack of dynamic adjustment existing in the prior art, the purpose of the present invention is to provide a UAV inspection method and system based on artificial intelligence.

[0008] The technical solution adopted by the present invention is as follows:

[0009] A UAV inspection method based on artificial intelligence includes the following steps:

[0010] A cloud data center uses artificial intelligence algorithms to construct a UAV path planning model, an inspection image analysis model, and an inspection plan optimization model, and deploys them to all edge computing gateways;

[0011] An edge computing gateway, according to a preset inspection plan, uses the UAV path planning model to perform UAV path planning, generates a real-time UAV inspection path, and sends it to the UAVs within the communication range;

[0012] UAVs, according to the real-time UAV inspection path, perform inspections, collect real-time inspection image data during the inspection process, and send it to the edge computing gateways within the communication range;

[0013] The edge computing gateway analyzes the real-time inspection images according to the real-time inspection image data, uses the inspection image analysis model to perform inspection image analysis, and obtains the real-time inspection image analysis results;

[0014] The edge computing gateway optimizes the preset inspection plan according to the real-time inspection image analysis results, uses the inspection plan optimization model to obtain the optimized inspection plan, and re-plans the UAV path.

[0015] Furthermore, the cloud data center uses artificial intelligence algorithms to construct a UAV path planning model, an inspection image analysis model, and an inspection plan optimization model, and deploys them to all edge computing gateways, including the following steps:

[0016] The cloud data center sets several historical inspection plans, and constructs a UAV path planning model according to several historical inspection plans using the swarm intelligence optimization algorithm;

[0017] Collect several historical inspection image data, and preprocess several historical inspection image data to obtain several preprocessed historical inspection image data;

[0018] According to several preprocessed historical inspection image data, use the image recognition algorithm to construct an inspection image analysis model, and generate several historical inspection image analysis results;

[0019] According to several historical inspection image analysis results, use the reinforcement learning algorithm to construct an inspection plan optimization model, and generate several historical inspection plan optimization experiences;

[0020] Extract the model metadata of the UAV path planning model, the inspection image analysis model, and the inspection plan optimization model, and send the model metadata to all edge computing gateways connected to the cloud data center.

[0021] Furthermore, the UAV path planning model is constructed based on the ISSA algorithm, and the UAV path planning model includes a sequentially connected inspection plan parsing module, an initialization module, an iterative optimization module, and a vector decoding module.

[0022] Furthermore, the inspection image analysis model is constructed based on the CNN-YOLOv6-DBN algorithm, and the inspection image analysis model includes an image feature extraction module constructed based on the CNN algorithm, an object detection module constructed based on the YOLOv6 algorithm, and an inspection image analysis module constructed based on the DBN algorithm, which are connected in sequence.

[0023] Furthermore, the inspection plan optimization model is constructed based on the PPO algorithm, and the inspection plan optimization model is provided with a policy network, an agent, and an experience replay pool.

[0024] Further, based on the analysis results of several historical inspection images, use the reinforcement learning algorithm to construct an inspection plan optimization model and generate several historical inspection plan optimization experiences, including the following steps:

[0025] Take the inspection plan optimization problem as the simulation environment of the reinforcement learning algorithm, and use the PPO algorithm to construct a policy network agent;

[0026] Use the experience replay mechanism to construct an experience replay pool, and define the action space of the agent according to several preset inspection plan optimization actions;

[0027] Analyze the historical inspection image analysis results to obtain several historical inspection image analysis states, and define the state space of the agent according to several historical inspection image analysis states;

[0028] Define the reward function of the agent according to the influence of the preset inspection plan optimization action on the historical inspection image analysis state;

[0029] Combine the experience replay pool, policy network, agent, action space, state space, and reward function to obtain an initial inspection plan optimization model;

[0030] Optimize and train the initial inspection plan optimization model according to several historical inspection image analysis results to obtain the final inspection plan optimization model and generate several historical inspection plan optimization experiences;

[0031] Store several historical inspection plan optimization experiences in the experience replay pool.

[0032] Further, the edge computing gateway uses the drone path planning model to perform drone path planning according to the preset inspection plan, generates a real-time drone inspection path, and sends it to the drones within the communication range, including the following steps;

[0033] The edge computing gateway uses the inspection plan parsing module of the drone path planning model to parse the preset inspection plan to obtain the real-time optimization target, real-time start position, real-time end position, and several real-time inspection positions;

[0034] Collect the real-time three-dimensional map of the area to be inspected, and based on the real-time three-dimensional map, according to the real-time start position, real-time end position, and several real-time inspection positions, use the initialization module of the drone path planning model to perform population initialization to obtain an initial ISSA population including several initial ISSA individuals;

[0035] Set the real-time fitness function according to the real-time optimization target, and based on the real-time fitness function, use the iterative optimization module of the drone path planning model to perform iterative optimization on the initial ISSA population to obtain the optimal individual with the minimum real-time fitness value;

[0036] Use the vector decoding module of the UAV path planning model to decode the individual vector of the optimal individual, obtain the optimal real-time UAV inspection path, and send the optimal real-time UAV inspection path to the UAVs within the communication range.

[0037] Furthermore, the edge computing gateway uses the inspection image analysis model to perform inspection image analysis on the real-time inspection image data, and the steps are as follows:

[0038] The edge computing gateway uses the image feature extraction module of the inspection image analysis model to extract the real-time image features of the real-time inspection image data;

[0039] Use the target detection module of the inspection image analysis model to perform target detection on the real-time inspection image data according to the real-time image features, and obtain several real-time detection targets;

[0040] Use the inspection image analysis module of the inspection image analysis model to perform inspection image analysis on several real-time detection targets according to the real-time image features, and obtain the real-time inspection image analysis results.

[0041] Furthermore, the edge computing gateway uses the inspection plan optimization model to optimize the preset inspection plan according to the real-time inspection image analysis results, obtain the optimized inspection plan, and re-plan the UAV path, and the steps are as follows:

[0042] The edge computing gateway analyzes the real-time inspection image analysis results to obtain several real-time inspection image analysis states, and updates the state space of the agent of the inspection plan optimization model according to the several real-time inspection image analysis states to obtain the updated state space;

[0043] Randomly extract several historical inspection plan optimization experiences from the experience replay pool of the inspection plan optimization model, obtain several real-time inspection plan optimization actions according to the several historical inspection plan optimization experiences, and update the action space of the agent of the inspection plan optimization model according to the several real-time inspection plan optimization actions to obtain the updated action space;

[0044] Based on the updated state space and the updated action space, use the agent of the inspection plan optimization model, the control strategy network, to generate the real-time inspection plan optimization strategy, and optimize the preset inspection plan according to the real-time inspection plan optimization strategy to obtain the optimized inspection plan, and re-plan the UAV path.

[0045] An unmanned aerial vehicle (UAV) inspection system based on artificial intelligence for implementing a UAV inspection method, characterized in that: the system includes a cloud data center, a number of edge computing gateways, and a number of UAVs. The cloud data center is communicatively connected to the number of edge computing gateways respectively, and each of the edge computing gateways is communicatively connected to a number of UAVs within its communication range.

[0046] The beneficial effects of the present invention are as follows:

[0047] The present invention discloses a UAV inspection method and system based on artificial intelligence. The UAV path planning model constructed based on the artificial intelligence algorithm can dynamically adjust the inspection path according to the real-time environmental changes, improving the inspection efficiency and adaptability. The UAV can effectively plan the path under complex terrain and changing meteorological conditions, avoiding obstacles and adverse weather areas to ensure the inspection safety. The inspection image analysis model constructed based on the artificial intelligence algorithm can process high-resolution images, maintaining the image quality even under long-distance or high-speed flight conditions, improving the accuracy of target recognition and the inspection effect. The inspection plan optimization model constructed based on the artificial intelligence algorithm can dynamically adjust the inspection plan according to the real-time inspection image analysis results, responding to emergencies in a timely manner and improving the inspection flexibility.

[0048] Other beneficial effects of the present invention will be further described in the specific implementation manners. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flowchart of the UAV inspection method based on artificial intelligence in the present invention.

[0050] Figure 2 is a structural block diagram of the UAV inspection system based on artificial intelligence in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0051] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0052] Embodiment 1:

[0053] As Figure 1 shown, this embodiment provides a UAV inspection method based on artificial intelligence, including the following steps:

[0054] S1: The cloud data center uses artificial intelligence algorithms to construct a UAV path planning model, an inspection image analysis model, and an inspection plan optimization model, and deploys them to all edge computing gateways, including the following steps:

[0055] S1-1: The cloud data center sets a number of historical inspection plans, and constructs a UAV path planning model according to the number of historical inspection plans using a swarm intelligence optimization algorithm;

[0056] The UAV path planning model is constructed based on the Improved Sparrow Search Algorithm (ISSA), and the UAV path planning model includes a patrol plan analysis module, an initialization module, an iterative optimization module, and a vector decoding module connected in sequence;

[0057] The patrol plan analysis module is used to analyze the input patrol plan, determine the optimization objectives and constraints of the patrol task, provide clear goals and limitations for subsequent path planning, ensure the practicality and effectiveness of the planning results, improve the pertinence and applicability of the algorithm, and adapt to different patrol task requirements; The initialization module is used to initialize the population according to the patrol plan analysis results. Usually, methods such as chaotic mapping are used to generate initial solutions, provide diverse initial solutions, increase the global search ability of the algorithm, and ensure the diversity and ergodicity of the population through methods such as chaotic mapping to avoid local optima; The iterative optimization module is used to perform iterative optimization using the Improved Sparrow Search Algorithm (ISSA), continuously update the individual positions in the population, and gradually approach the optimal solution through the iterative process to improve the quality of path planning. The unique mechanisms of the ISSA algorithm (such as the roles of discoverers, joiners, and predators) can enhance the exploration and exploitation capabilities of the population and balance global search and local search; The vector decoding module is used to decode the optimal individual finally obtained by the algorithm (usually a vector encoded in binary or real numbers) into a specific UAV patrol path, ensure that the path meets the actual requirements of UAV operations, such as path continuity and reachability, and transform the optimization results of the algorithm into an executable patrol path to realize the practical application of the algorithm;

[0058] S1-2: Collect a number of historical patrol image data, and preprocess the number of historical patrol image data to obtain a number of preprocessed historical patrol image data;

[0059] S1-3: According to a number of preprocessed historical patrol image data, use an image recognition algorithm to construct a patrol image analysis model and generate a number of historical patrol image analysis results;

[0060] The patrol image analysis model is constructed based on the Convolutional Neural Networks (CNN)-YOLOv6-Deep Belief Network (DBN) algorithm, and the patrol image analysis model includes an image feature extraction module constructed based on the CNN algorithm, an object detection module constructed based on the YOLOv6 algorithm, and a patrol image analysis module constructed based on the DBN algorithm connected in sequence;

[0061] The image feature extraction module automatically extracts local features in the image through a series of convolutional layers and pooling layers, and can effectively extract high-level features from the original image that are helpful for subsequent tasks; the target detection module uses the image features extracted by CNN and outputs detection results through the detection network of YOLOv6, achieving fast and accurate detection of targets in the image, suitable for scenarios with high real-time requirements. Through an end-to-end design, the detection process is simplified and the detection efficiency is improved; the inspection image analysis module receives the target detection results from YOLOv6 and further analyzes the attributes and states of the targets, enabling a more in-depth analysis of the detected targets;

[0062] S1-4: According to the analysis results of several historical inspection images, use the reinforcement learning algorithm to construct an inspection plan optimization model and generate several historical inspection plan optimization experiences;

[0063] The inspection plan optimization model is constructed based on the Proximal Policy Optimization (PPO) algorithm, and the inspection plan optimization model is set with a policy network, an agent, and an experience replay pool;

[0064] The policy network is the core of the PPO algorithm. It is used to learn an optimal policy, that is, the probability distribution of selecting the best action given the current state. By learning historical data, the policy network can provide effective decision support for the agent; the agent is the decision-making entity in the inspection plan optimization model. It selects actions according to the probability distribution provided by the policy network. The agent collects experiences through interactions with the environment, and these experiences are used to update the action space and state space. The agent can autonomously explore the space of inspection plans and find better inspection strategies; the experience replay pool is used to store the interaction experiences of the agent in the environment. These experiences include states, actions, rewards, and the next state. During the training process, the data in the experience replay pool is randomly extracted for updating the policy network. The experience replay pool breaks the correlation between data, reduces the variance in neural network training, and improves the stability and efficiency of learning;

[0065] S1-5: Extract the model metadata of the UAV path planning model, the inspection image analysis model, and the inspection plan optimization model, and send the model metadata to all edge computing gateways connected to the cloud data center;

[0066] S2: The edge computing gateway uses the UAV path planning model to perform UAV path planning according to the preset inspection plan, generates a real-time UAV inspection path, and sends it to the UAVs within the communication range, including the following steps:

[0067] S2-1: Take the inspection plan optimization problem as the simulation environment of the reinforcement learning algorithm, and use the PPO algorithm to construct a policy network agent;

[0068] S2-2: Use the experience replay mechanism to construct an experience replay pool, optimize actions according to several preset inspection plans, and define the action space of the agent;

[0069] S2-3: Analyze the historical inspection image analysis results to obtain several historical inspection image analysis states, and define the state space of the agent according to the several historical inspection image analysis states;

[0070] S2-4: Define the reward function of the agent according to the influence of the optimized actions according to the preset inspection plan on the historical inspection image analysis state;

[0071] S2-5: Combine the experience replay pool, policy network, agent, action space, state space, and reward function to obtain an initial inspection plan optimization model;

[0072] S2-6: Optimize and train the initial inspection plan optimization model according to several historical inspection image analysis results to obtain the final inspection plan optimization model, and generate several historical inspection plan optimization experiences;

[0073] S2-7: Store the several historical inspection plan optimization experiences in the experience replay pool;

[0074] S3: The drone conducts inspections according to the real-time drone inspection path, collects real-time inspection image data during the inspection, and sends it to the edge computing gateway within the communication range, including the following steps;

[0075] S3-1: The edge computing gateway uses the inspection plan parsing module of the drone path planning model to parse the preset inspection plan to obtain the real-time optimization target, real-time start position, real-time end position, and several real-time inspection positions;

[0076] In this embodiment, the real-time optimization target is to minimize the inspection cost;

[0077] S3-2: Collect the real-time three-dimensional map of the area to be inspected, and based on the real-time three-dimensional map, according to the real-time start position, real-time end position, and several real-time inspection positions, use the initialization module of the drone path planning model to perform population initialization to obtain an initial ISSA population including several initial ISSA individuals;

[0078] Use the initialization module of the drone path planning model to initialize with the Circle chaotic mapping sequence to obtain the initial ISSA population, and the initial ISSA population includes several initial ISSA individuals;

[0079] The formula is:

[0080]

[0081] In the formula, X' c is the initial ISSA individual of the Circle chaotic map; X c * is the randomly generated initial ISSA individual; c is the ISSA individual indicator;

[0082] S3-3: According to the real-time optimization objective, set the real-time fitness function, and based on the real-time fitness function, use the iterative optimization module of the UAV path planning model to perform iterative optimization on the initial ISSA population to obtain the optimal individual with the minimum real-time fitness value, including the following steps:

[0083] S3-3-1: According to the real-time optimization objective, set the real-time fitness function;

[0084] The formula of the real-time fitness function is:

[0085] f(X c ) = min H = α·A(X c ) + β·T(X c ) + ζ·L(X c ) + ψ

[0086] In the formula, f(X c ) is the real-time fitness value of the ISSA individual X c ; L(X c ) is the inspection distance cost of the ISSA individual X c ; A(X c ) is the inspection resource cost of the ISSA individual X c ; T(X c ) is the inspection time cost of the ISSA individual X c ; α, β, are all weight coefficients; X c is the ISSA individual variable; c is the ISSA individual indicator; H is the real-time optimization objective; ψ is the collision penalty parameter;

[0087] S3-3-2: Use the real-time fitness function to obtain the real-time fitness value of each initial ISSA individual in the initial ISSA population;

[0088] S3-3-3: According to the fitness values of the initial ISSA individuals, sort the initial ISSA individuals to obtain the initial discoverer, the initial joiner, and the initial predator;

[0089] S3-3-4: Update the initial ISSA population to obtain the updated ISSA population; the updated ISSA population includes the updated discoverer, the updated joiner, and the updated predator;

[0090] The update formula for the discoverer is as follows:

[0091]

[0092] Wherein, are the c-th discoverer ISSA individuals in the (t + 1)-th and t-th iterations respectively; t max is the maximum number of iterations; ξ is a random number between 0 and 1; Q is a random number with a normal distribution; L is a 1×D matrix whose elements are all 1; R 2 is the warning value; ST is the safety threshold;

[0093] The update formula for the joiner is as follows:

[0094]

[0095] Wherein, are the c-th joiner ISSA individuals in the (t + 1)-th and t-th iterations respectively; is the best position occupied by the exposed individual; is the current worst position; ξ is a random number between 0 and 1; L is a 1×D matrix whose elements are all 1 or -1; c is the sparrow indicator; h is the total number of ISSA individuals; A + is the update parameter;

[0096] The update formula for the predator is as follows:

[0097]

[0098] Wherein, are the c-th predator ISSA individuals in the (t + 1)-th and t-th iterations respectively; δ is the step size control parameter, and δ = a"·γ", a" is the convergence factor, and γ" is a non-zero positive real number for step size control; is the current best position; f c 、f g 、f w are the current, best, and worst fitnesses of the ISSA individual respectively; γ is the minimum constant to prevent the denominator from being 0;

[0099]

[0100] Wherein, a" is the convergence factor; tanh(.) is the hyperbolic tangent function; a max 、a min are the maximum and minimum values of the convergence factor respectively; λ is the decreasing rate parameter, k" is the decreasing period parameter, λ = -2π, k" = π;

[0101] S3-3-5: Use the dynamic reverse learning algorithm to perform dynamic reverse learning on the updated ISSA population to generate a dynamically reversed ISSA population;

[0102] The formula is:

[0103]

[0104] In the formula, is the dynamically reversed ISSA individual; γ * is the decreasing inertia coefficient; ub is the upper limit of the search space in the constraint condition; lb is the lower limit of the search space in the constraint condition; is the updated ISSA individual;

[0105] S3-3-6: According to the real-time fitness function, calculate the real-time fitness values of all ISSA individuals in the updated ISSA population and the dynamically reversed ISSA population, and take the ISSA individual with the minimum real-time fitness value as the optimal individual;

[0106] S3-3-7: If the iteration number of the algorithm reaches the maximum iteration number or the fitness value of the optimal individual meets the requirements, output the optimal individual;

[0107] S3-4: Use the vector decoding module of the UAV path planning model to decode the individual vector of the optimal individual to obtain the optimal real-time UAV inspection path, and send the optimal real-time UAV inspection path to the UAVs within the communication range;

[0108] S4: The edge computing gateway performs inspection image analysis on the real-time inspection image data using the inspection image analysis model, and the real-time inspection image analysis results are obtained, including the following steps:

[0109] S4-1: The edge computing gateway uses the image feature extraction module of the inspection image analysis model to extract the real-time image features of the real-time inspection image data; it can extract key features from the original image that are helpful for subsequent analysis. These features are more representative and discriminative than the original pixel data, greatly reducing the data dimension and simplifying the computational complexity of subsequent processing steps;

[0110] S4-2: Use the target detection module of the inspection image analysis model to perform target detection on the real-time inspection image data according to the real-time image features to obtain several real-time detection targets; the design of the YOLOv6 algorithm makes it very suitable for real-time target detection. It can complete the detection of multiple targets in the image in a short time. The algorithm can accurately identify and locate the targets in the image, providing accurate information for subsequent analysis. It can process multiple targets in the image simultaneously and is suitable for inspection tasks in complex scenarios;

[0111] S4-3: The inspection image analysis module using the inspection image analysis model analyzes the inspection images of several real-time detection targets based on the real-time image features to obtain the real-time inspection image analysis results;

[0112] The real-time inspection image analysis results include:

[0113] 1) Target detection results: Target location: The bounding box coordinates of the detected target in the image; Target category: The recognized target category, such as transmission line, tower, insulator, drone, bird, tree, etc.; Target confidence: The confidence score of the algorithm for the detected target belonging to a specific category

[0114] 2) Target status analysis: Anomaly detection: Identify whether there are anomalies in the target, such as damage, defect, foreign object attachment, etc.; Damage assessment: Evaluate the degree of damage to the target, which may include the location, type, and severity of the damage; Functional status: Evaluate the functional status of the target, such as whether the device is operating normally;

[0115] 3) Safety risk assessment: Safety hazard: Identify possible safety hazards, such as line slack, equipment detachment, etc.; Risk level: Give a risk assessment level according to the severity of the safety hazard;

[0116] S5: The edge computing gateway optimizes the preset inspection plan using the inspection plan optimization model according to the real-time inspection image analysis results to obtain an optimized inspection plan, and re-plans the UAV path, including the following steps:

[0117] S5-1: The edge computing gateway analyzes the real-time inspection image analysis results to obtain several real-time inspection image analysis states, and updates the state space of the agent of the inspection plan optimization model according to the several real-time inspection image analysis states to obtain an updated state space; Ensuring that the agent can make decisions based on the latest inspection situation improves the accuracy and real-time performance of the decision-making;

[0118] S5-2: Randomly extract several historical inspection plan optimization experiences from the experience replay pool of the inspection plan optimization model, obtain several real-time inspection plan optimization actions according to the several historical inspection plan optimization experiences, and update the action space of the agent of the inspection plan optimization model according to the several real-time inspection plan optimization actions to obtain an updated action space; Experience utilization: By reusing historical experiences, the learning efficiency is improved, the amount of data required for learning is reduced, and the updated action space contains more possible optimization actions, which helps to find a better inspection plan;

[0119] S5-3: Based on the updated state space and action space, use the agent of the inspection plan optimization model and the control policy network to generate a real-time inspection plan optimization strategy. According to the real-time inspection plan optimization strategy, optimize the preset inspection plan to obtain an optimized inspection plan, and re-plan the UAV path. It can adjust the inspection plan in real time according to the current inspection situation, improve the inspection efficiency and response speed, reduce manual intervention, realize the automatic optimization of the inspection plan, and improve the intelligent level of the overall inspection system;

[0120] The real-time inspection plan optimization strategy includes:

[0121] 1) Path adjustment action: Change the flight trajectory of the UAV to avoid obstacles or bad weather areas; optimize the flight path to reduce flight time or energy consumption;

[0122] 2) Speed adjustment: Adjust the flight speed of the UAV according to the requirements of the inspection task and environmental conditions;

[0123] 3) Task priority adjustment: According to the real-time analysis results, re-arrange the priorities of the inspection tasks, such as giving priority to inspecting potential problem areas.

[0124] Embodiment 2:

[0125] As Figure 2 shown, this embodiment provides an AI-based UAV inspection system for implementing the UAV inspection method, characterized in that: the system includes a cloud data center, several edge computing gateways, and several UAVs. The cloud data center is respectively communicatively connected to several edge computing gateways, and each edge computing gateway is respectively communicatively connected to several UAVs within its communication range;

[0126] The cloud data center is used to use artificial intelligence algorithms to construct a UAV path planning model, an inspection image analysis model, and an inspection plan optimization model, and deploy them to all edge computing gateways;

[0127] The edge computing gateway is used to perform UAV path planning according to the preset inspection plan, using the UAV path planning model, generate a real-time UAV inspection path, and send it to the UAVs within its communication range; perform inspection image analysis on the real-time inspection image data, using the inspection image analysis model, to obtain a real-time inspection image analysis result; according to the real-time inspection image analysis result, use the inspection plan optimization model to optimize the preset inspection plan to obtain an optimized inspection plan, and re-plan the UAV path;

[0128] The UAV is used to perform inspections according to the real-time UAV inspection path, collect real-time inspection image data during the inspection process, and send it to the edge computing gateway within its communication range.

[0129] The present invention discloses a method and system for drone inspection based on artificial intelligence. The drone path planning model constructed based on artificial intelligence algorithms can dynamically adjust the inspection path according to real-time environmental changes, improving the inspection efficiency and adaptability. The drone can effectively plan the path under complex terrains and changing meteorological conditions, avoiding obstacles and adverse weather areas to ensure the safety of inspection. The inspection image analysis model constructed based on artificial intelligence algorithms can process high-resolution images, maintaining the image quality even under long-distance or high-speed flight conditions, improving the accuracy of target recognition and the effect of inspection. The inspection plan optimization model constructed based on artificial intelligence algorithms can dynamically adjust the inspection plan according to the analysis results of real-time inspection images, promptly responding to emergencies and improving the flexibility of inspection.

[0130] The present invention is not limited to the above optional embodiments, and anyone can obtain other various forms of products under the inspiration of the present invention. The above specific embodiments should not be construed as limiting the protection scope of the present invention. The protection scope of the present invention should be defined by the claims, and the specification can be used to interpret the claims.

Claims

1. An artificial intelligence-based drone inspection method, characterized in that: The steps include: The cloud data center uses artificial intelligence algorithms to build drone path planning models, inspection image analysis models, and inspection plan optimization models, and deploys them to all edge computing gateways; The edge computing gateway uses the drone path planning model to plan the drone path according to the preset inspection plan, generate a real-time drone inspection path, and send it to the drones within the communication range; The drone conducts inspections according to the real-time drone inspection path, collects real-time inspection image data during the inspection process, and sends it to the edge computing gateway within the communication range; The edge computing gateway uses the inspection image analysis model to perform inspection image analysis based on the real-time inspection image data to obtain real-time inspection image analysis results; The edge computing gateway optimizes the preset inspection plan based on the real-time inspection image analysis results and uses the inspection plan optimization model to obtain the optimized inspection plan and re-plan the drone path.

2. The artificial intelligence-based drone inspection method according to claim 1, characterized in that: The cloud data center uses artificial intelligence algorithms to build drone path planning models, inspection image analysis models, and inspection plan optimization models, and deploys them to all edge computing gateways, including the following steps: The cloud data center sets up several historical inspection plans, and uses swarm intelligence optimization algorithms to build a drone path planning model based on these historical inspection plans; Collecting a number of historical inspection image data, and preprocessing the number of historical inspection image data to obtain a number of preprocessed historical inspection image data; Based on a number of pre-processed historical inspection image data, an inspection image analysis model is constructed using an image recognition algorithm, and a number of historical inspection image analysis results are generated; Based on the analysis results of several historical inspection images, a reinforcement learning algorithm is used to build an inspection plan optimization model and generate several historical inspection plan optimization experiences; The model metadata of the drone path planning model, inspection image analysis model, and inspection plan optimization model are extracted and sent to all edge computing gateways connected to the cloud data center.

3. The artificial intelligence-based drone inspection method according to claim 2 is characterized in that: The UAV path planning model is constructed based on the ISSA algorithm, and the UAV path planning model includes an inspection plan parsing module, an initialization module, an iterative optimization module and a vector decoding module which are connected in sequence.

4. The artificial intelligence-based drone inspection method according to claim 3 is characterized in that: The inspection image analysis model is constructed based on the CNN-YOLOv6-DBN algorithm, and the inspection image analysis model includes an image feature extraction module constructed based on the CNN algorithm, a target detection module constructed based on the YOLOv6 algorithm, and an inspection image analysis module constructed based on the DBN algorithm, which are connected in sequence.

5. The artificial intelligence-based drone inspection method according to claim 3 is characterized in that: The inspection scheme optimization model is constructed based on the PPO algorithm, and the inspection scheme optimization model is provided with a strategy network, an intelligent agent and an experience replay pool.

6. The artificial intelligence-based drone inspection method according to claim 5, characterized in that: Based on the analysis results of several historical inspection images, a reinforcement learning algorithm is used to build an inspection plan optimization model and generate several historical inspection plan optimization experiences, including the following steps: The inspection plan optimization problem is used as a simulation environment for the reinforcement learning algorithm, and the PPO algorithm is used to build a policy network agent. Use the experience replay mechanism to build an experience replay pool, optimize actions based on several preset inspection plans, and define the action space of the agent; Analyze the historical inspection image analysis results to obtain several historical inspection image analysis states, and define the state space of the intelligent agent based on the several historical inspection image analysis states; According to the preset inspection plan, the optimization action affects the historical inspection image analysis status and defines the reward function of the intelligent agent; Combining the experience replay pool, policy network, intelligent agent, action space, state space and reward function, we get the initial inspection plan optimization model. According to the analysis results of several historical inspection images, the initial inspection plan optimization model is optimized and trained to obtain the final inspection plan optimization model, and several historical inspection plan optimization experiences are generated; Store several historical inspection plan optimization experiences in the experience replay pool.

7. The artificial intelligence-based drone inspection method according to claim 6 is characterized in that: The edge computing gateway, based on the preset inspection plan, uses the drone path planning model to perform drone path planning, generate a real-time drone inspection path, and send it to the drones within the communication range, including the following steps; The edge computing gateway uses the inspection plan parsing module of the drone path planning model to parse the preset inspection plan and obtain the real-time optimization target, real-time start position, real-time end position, and several real-time inspection positions; A real-time three-dimensional map of the area to be inspected is collected, and based on the real-time three-dimensional map, the population is initialized using an initialization module of the UAV path planning model according to the real-time start position, the real-time end position and several real-time inspection positions, to obtain an initial ISSA population including several initial ISSA individuals; According to the real-time optimization goal, a real-time fitness function is set, and based on the real-time fitness function, the iterative optimization module of the UAV path planning model is used to iteratively optimize the initial ISSA population to obtain the optimal individual with the minimum real-time fitness value; The vector decoding module of the UAV path planning model is used to decode the individual vector of the optimal individual to obtain the optimal real-time UAV inspection path, and the optimal real-time UAV inspection path is sent to the UAVs within the communication range.

8. The artificial intelligence-based drone inspection method according to claim 7, characterized in that: The edge computing gateway uses the inspection image analysis model to perform inspection image analysis based on the real-time inspection image data to obtain the real-time inspection image analysis results, including the following steps: The edge computing gateway uses the image feature extraction module of the inspection image analysis model to extract real-time image features of the real-time inspection image data; Using the target detection module of the inspection image analysis model, target detection is performed on the real-time inspection image data according to the real-time image features to obtain several real-time detection targets; The inspection image analysis module of the inspection image analysis model is used to perform inspection image analysis on a number of real-time detection targets according to real-time image features to obtain real-time inspection image analysis results.

9. The artificial intelligence-based drone inspection method according to claim 8, characterized in that: The edge computing gateway optimizes the preset inspection plan based on the real-time inspection image analysis results and uses the inspection plan optimization model to obtain the optimized inspection plan and re-plan the drone path, including the following steps: The edge computing gateway analyzes the real-time inspection image analysis results to obtain a number of real-time inspection image analysis states, and updates the state space of the intelligent agent of the inspection scheme optimization model according to the real-time inspection image analysis states to obtain an updated state space; A number of historical inspection scheme optimization experiences are randomly extracted from the experience playback pool of the inspection scheme optimization model, and a number of real-time inspection scheme optimization actions are obtained based on the historical inspection scheme optimization experiences. The action space of the intelligent agent of the inspection scheme optimization model is updated based on the real-time inspection scheme optimization actions to obtain an updated action space. Based on the updated state space and the updated action space, the intelligent agent of the inspection plan optimization model is used to control the strategy network, generate a real-time inspection plan optimization strategy, and optimize the preset inspection plan according to the real-time inspection plan optimization strategy to obtain the optimized inspection plan, and re-plan the UAV path.

10. An artificial intelligence-based drone inspection system, used to implement the drone inspection method according to any one of claims 1 to 9, characterized in that: The system includes a cloud data center, several edge computing gateways and several drones. The cloud data center is respectively communicated with the several edge computing gateways, and each of the edge computing gateways is respectively communicated with several drones within the communication range.

Images