Unmanned aerial vehicle cruising method and system based on deep learning artificial intelligence image recognition algorithm

Abstract

The invention discloses an unmanned aerial vehicle cruising method and system based on a deep learning artificial intelligence image recognition algorithm. According to the method, an unmanned aerial vehicle carrying an improved YOLOv7-SwinT target recognition model collects real-time image data of an inspection area, and the model fuses a single-stage target detection architecture of YOLOv7 and a visual feature extraction network of Swin Transform. Progressive target detection is realized by adopting a three-level recognition architecture, wherein the progressive target detection comprises primary anomaly detection based on lightweight CNN, intermediate accurate positioning in combination with an attention mechanism and advanced target classification of multi-sensor data fusion. And the system combines the electric quantity of the unmanned aerial vehicle, the environmental condition and the task priority according to the identification result, generates a dynamic inspection path through an adaptive path planning algorithm, and realizes multi-vehicle collaborative operation by using an intelligent task allocation algorithm. In the inspection process, sensor data are processed in real time through edge computing equipment, and charging scheduling is optimized by adopting an intelligent energy management system. The target recognition precision and the cruising efficiency of the unmanned aerial vehicle in a complex environment are remarkably improved, and the method is suitable for application scenes such as electric power inspection and security monitoring.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of unmanned aerial vehicle intelligent inspection, and specifically relates to an unmanned aerial vehicle cruising method and system based on a deep learning artificial intelligence image recognition algorithm, and is particularly suitable for intelligent inspection of large-scale infrastructures such as offshore wind farms, power facilities, and pipelines. BACKGROUND

[0002] With the rapid development of unmanned aerial vehicle technology, its application in the fields of power inspection, security monitoring, and agricultural plant protection is increasingly widespread. However, existing unmanned aerial vehicle cruising systems still face many technical challenges:

[0003] Insufficient target detection accuracy: Traditional unmanned aerial vehicle inspection systems mostly use target detection algorithms based on convolutional neural networks, such as YOLOv3, Faster R-CNN, etc. These algorithms have low detection accuracy for small targets in complex backgrounds, making it difficult to meet the high-precision inspection requirements. Especially in harsh conditions such as long distances and low light, the detection performance significantly decreases.

[0004] Difficulties in multi-source data fusion: Existing systems usually rely only on a single visual sensor, lacking effective fusion of multi-modal data such as infrared thermal imaging and laser radar, resulting in insufficient robustness of target recognition in complex environments.

[0005] Low efficiency of real-time path planning: Traditional path planning algorithms (such as A*, RRT, etc.) are difficult to dynamically adapt to environmental changes and task requirements, and cannot achieve optimal path planning under multiple constraints such as power, environmental conditions, and task priority.

[0006] Limited computing resources: The computing power of unmanned aerial vehicle platforms is limited, and existing deep learning models have large parameters and high computational complexity, making it difficult to achieve real-time processing on edge devices.

[0007] Energy management is extensive: The current unmanned aerial vehicle energy management system lacks intelligent scheduling capability and cannot dynamically optimize the charging path according to the task progress and power state, resulting in low inspection efficiency.

[0008] To address the above problems, although some research has attempted to improve the target detection algorithm or optimize the path planning strategy, a complete solution has not yet been formed. In particular, there is still a lack of systematic innovation in multi-sensor data fusion, lightweight model design, and intelligent energy scheduling. Therefore, it is urgent to develop an unmanned aerial vehicle cruising system that integrates efficient target recognition, intelligent path planning, and optimized energy management to meet the growing demand of industry applications. SUMMARY

[0009] Therefore, it is necessary to provide a UAV cruising method and system based on a deep learning artificial intelligence image recognition algorithm, aiming to solve the technical defects of the existing UAV cruising system in target detection accuracy, multi-source data fusion, path planning efficiency, etc. The present application significantly improves the inspection efficiency and target recognition accuracy of the UAV in complex environments through an innovative three-level recognition architecture, multi-modal data fusion technology and adaptive path planning algorithm.

[0010] In a first aspect, the embodiments of the present application provide a UAV cruising method based on a deep learning artificial intelligence image recognition algorithm, comprising:

[0011] S1: collecting real-time image data of the inspection area by a UAV equipped with an improved YOLOv7-SwinT target recognition model, wherein the improved YOLOv7-SwinT target recognition model combines the single-stage target detection architecture of YOLOv7 and the visual feature extraction network of SwinTransformer, and adopts multi-sensor data fusion technology;

[0012] S2: based on the target information identified in step S1, combining the current power of the UAV, environmental conditions and task priority, generating a dynamic inspection path through an adaptive path planning algorithm;

[0013] S3: according to the inspection path generated in step S2, assigning inspection tasks to multiple UAVs through an intelligent task allocation algorithm, and monitoring the task execution in real time;

[0014] S4: during the inspection process, real-time processing of sensor data through an edge computing device, and transmitting key data to the control center;

[0015] S5: based on the real-time state of the UAV and the task progress, optimizing the scheduling of the charging base station and the energy allocation of the UAV through an intelligent energy management system; wherein the improved YOLOv7-SwinT target recognition model in step S1 adopts a three-level recognition architecture, including primary anomaly detection based on lightweight CNN, secondary accurate positioning combined with attention mechanism, and advanced target classification based on multi-sensor data fusion.

[0016] Optionally, in an implementation manner of the first aspect of the present application, the improved YOLOv7-SwinT target recognition model specifically comprises:

[0017] The YOLOv7 target detection framework includes: a backbone network constructed by deep separable convolution, an adaptive anchor point size selection module, and a multi-scale feature fusion pyramid structure.

[0018] The Swin Transformer feature extraction network comprises a hierarchical window attention mechanism, a cross-window connection module, a local perception enhancement module and a multi-sensor data fusion layer, and realizes feature-level fusion of visual images, infrared thermal images and laser radar point cloud data, feature weight distribution based on an attention mechanism and spatiotemporal alignment of multi-modal data correlation.

[0019] Optionally, in an implementation form of the first aspect of the present application, the deep separable convolution constructed backbone network comprises:

[0020] The feature extraction network is configured to extract multi-scale features from the input image, wherein at least one variability attention module is configured to enhance the correlation between distant pixels and improve the feature extraction capability of small targets.

[0021] The feature fusion network is configured to integrate the multi-scale features output by the feature extraction network, wherein at least one deformable convolution module is configured to expand the receptive field of the convolution kernel and reduce the model parameter quantity.

[0022] The detection head is configured to perform target detection based on the fused features and output the class, position and confidence of the target.

[0023] The loss calculation module is configured to calculate the loss function of the model during the training process, including: a positioning loss, which is based on EIoU to calculate the matching degree of the predicted box and the real box; a target confidence loss, which is used to measure the prediction accuracy of the model on the existence of the target; and a class prediction loss, which uses a label smoothing cross-entropy loss to optimize the classification result.

[0024] The system adopts Mosaic data enhancement, transfer learning and fine-tuning strategies in the training stage to improve the detection accuracy of small targets in the unmanned aerial vehicle patrol scene.

[0025] Optionally, in an implementation form of the first aspect of the present application, the loss function of the YOLOv7 is a multi-task loss function, which is composed of three parts: a positioning loss L CIoU , a target confidence loss L obj and a class prediction loss L cls .

[0026] The YOLOv7 uses an improved IoU loss function to calculate the positioning bounding box to obtain the regression error, and the positioning loss L CIoU is as follows:

[0027]

[0028] IoU is the intersection over union of the predicted box and the real box, and ρ 2 (b,b gt) is the Euclidean distance between the predicted box center point and the real box center point, l is the diagonal length of the minimum bounding rectangle, v is the consistency measure of the aspect ratio; α is the weight coefficient;

[0029] Target confidence loss L obj To measure whether the predicted box contains the target, the binary cross-entropy loss is used, and the formula is as follows:

[0030]

[0031] S 2 is the number of grids, B is the number of anchor boxes per grid, is an indicator function, which is 1 if anchor box j is responsible for detecting the target, otherwise it is 0, is the predicted confidence score;

[0032] Class prediction loss L cls To measure the accuracy of the predicted class, the binary cross-entropy loss is used, and the formula is as follows:

[0033]

[0034] where: is the real class label, is the predicted class probability, classes is the total number of classes, and c represents the class;

[0035] Total loss function L of YOLOv7 total is the weighted sum of the above three loss functions:

[0036] L total =λ coord L CIoU +λ obj L obj +λ cls L cls ,

[0037] where λ coord , λ obj , λ cls are hyperparameters used to balance the weights of different tasks;

[0038] When the YOLOv7 framework includes auxiliary heads, its total loss function is L total l :

[0039] L total l =L total +0.25L aux ,

[0040] where L aux is the loss function of the auxiliary head.

[0041] Optionally, in an implementation form of the first aspect of the application, the adaptive anchor size selection module comprises:

[0042] a K-means clustering unit configured to perform clustering analysis on the width and height sizes of all real bounding boxes in the training data set to generate an initial anchor size set;

[0043] an intersection over union optimization unit configured to calculate the matching degree of the predicted bounding box and the real bounding box by using an improved intersection over union IoU measurement method, and dynamically adjust the distribution of the anchor sizes;

[0044] a scale-sensitive screening unit configured to assign a specific scale anchor set to each feature layer according to the receptive field range of different feature layers in the target detection task;

[0045] a gradient backpropagation adaptation unit configured to incorporate the anchor size parameter into the network training process, and optimize the distribution of the anchor sizes through end-to-end learning;

[0046] wherein the gradient backpropagation adaptation unit automatically generates anchor sizes adapted to different detection scales by analyzing the geometric characteristics of the target data set, and improves the detection accuracy of the model for multi-scale targets.

[0047] Optionally, in an implementation form of the first aspect of the application, the multi-scale feature fusion pyramid structure comprises:

[0048] a bottom-up feature extraction path configured to extract multi-scale features of the input image layer by layer through a convolutional neural network to generate feature maps of different resolutions;

[0049] a top-down feature fusion path configured to perform step-by-step fusion of high-level semantic features and bottom-layer high-resolution features through up-sampling operation;

[0050] a horizontal connection module configured to adjust the number of feature channels by using a 1x1 convolution kernel to realize cross-scale connection of feature maps of different levels;

[0051] a feature enhancement unit configured to introduce an attention mechanism in the feature fusion process to dynamically adjust the weight distribution of different scale features;

[0052] a multi-level prediction output layer configured to set a detection head on each scale feature map after fusion to realize synchronous detection of small, medium and large targets of different sizes;

[0053] wherein the multi-scale feature fusion pyramid structure realizes complementary fusion of deep and shallow features through a bidirectional feature transmission path to improve the detection performance of the model for multi-scale targets.

[0054] Optionally, in an implementation form of the first aspect of the present application, based on the target information identified in step S1, in combination with the current power of the UAV, environmental conditions and task priority, a dynamic inspection path is generated through an adaptive path planning algorithm, including:

[0055] S2.1: Based on the target information identified in step S1, an inspection target priority evaluation model is established, which comprehensively considers: target abnormality degree score, confidence score output based on deep learning model; target type weight, pre-set priority coefficient of different types of targets; target timeliness parameter, time sensitivity factor calculated according to target change rate;

[0056] S2.2: A multi-constraint path planning model is constructed, and the input parameters include:

[0057] The current power state and estimated endurance time of the UAV; environmental condition parameters, including wind speed, rainfall intensity, and visibility; task priority matrix, calculated from step S2.1; geographic fence restriction conditions;

[0058] S2.3: An improved genetic algorithm is used to solve the optimal path, specifically including:

[0059] a) Initialization of population: generating N feasible paths as initial solutions;

[0060] b) Fitness calculation: constructing a multi-objective fitness function based on path length, task coverage, and energy consumption balance;

[0061] c) Selection operation: combining elite preservation strategy and roulette wheel selection;

[0062] d) Cross variation: designing a directional variation operator for inspection tasks;

[0063] e) Dynamic adjustment: real-time monitoring of environmental changes to trigger path re-planning;

[0064] S2.4: Output the optimized inspection path and dynamically update it to the UAV cruise system.

[0065] Optionally, in an implementation form of the first aspect of the present application, the three-level recognition architecture includes primary anomaly detection based on lightweight CNN, middle-level accurate positioning combined with attention mechanism, and high-level target classification based on multi-sensor data fusion, including:

[0066] The primary anomaly detection module uses a lightweight CNN network to quickly extract features from the input image, and realizes preliminary screening of abnormal areas through low-complexity convolutional layers;

[0067] The middle-level accurate positioning module is constructed based on a Swin Transformer structure and includes: a window multi-head self-attention unit for capturing long-distance spatial dependency; a cross-window feature interaction unit for realizing global context modeling through a shift window mechanism; and a dynamic receptive field adjustment unit for adaptively adjusting the attention range according to the target scale.

[0068] The advanced target classification module integrates a multi-source sensor data fusion mechanism and specifically includes: a visible light and infrared feature alignment unit adopting a cross-modal feature mapping network; a spatio-temporal context fusion unit for integrating time-series sensor data through a 3D convolution; and a multi-task decision layer for synchronously outputting target category, confidence and spatial position information.

[0069] The cascaded feature transmission path realizes feature reuse and enhancement among the three-level modules through a residual connection and a feature pyramid structure.

[0070] The architecture realizes high-precision detection and classification of multi-scale targets in a UAV inspection scene under the condition of limited computing resources through a progressive recognition strategy.

[0071] In a second aspect, the embodiments of the present application provide a UAV cruising system based on a deep learning artificial intelligence image recognition algorithm, which is applied to the UAV cruising method based on the deep learning artificial intelligence image recognition algorithm as described in the first aspect, and the system includes:

[0072] An image acquisition module is configured to acquire real-time image data of an inspection area by a UAV carrying an improved YOLOv7-SwinT target recognition model, the improved YOLOv7-SwinT target recognition model fuses a single-stage target detection architecture of YOLOv7 and a visual feature extraction network of Swin Transformer, and adopts a multi-sensor data fusion technology.

[0073] A path planning module is configured to generate a dynamic inspection path through an adaptive path planning algorithm based on the target information identified in step S1, in combination with the current power of the UAV, environmental conditions and task priorities.

[0074] A task allocation module is configured to allocate inspection tasks to multiple UAVs through an intelligent task allocation algorithm according to the inspection path generated in step S2, and to monitor the task execution in real time.

[0075] A data processing module is configured to process sensor data in real time through an edge computing device during the inspection process, and to transmit key data to a control center.

[0076] An energy management module is configured to optimize the scheduling of charging stations and the distribution of energy of the UAVs based on the real-time status of the UAVs and the progress of the tasks through an intelligent energy management system.

[0077] In a third aspect, an electronic device is provided, comprising:

[0078] a processor;

[0079] a memory for storing processor-executable instructions;

[0080] The processor is configured to implement the UAV cruising method based on the deep learning artificial intelligence image recognition algorithm when executing the instructions.

[0081] In a fourth aspect, a computer-readable storage medium is provided, which stores a program instructing a device to execute the UAV cruising method based on the deep learning artificial intelligence image recognition algorithm.

[0082] A UAV cruising method and system based on a deep learning artificial intelligence image recognition algorithm are disclosed. The method collects real-time image data of the inspection area by a UAV equipped with an improved YOLOv7-SwinT target recognition model. The model combines the single-stage target detection architecture of YOLOv7 and the visual feature extraction network of Swin Transformer, and adopts a three-level recognition architecture to realize progressive target detection: primary abnormality detection based on lightweight CNN, intermediate accurate positioning combined with attention mechanism, and advanced target classification based on multi-sensor data fusion. The system generates a dynamic inspection path through an adaptive path planning algorithm based on the recognition results, the power of the UAV, the environmental conditions, and the task priority, and realizes multi-UAV collaborative operation through an intelligent task allocation algorithm. In the inspection process, the sensor data are processed in real time by an edge computing device, and the charging scheduling is optimized through an intelligent energy management system. The method significantly improves the target recognition accuracy and cruising efficiency of the UAV in complex environments, and is suitable for applications such as power inspection and security monitoring.

[0083] Advantages:

[0084] (1)Target detection performance is significantly improved. Through a three-stage progressive recognition architecture (lightweight CNN preliminary screening + SwinTransformer precise positioning + multi-sensor fusion classification), the small target detection accuracy is improved; the improved CIoU loss function introduces the aspect ratio consistency measurement to improve the boundary box positioning accuracy. Multi-modal data fusion technology improves the recognition robustness in complex environments.

[0085] (2) System response efficiency is optimized. The combination of deep separable convolution and deformable convolution reduces the number of parameters and improves the inference speed. The edge computing architecture realizes millisecond-level real-time data processing, reduces the system response delay, and improves the overall inspection efficiency through multi-machine cooperative task allocation algorithm.

[0086] (3) Path planning is intelligent. The improved genetic algorithm combined with the multi-objective optimization model shortens the path planning time, and the dynamic re-planning mechanism improves the environmental adaptability and responds to sudden conditions such as wind speed changes in real time. The task priority evaluation model improves the key target detection rate.

[0087] (4) Energy management is refined. The intelligent charging scheduling system prolongs the effective operation time of the unmanned aerial vehicle, the energy consumption prediction algorithm realizes accurate power management, reduces the invalid flight distance, and the multi-machine cooperative charging strategy improves the overall energy utilization rate.

[0088] (5) Application adaptability is enhanced. Visible light / infrared / laser radar multi-source data input is supported, environmental adaptability is improved, modular design can flexibly configure detection models according to different task requirements, lightweight models are adapted to mainstream edge computing devices, and deployment costs are reduced. BRIEF DESCRIPTION OF DRAWINGS

[0089] Figure 1 The unmanned aerial vehicle cruising method based on the deep learning artificial intelligence image recognition algorithm provided by an embodiment of the present application is shown in the flowchart.

[0090] Figure 2 The YOLOv7-SwinT model structure diagram provided by an embodiment of the present application is shown.

[0091] Figure 3 The unmanned aerial vehicle cruising system module schematic diagram based on the deep learning artificial intelligence image recognition algorithm provided by an embodiment of the present application is shown.

[0092] Figure 4 The electronic device schematic diagram provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0093] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application.

[0094] It should be noted that the "at least one" in the embodiments of the present application means one or more, and more means two or more. Unless otherwise defined, all technical and scientific terms used in the present application have the same meanings as those commonly understood by those skilled in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments and are not intended to limit the present application.

[0095] It should be noted that the "first", "second", and the like in the embodiments of the present application are only used for the purpose of distinguishing description, and cannot be understood as indicating or implying relative importance, nor can it be understood as indicating or implying sequence. The features limited by "first", "second" can explicitly or implicitly include one or more of the features. In the description of the embodiments of the present application, the words "exemplary" or "for example" are used to mean serving as an example, illustration, or description. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the use of "exemplary" or "for example" is intended to present the relevant concept in a specific manner.

[0096] Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0097] Embodiment one

[0098] Figure 1 The flowchart of the unmanned aerial vehicle cruising method based on the deep learning artificial intelligence image recognition algorithm provided by an embodiment of the present application is shown.

[0099] As shown in Figure 1 , an unmanned aerial vehicle cruising method based on a deep learning artificial intelligence image recognition algorithm comprises:

[0100] S1: Collecting real-time image data of a patrol area by an unmanned aerial vehicle carrying an improved YOLOv7-SwinT target recognition model, wherein the improved YOLOv7-SwinT target recognition model fuses the single-stage target detection architecture of YOLOv7 and the visual feature extraction network of SwinTransformer, and adopts a multi-sensor data fusion technology.

[0101] The improved YOLOv7-SwinT target recognition model is used for image acquisition, wherein the improved YOLOv7-SwinT model: the model fuses the single-stage target detection architecture of YOLOv7 and the visual feature extraction network of Swin Transformer, and simultaneously adopts a multi-sensor data fusion technology (visual image, infrared thermal imaging, and laser radar point cloud data).

[0102] The three-level recognition architecture includes: primary abnormality detection based on a lightweight CNN for quickly identifying abnormal targets; middle-level accurate positioning combined with an attention mechanism to improve target positioning accuracy; and high-level target classification through multi-sensor data fusion to realize feature-level fusion and spatiotemporal alignment of multi-modal data correlation.

[0103] Figure 2 A YOLOv7-SwinT model structure diagram provided for an embodiment of the present application is shown in FIG. 1. Figure 2 As shown in FIG. 1, specifically, the improved YOLOv7-SwinT target recognition model specifically includes:

[0104] The YOLOv7 target detection framework includes a backbone network constructed by a depth separable convolution, an adaptive anchor point size selection module, and a multi-scale feature fusion pyramid structure.

[0105] In the embodiment, the YOLOv7 framework is mainly divided into three parts, namely, a backbone network constructed by a depth separable convolution to improve computational efficiency, an adaptive anchor point size selection module to optimize anchor box matching and improve detection accuracy, and a multi-scale feature fusion pyramid structure to enhance the detection ability of the model for targets of different scales.

[0106] The Swin Transformer feature extraction network includes a hierarchical window attention mechanism, a cross-window connection module, a local perception enhancement module, and a multi-sensor data fusion layer, which realizes feature-level fusion of visual images, infrared thermal imaging, and laser radar point cloud data, feature weight distribution based on an attention mechanism, and spatiotemporal alignment of multi-modal data correlation.

[0107] In the Swin Transformer feature extraction network, the hierarchical window attention mechanism is used to improve the feature extraction capability, the cross-window connection module is used to enhance the feature fusion effect, the local perception enhancement module is used to improve the perception ability of the model for local details, and the multi-sensor data fusion layer is used to realize feature-level fusion of visual images, infrared thermal imaging, and laser radar point cloud data, feature weight distribution based on an attention mechanism, and spatiotemporal alignment of multi-modal data correlation.

[0108] The improved YOLOv7-SwinT target recognition model in the step S1 adopts a three-level recognition architecture, including primary anomaly detection based on a lightweight CNN, middle-level accurate positioning combined with an attention mechanism, and high-level target classification based on multi-sensor data fusion.

[0109] Specifically, the three-level recognition architecture includes primary anomaly detection based on a lightweight CNN, middle-level accurate positioning combined with an attention mechanism, and high-level target classification based on multi-sensor data fusion, and includes the following.

[0110] The primary anomaly detection module adopts a lightweight CNN network to quickly extract features from an input image, and realizes preliminary screening of an abnormal area through a low-complexity convolutional layer.

[0111] The middle-level accurate positioning module is constructed based on a Swin Transformer structure and includes a window multi-head self-attention unit for capturing long-distance spatial dependency, a cross-window feature interaction unit for realizing global context modeling through a shift window mechanism, and a dynamic receptive field adjustment unit for adaptively adjusting the attention range according to the target scale.

[0112] The high-level target classification module integrates a multi-source sensor data fusion mechanism and specifically includes a visible light and infrared feature alignment unit adopting a cross-modal feature mapping network, a space-time context fusion unit integrating time-series sensor data through a 3D convolution, and a multi-task decision layer synchronously outputting target category, confidence and spatial position information.

[0113] A cascaded feature transmission path realizes feature reuse and enhancement among the three-level modules through a residual connection and a feature pyramid structure.

[0114] The architecture realizes high-precision detection and classification of multi-scale targets in a UAV inspection scene under the condition of limited computing resources through a progressive recognition strategy.

[0115] The backbone network constructed by the depth separable convolution includes:

[0116] A feature extraction network is configured to extract multi-scale features from an input image, wherein at least one variability attention module is included to enhance the correlation between distant pixels and improve the feature extraction capability of small targets.

[0117] A feature fusion network is configured to integrate the multi-scale features output by the feature extraction network, wherein at least one deformable convolution module is included to expand the receptive field of the convolution kernel and reduce the model parameter amount.

[0118] A detection head is configured to perform target detection based on the fused features and output the category, position and confidence of the target.

[0119] The loss calculation module is configured to calculate a loss function of the model during the training process, including: a positioning loss, which is configured to calculate a matching degree of a predicted bounding box and a real bounding box based on EIoU; a target confidence loss, which is configured to measure a prediction accuracy of the model on whether a target exists; and a category prediction loss, which is configured to optimize a classification result by using a label smoothing cross-entropy loss.

[0120] In the training stage, the system adopts a Mosaic data enhancement, a transfer learning and a fine-tuning strategy to improve the detection accuracy of small targets in a UAV patrol scene.

[0121] The loss function of the YOLOv7 is a multi-task loss function, which is composed of three parts: a positioning loss L CIoU , a target confidence loss L obj and a category prediction loss L cls .

[0122] The YOLOv7 uses an improved IoU loss function to calculate a positioning bounding box, obtains a regression error, and the positioning loss L CIoU is as follows:

[0123]

[0124] IoU is an intersection over union of a predicted bounding box and a real bounding box, ρ 2 (b,b gt ) is an Euclidean distance between a center point of the predicted bounding box and a center point of the real bounding box, l is a diagonal length of a minimum circumscribed rectangle, and v is a consistency measure of an aspect ratio; and α is a weight coefficient.

[0125] The target confidence loss L obj is configured to measure whether a predicted bounding box contains a target, and is calculated by using a binary cross-entropy loss, and the formula is as follows:

[0126]

[0127] S 2 is a grid number, B is an anchor box number of each grid, is an indicator function, which is 1 if an anchor box j is responsible for detecting a target, and is 0 otherwise, is a predicted confidence score.

[0128] The category prediction loss L cls is configured to measure an accuracy of a predicted category, and is calculated by using a binary cross-entropy loss, and the formula is as follows:

[0129]

[0130] wherein: is a real category label, is a predicted category probability, classes is a total number of categories, and c represents a category.

[0131] The total loss function L of YOLOv7 total is the weighted sum of the above three loss functions:

[0132] L total =λ coord L CIoU +λ obj L obj +λ cls L cls ,

[0133] where λ coord , λ obj , λ cls are hyperparameters to balance the weights of different tasks.

[0134] When the YOLOv7 framework includes an auxiliary head, its total loss function is L total l :

[0135] L total l =L total +0.25L aux ,

[0136] where L aux is the loss function of the auxiliary head.

[0137] The adaptive anchor size selection module comprises:

[0138] A K-means clustering unit is configured to perform clustering analysis on the width and height sizes of all real annotation boxes in a training data set to generate an initial anchor size set.

[0139] An intersection over union optimization unit is configured to calculate the matching degree of a predicted box and a real box by using an improved intersection over union IoU measurement method, and dynamically adjust the distribution of anchor sizes.

[0140] A scale-sensitive screening unit is configured to assign a specific scale anchor set to each feature layer according to the receptive field range of different feature layers in a target detection task.

[0141] The gradient back propagation adaptation unit incorporates the anchor point size parameter into the network training process and optimizes the distribution of the anchor point size through end-to-end learning.

[0142] The gradient back propagation adaptation unit automatically generates the anchor point size adapted to different detection scales by analyzing the geometric characteristics of the target data set, thereby improving the detection accuracy of the model for multi-scale targets.

[0143] The multi-scale feature fusion pyramid structure comprises:

[0144] The bottom-up feature extraction path is used to extract multi-scale features of the input image layer by layer through a convolutional neural network to generate feature maps of different resolutions.

[0145] The top-down feature fusion path is used to gradually fuse high-level semantic features with bottom-layer high-resolution features through upsampling operations.

[0146] The lateral connection module adjusts the number of feature channels using a 1x1 convolution kernel to realize cross-scale connection of feature maps of different levels.

[0147] The feature enhancement unit introduces an attention mechanism in the feature fusion process to dynamically adjust the weight distribution of different scale features.

[0148] The multi-level prediction output layer sets detection heads on each scale feature map after fusion to realize synchronous detection of small, medium, and large targets of different sizes.

[0149] The multi-scale feature fusion pyramid structure realizes complementary fusion of deep and shallow layer features through a bidirectional feature transmission path to improve the detection performance of the model for multi-scale targets.

[0150] The improved YOLOv7-SwinT model realizes efficient and accurate target recognition, and the adaptive path planning algorithm and intelligent task allocation algorithm are combined to realize multi-unmanned aerial vehicle cooperative inspection. The method has innovation in target recognition, path planning, task allocation and energy management, and is suitable for unmanned aerial vehicle inspection tasks in complex environment.

[0151] S2: Based on the target information identified in step S1, the current power of the unmanned aerial vehicle, the environmental conditions and the task priority are combined to generate a dynamic inspection path through an adaptive path planning algorithm.

[0152] Among them, the adaptive path planning algorithm can be to combine target information and environmental conditions: based on the identified target information, the current power of the unmanned aerial vehicle, the environmental conditions (wind speed, rainfall intensity, visibility) and the task priority are combined to generate a dynamic inspection path. And construct a multi-constraint path planning model, where the input parameters can be the current power state of the unmanned aerial vehicle, the environmental condition parameters (wind speed, rainfall intensity, visibility), the task priority matrix, and the geographic fence restriction conditions. The output can be the optimized inspection path, which is dynamically updated to the unmanned aerial vehicle flight control system.

[0153] Specifically, the S2: based on the target information identified in step S1, the current power of the unmanned aerial vehicle, the environmental conditions and the task priority are combined to generate a dynamic inspection path through an adaptive path planning algorithm, including:

[0154] S2.1: Based on the target information identified in step S1, an inspection target priority evaluation model is established, which comprehensively considers: target abnormality degree score, confidence score output based on deep learning model; target type weight, pre-set priority coefficient of different types of targets; target timeliness parameter, time sensitivity factor calculated according to target change rate. The evaluation model comprehensively considers the target abnormality degree score, the target type weight and the target timeliness parameter. The target abnormality degree score is based on the confidence score output by the deep learning model, the target type weight is the pre-set priority coefficient of different types of targets, and the target timeliness parameter is the time sensitivity factor calculated according to the target change rate.

[0155] S2.2: A multi-constraint path planning model is constructed, and the input parameters include:

[0156] The current power state and estimated endurance time of the unmanned aerial vehicle; environmental condition parameters, including wind speed, rainfall intensity, and visibility; task priority matrix, calculated from step S2.1; geographic fence restriction conditions. The input parameters include the current power state and estimated endurance time of the unmanned aerial vehicle, environmental condition parameters (wind speed, rainfall intensity, visibility), task priority matrix and geographic fence restriction conditions. These parameters jointly determine the feasibility and optimization direction of the inspection path.

[0157] S2.3: Solving the optimal path using improved genetic algorithm, specifically including:

[0158] a) Initialization of population: generating N feasible paths as initial solutions;

[0159] b) Fitness calculation: constructing a multi-objective fitness function based on path length, task coverage, and energy balance;

[0160] c) Selection operation: combining elite preservation strategy and roulette wheel selection;

[0161] d) Crossover and mutation: designing a directional mutation operator for inspection tasks;

[0162] e) Dynamic adjustment: real-time monitoring of environmental changes to trigger path re-planning.

[0163] S2.4: Output the optimized inspection path and dynamically update it to the UAV cruise system. The optimized inspection path is dynamically updated to the UAV flight control system to ensure that the UAV can execute the task in real time.

[0164] S3: According to the inspection path generated in step S2, the intelligent task allocation algorithm is used to allocate inspection tasks to multiple UAVs, and the task execution situation is monitored in real time.

[0165] Specifically, multi-UAV task allocation: according to the generated inspection path, the intelligent task allocation algorithm is used to allocate tasks to multiple UAVs, and the task execution situation is monitored in real time.

[0166] The choice of multi-UAV task allocation algorithm is crucial to ensure efficient task execution. The problems faced by task allocation algorithms include computational complexity, task coupling, problem size, time limit, and heterogeneity. To address these problems, a variety of algorithms can be selected, such as auction-based algorithms, ant colony optimization (ACO), particle swarm optimization (PSO) algorithms, etc.

[0167] In practical applications, task allocation usually involves the following steps: task submission and observation: Follower_UAVs submit observation information to Leader_UAV, Leader_UAV generates a task list, and conducts task auction according to task execution cost. Task allocation and path planning: according to the task allocation result, Follower_UAVs judge the distance to the target, if the distance is less than the set value, then directly launch the missile; otherwise, use model predictive control (MPC) algorithm to plan the attack path. Dynamic adjustment and re-planning: during task execution, if the UAV fails or the environmental conditions change, the system needs to adjust the task allocation and path planning in real time.

[0168] During task execution, the system needs to monitor the status of the UAVs and the progress of the tasks in real time. After each UAV completes the inspection, it uploads the inspection data to the central server to generate raw inspection data. Then, based on the raw inspection data, a data integration algorithm is used to process and integrate the inspection data of multiple UAVs, generating integrated inspection data. In addition, the system also generates inspection reports through a report generation method and ensures the accuracy and completeness of the report content through a report review algorithm.

[0169] The design of the intelligent task allocation algorithm needs to consider multiple factors, including the availability of UAVs, endurance, task type, geographic location, etc. Market mechanisms can be used to achieve dynamic allocation of tasks, and consensus mechanisms can be used to solve conflict problems. In addition, algorithms based on improved self-attention mechanisms and adaptive task planning can be used to achieve efficient task allocation in dynamic environments.

[0170] In multi-UAV cooperative operation, task allocation and path planning need to consider the cooperation and communication between UAVs. After completing task allocation, the UAVs will fly to the maintenance area, visit the assigned tasks in turn, and return to the nest autonomously after completing the tasks. Through information exchange and task synchronization between UAVs, multi-UAV cooperative operation is achieved.

[0171] S4: During the inspection process, the edge computing device processes sensor data in real time and transmits key data to the control center.

[0172] Specifically, edge computing and data transmission are used. The edge computing device is used to process sensor data in real time during the inspection process through the edge computing device and transmit key data to the control center.

[0173] In S4, the edge computing device is used for real-time processing of sensor data during the inspection process and transmits key data to the control center. This process has the following key features: 1. Real-time data processing: edge computing devices can preliminarily process sensor data locally, reducing data transmission time to the cloud and reducing latency. For example, in unmanned aerial vehicle inspection, edge computing can reduce the time of data transmission to the cloud, thereby improving the efficiency of data processing. 2. Data preprocessing and filtering: edge computing devices preprocess and filter sensor data, uploading only key information to the cloud, thereby reducing network bandwidth usage and improving data transmission efficiency. 3. Local decision and response: in some cases, edge computing devices can make quick decisions based on local data. Edge computing nodes can analyze device sensor data in real time, trigger local alarms immediately upon detecting anomalies, and quickly push abnormal information to the terminal devices of safety management personnel. 4. Data transmission to the control center: key data processed by the edge computing device will be transmitted to the control center for further analysis and processing. In the remote inspection of substations, sensor data is first preprocessed by terminal devices, and then the preliminarily processed data is transmitted to the data processing server in the substation for further processing and integration. 5. Improve system performance: through edge computing, the real-time performance and accuracy of the system can be significantly improved. In the intelligent operation and maintenance of the power system, the combination of edge computing and sensor technology enables the data collected by sensors to be preliminarily processed and analyzed locally, thereby reducing the requirement for data transmission to remote centers, significantly reducing latency, and enhancing the real-time performance of data processing.

[0174] The application of edge computing in the inspection process not only improves the efficiency of data processing, but also enhances the real-time performance and accuracy of the system, providing strong support for the efficient execution of inspection tasks.

[0175] S5: Based on the real-time status and task progress of the unmanned aerial vehicle, the intelligent energy management system optimizes the scheduling of the charging base station and the allocation of energy for the unmanned aerial vehicle.

[0176] Optimizing charging scheduling and energy allocation, specifically: based on the real-time status and task progress of the unmanned aerial vehicle, the intelligent energy management system optimizes the scheduling of the charging base station and the allocation of energy for the unmanned aerial vehicle.

[0177] In S5, the intelligent energy management system optimizes the scheduling of charging stations and the allocation of energy to drones by monitoring their status and task progress in real-time. This process involves several key technologies and methods, as follows: 1. Real-time monitoring and data analysis: The intelligent energy management system monitors the battery voltage, current, temperature, and other key parameters of the drone in real-time through sensors and flight computers such as Pixhawk Px4 and Raspberry Pi. These data provide accurate support for energy management, ensuring that the system can respond quickly according to the real-time status of the drone. 2. Dynamic energy allocation: The system dynamically allocates energy to different task components according to the task priority and current energy status of the drone. When the drone performs a patrol task, the system will prioritize the energy supply of key sensors and communication modules, while reducing the energy consumption of unnecessary devices. In addition, the system can further reduce the energy consumption of the drone by adjusting the flight height, speed, and other parameters. 3. Charging station scheduling: When the battery level of the drone is lower than the preset threshold (such as 20%), the system will automatically trigger the charging process. The drone locates the nearest charging station through GPS navigation and uses image processing technology to identify available charging units. Then, the drone autonomously lands at the charging station and continues to perform tasks after charging. This autonomous charging mechanism not only reduces the need for manual intervention, but also improves the continuity and efficiency of tasks. 4. Optimization algorithm and model: In order to achieve efficient energy management, the system uses a variety of optimization algorithms, including Ruled-based EMS, Optimization-based EMS, and Intelligent-based EMS. For example, the Optimization-based EMS can use dynamic programming, model predictive control, and other methods to optimize the flight path and charging strategy of the drone, thereby minimizing energy consumption and extending flight time. 5. Intelligent scheduling and prediction: The intelligent energy management system also combines artificial intelligence and big data technology to analyze and predict the task demand, flight environment, and energy status of the drone in real-time. In this way, the system can plan the flight path and charging strategy of the drone in advance to ensure sufficient energy support during critical task stages. 6. Energy recovery and reuse: In order to further improve energy utilization efficiency, the system also develops an intelligent energy recovery mechanism. For example, the remaining energy generated by the drone during flight can be used to power other devices or charge the battery through wireless charging technology, thereby reducing energy waste.

[0178] The intelligent energy management system realizes efficient management of the scheduling of charging stations and the allocation of energy to drones through real-time monitoring, dynamic allocation, optimization algorithms, and intelligent scheduling, thereby significantly improving the endurance and task execution efficiency of drones.

[0179] Embodiment Two

[0180] As Figure 3 shown, the present application provides a UAV cruising system based on deep learning artificial intelligence image recognition algorithm, applied to the UAV cruising method based on deep learning artificial intelligence image recognition algorithm as described in Embodiment One, comprising: an image acquisition module 11, a path planning module 12, a task allocation module 13, a data processing module 14, and an energy management module 15.

[0181] It can be understood that in the present embodiment, the image acquisition module 11 is used to collect real-time image data of the inspection area by the UAV equipped with an improved YOLOv7-SwinT target recognition model, which integrates the single-stage target detection architecture of YOLOv7 and the visual feature extraction network of Swin Transformer, and adopts multi-sensor data fusion technology.

[0182] It can be understood that in the present embodiment, the path planning module 12 is used to generate a dynamic inspection path through an adaptive path planning algorithm based on the target information identified in step S1, combined with the current power of the UAV, environmental conditions and task priority.

[0183] It can be understood that in the present embodiment, the task allocation module 13 is used to allocate inspection tasks to multiple UAVs through an intelligent task allocation algorithm according to the inspection path generated in step S2, and to monitor the task execution in real time.

[0184] It can be understood that in the present embodiment, the data processing module 14 is used to process sensor data in real time through an edge computing device during the inspection process, and to transmit key data to the control center.

[0185] It can be understood that in the present embodiment, the energy management module 15 is used to optimize the scheduling of charging stations and the allocation of UAV energy through an intelligent energy management system based on the real-time status of the UAV and the task progress; wherein the improved YOLOv7-SwinT target recognition model in step S1 adopts a three-level recognition architecture, including primary anomaly detection based on lightweight CNN, intermediate precise positioning combined with attention mechanism, and advanced target classification based on multi-sensor data fusion.

[0186] The present application has made breakthrough progress in detection accuracy, response speed, energy efficiency, etc. through innovative algorithm architecture and system design, and is particularly suitable for power inspection, border monitoring, disaster rescue and other application scenarios with strict requirements on real-time and reliability, and has significant technical advantages and application value.

[0187] Figure 4An electronic device is provided in an embodiment of the present application. As shown in Figure 4 The electronic device at least includes the following parts: a processor 101 and a memory 100, a communication interface 103, and a bus 102.

[0188] In an embodiment of the present application, the memory 100 is configured to store processor-executable instructions, and the processor 101 is configured to execute the instructions to implement the method of the first aspect.

[0189] In an embodiment of the present application, a computer-readable storage medium includes instructions, and the instructions instruct the device to perform the method of the first aspect. For example, the instructions instruct the device to perform the method shown in the flow steps in the Figure 1

[0190] The program that works in the electronic device related to an embodiment of the present application can be a program (a program that makes a computer function) that controls a central processing unit (CPU) and the like to realize the functions of the above-described embodiments related to one aspect of the present application. Then, the information processed by these devices is temporarily stored in a random access memory (RAM) when it is processed, and then stored in various ROMs such as a read-only memory (Flash ROM), a hard disk drive (HDD), and the like, and read out, corrected, and written by a CPU as needed.

[0191] Note that a part of the electronic device of the above-described embodiments can also be realized by a computer. In this case, a program for realizing the control function can be recorded in a computer-readable recording medium, and realized by reading the program recorded in the recording medium into a computer and executing it.

[0192] Note that the "computer" referred to here means a computer built into the electronic device, and a computer including hardware such as an OS and a peripheral device. In addition, the "computer-readable recording medium" means a removable medium such as a floppy disk, a magneto-optical disk, a ROM, a CD-ROM, and the like, a storage device such as a hard disk built into a computer.

[0193] Furthermore, the "computer-readable recording medium" can include a medium that dynamically stores a program for a short period of time, such as a communication line in the case of transmitting a program via a network such as the Internet or a communication line such as a telephone line, and a medium that stores a program for a fixed period of time, such as a volatile memory inside a computer that is a server or a client in this case. In addition, the above-described program can be a program for realizing a part of the above-described functions, and can also be a program that can realize the above-described functions by being combined with a program already recorded in a computer.​

[0194] Furthermore, the electronic device in the above-described embodiments can also be realized as an assembly (device group) of a plurality of devices. Each device constituting the device group can have each function or a part or all of each functional block of the electronic device of the above-described embodiments. As the device group, all of each function or each functional block of the electronic device can be possessed.

[0195] Those skilled in the art will recognize that the above-described embodiments are merely illustrative of the application and should not be considered restrictive of the application's scope, which is defined by the appended claims, and their equivalents.

Claims

1. A method for drone patrol based on deep learning artificial intelligence image recognition algorithm, characterized in that, The method includes: S1: Real-time image data of the inspection area is collected by a drone equipped with an improved YOLO v7-S win T target recognition model. The improved YOLO v7-S win T target recognition model integrates the single-stage target detection architecture of YOLO v7 and the visual feature extraction network of S win Transformer, and adopts multi-sensor data fusion technology. S2: Based on the target information identified in step S1, and combined with the drone's current battery level, environmental conditions, and task priority, a dynamic inspection path is generated through an adaptive path planning algorithm. S3: Based on the inspection path generated in step S2, assign inspection tasks to multiple drones using an intelligent task allocation algorithm, and monitor the task execution status in real time. S4: During the inspection process, sensor data is processed in real time through edge computing devices, and key data is transmitted to the control center. S5: Based on the real-time status and mission progress of the UAV, optimize the scheduling of charging base stations and the energy allocation of the UAV through the intelligent energy management system; wherein, the improved YOLO v7-S win T target recognition model in step S1 adopts a three-level recognition architecture, including primary anomaly detection based on lightweight CNN, intermediate precise localization combined with attention mechanism, and advanced target classification based on multi-sensor data fusion.

2. The drone cruise method based on a deep learning artificial intelligence image recognition algorithm according to claim 1, characterized in that, The improved YOLO v7-S win T target recognition model specifically includes: The YOLO v7 object detection framework includes: a backbone network constructed from depthwise separable convolutions, an adaptive anchor size selection module, and a multi-scale feature fusion pyramid structure. The S win Transformer feature extraction network includes a hierarchical window attention mechanism, a cross-window connection module, a local perception enhancement module, and a multi-sensor data fusion layer. It achieves feature-level fusion of visual images, infrared thermal imaging, and LiDAR point cloud data, feature weight allocation based on the attention mechanism, and spatiotemporally aligned multimodal data association.

3. The drone cruise method based on a deep learning artificial intelligence image recognition algorithm according to claim 2, characterized in that, The backbone network constructed from depthwise separable convolutions includes: A feature extraction network is used to extract multi-scale features from an input image, which includes at least one variable attention module to enhance the correlation between distant pixels and improve the feature extraction capability of small targets; A feature fusion network is used to integrate the multi-scale features output by the feature extraction network, wherein at least one deformable convolutional module is included to expand the receptive field of the convolutional kernel and reduce the number of model parameters; The detection head is used to detect targets based on the fused features and outputs the target's category, location, and confidence level. The loss calculation module is used to calculate the model's loss function during training, including: localization loss, which calculates the matching degree between the predicted bounding box and the ground truth bounding box based on EIo U; target confidence loss, which measures the accuracy of the model's prediction of whether the target exists; and category prediction loss, which optimizes the classification result using label smoothing cross-entropy loss. The system employs Mosaic data augmentation, transfer learning, and fine-tuning strategies during the training phase to improve the model's detection accuracy for small targets in drone patrol scenarios.

4. The drone cruise method based on a deep learning artificial intelligence image recognition algorithm according to claim 3, characterized in that, The loss function of YOL O v7 is a multi-task loss function, consisting of three parts: localization loss L CIoU Target confidence loss L obj And category prediction loss L cls , In YOL O v7, an improved IoU loss function is used to calculate the localization bounding box, and the regression error and localization loss L are obtained. CIoU The formula is as follows: IoU is the intersection-union ratio between the predicted bounding box and the ground truth bounding box, ρ 2 (b,b gt ) is the Euclidean distance between the center point of the predicted bounding box and the center point of the ground truth bounding box, l is the diagonal length of the minimum bounding rectangle, v is the aspect ratio consistency measure, and α is the weighting coefficient. Target confidence loss L obj The binary cross-entropy loss is used to measure whether the predicted bounding box contains the target, as shown in the following formula: S 2 B is the number of grid cells, and B is the number of anchor frames per grid cell. This is an indicator function; it is 1 if anchor frame j is responsible for detecting the target, and 0 otherwise. It is the confidence score of the prediction; Category Prediction Loss L cls To measure the accuracy of the predicted category, a binary cross-entropy loss is used, as shown in the following formula: in: These are real category labels. This represents the predicted class probability, where classes is the total number of classes and c represents the number of classes. The total loss function L of YOLOv7 total It is the weighted sum of the three loss functions mentioned above: L total =λ coord L CIoU +λ obj L obj +λ cls L cls , Where, λ coord , λ obj , λ cls These are hyperparameters used to balance the weights of different tasks; When the YOLOv7 framework includes an auxiliary head, its total loss function is L. total l : L total l =L total +0.25L aux , Among them, L aux The loss function for the auxiliary head.

5. The drone cruise method based on a deep learning artificial intelligence image recognition algorithm according to claim 2, characterized in that, The adaptive anchor point size selection module includes: K-means clustering units are used to perform cluster analysis on the width and height dimensions of all ground truth bounding boxes in the training dataset to generate an initial set of anchor point dimensions. The Intersection over Union (IoU) optimization unit uses an improved IoU metric to calculate the matching degree between the predicted bounding box and the ground truth bounding box, and dynamically adjusts the distribution of anchor point sizes. The scale-sensitive screening unit assigns a set of anchor points of a specific scale to each feature level based on the receptive field range of different feature layers in the target detection task. The gradient backpropagation adaptation unit incorporates anchor size parameters into the network training process, optimizing the distribution of anchor size through end-to-end learning. The gradient backpropagation adaptation unit automatically generates anchor point sizes that adapt to different detection scales by analyzing the geometric characteristics of the target dataset, thereby improving the model's detection accuracy for multi-scale targets.

6. The drone cruise method based on a deep learning artificial intelligence image recognition algorithm according to claim 2, characterized in that, The multi-scale feature fusion pyramid structure includes: The bottom-up feature extraction pathway is used to extract multi-scale features of the input image layer by layer through a convolutional neural network to generate feature maps of different resolutions. A top-down feature fusion pathway is used to fuse high-level semantic features with low-level high-resolution features step by step through upsampling operations. The horizontal connection module uses a 1×1 convolution kernel to adjust the number of feature channels, enabling cross-scale connections between feature maps at different levels; The feature enhancement unit introduces an attention mechanism during the feature fusion process to dynamically adjust the weight distribution of features at different scales; A multi-level prediction output layer sets up detection heads on the fused feature maps at various scales to achieve simultaneous detection of targets of different sizes, including small, medium, and large. The multi-scale feature fusion pyramid structure achieves complementary fusion of deep and shallow features through a bidirectional feature transfer path, thereby improving the model's detection performance for multi-scale targets.

7. The drone cruise method based on a deep learning artificial intelligence image recognition algorithm according to claim 1, characterized in that, Step S2: Based on the target information identified in step S1, and combined with the drone's current battery level, environmental conditions, and task priority, a dynamic inspection path is generated through an adaptive path planning algorithm, including: S2.1: Based on the target information identified in step S1, establish an inspection target priority evaluation model. The evaluation model comprehensively considers: target anomaly score, based on the confidence score output by the deep learning model; target type weight, a pre-set priority coefficient for different types of targets; and target timeliness parameter, a time sensitivity factor calculated based on the target change rate. S2.2: Construct a multi-constraint path planning model. Input parameters include: The drone's current battery status and estimated flight time; environmental condition parameters, including wind speed, rainfall intensity, and visibility; task priority matrix, calculated in step S2.1; geofencing constraints; S2.3: An improved genetic algorithm is used to solve for the optimal path, specifically including: a) Initialize the population: Generate N feasible paths as initial solutions; b) Fitness calculation: Construct a multi-objective fitness function based on path length, task coverage, and energy balance; c) Selection operation: A combination of elite retention strategy and roulette wheel selection is adopted; d) Crossover mutation: Design targeted mutation operators for inspection tasks; e) Dynamic adjustment: Real-time monitoring of environmental changes triggers path replanning; S2.4: Output the optimized inspection path and dynamically update it to the UAV patrol system.

8. The drone cruise method based on a deep learning artificial intelligence image recognition algorithm according to claim 1, characterized in that, The three-level recognition architecture includes primary anomaly detection based on lightweight CNN, intermediate precise localization combined with attention mechanism, and high-level object classification based on multi-sensor data fusion, including: The primary anomaly detection module uses a lightweight CNN network to quickly extract features from the input image and achieves preliminary screening of anomaly regions through convolutional layers with low computational complexity. The intermediate-level precise localization module, built on the S win Transformer architecture, includes: a window multi-head self-attention unit for capturing long-distance spatial dependencies; a cross-window feature interaction unit for global context modeling through a shifting window mechanism; and a dynamic receptive field adjustment unit for adaptively adjusting the attention range according to the target scale. The advanced target classification module integrates a multi-source sensor data fusion mechanism, specifically including: a visible light and infrared feature alignment unit that uses a cross-modal feature mapping network; a spatiotemporal context fusion unit that integrates temporal sensor data through 3D convolution; and a multi-task decision layer that synchronously outputs target category, confidence level, and spatial location information. A cascaded feature transfer path enables feature reuse and enhancement among three-level modules through residual connections and a feature pyramid structure. The architecture employs a progressive recognition strategy to achieve high-precision detection and classification of multi-scale targets in UAV inspection scenarios under conditions of limited computing resources.

9. A drone patrol system based on a deep learning artificial intelligence image recognition algorithm, applied to the drone patrol method based on a deep learning artificial intelligence image recognition algorithm as described in any one of claims 1 to 8, characterized in that, The system includes: The image acquisition module is used to collect real-time image data of the inspection area by a drone equipped with an improved YOLO v7-S win T target recognition model. The improved YOLO v7-S win T target recognition model integrates the single-stage target detection architecture of YOLO v7 and the visual feature extraction network of S win T ranceformer, and adopts multi-sensor data fusion technology. The path planning module is used to generate a dynamic inspection path based on the target information identified in step S1, combined with the drone's current battery level, environmental conditions, and task priority, through an adaptive path planning algorithm. The task allocation module is used to allocate inspection tasks to multiple drones based on the inspection path generated in step S2 using an intelligent task allocation algorithm, and to monitor the task execution status in real time. The data processing module is used to process sensor data in real time through edge computing devices during the inspection process and transmit key data to the control center. The energy management module is used to optimize the scheduling of charging base stations and the energy allocation of drones through an intelligent energy management system based on the real-time status and mission progress of the drones. The improved YOLO v7-S win T target recognition model in step S1 adopts a three-level recognition architecture, including primary anomaly detection based on lightweight CNN, intermediate precise localization combined with attention mechanism, and advanced target classification based on multi-sensor data fusion.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that instructs the device to perform the drone cruise method based on a deep learning artificial intelligence image recognition algorithm as described in any one of claims 1 to 8.

Images