Engineering supervision method, system and equipment based on unmanned aerial vehicle AI inspection and medium

Abstract

The invention discloses an engineering supervision method, system, equipment and medium based on unmanned aerial vehicle AI inspection, and relates to the technical field of intelligent inspection and engineering safety supervision, and the method comprises the steps: collecting and preprocessing multi-source visual information and thermal imaging data, and carrying out the preliminary analysis of the preprocessed data; constructing a feature recognition engine through scene skeleton modeling and a deep optimization algorithm, inputting an analysis result into the feature recognition engine, and outputting an abnormal structure and a potential safety hazard region by the feature recognition engine; and packaging the abnormal structure and the potential safety hazard area data into a packet, sending the packet, and adjusting the inspection route of the unmanned aerial vehicle according to the abnormal structure and the potential safety hazard area data. According to the method, a feature recognition engine is constructed through three-dimensional scene skeleton modeling and a deep optimization algorithm, the recognition precision of an abnormal structure of a construction site is improved, and efficient adjustment and automatic response of an inspection task are realized through data encapsulation and intelligent scheduling strategy optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent inspection and engineering safety supervision, in particular to an engineering supervision method and system based on AI inspection of unmanned aerial vehicles. BACKGROUND

[0002] The safety supervision of the construction site is of great significance in the management of the construction site, so it is very important to carry out necessary safety quality management on the construction of the building project. The effect of safety quality management work has affected the enterprise credit, economic benefit and even the survival of the construction enterprise. The construction of the building has its own particularity. In the current situation of the construction industry, the types and styles of the building have tended to diversification, and the construction conditions will be affected by the climate, and the cross use of various construction technologies and processes in the construction process also puts forward new requirements for the safety quality management level in the construction process. At present, the safety quality inspection of the construction site is still mostly carried out by manual close-range safety inspection, which has the problems of too many and heavy tasks for the inspection personnel, the need for a large amount of time and energy for multiple and long-time inspection, the limitation of the work by the infrastructure, the time and manpower consumption for the erection and disassembly of the scaffold, the influence on the work efficiency, and the importance of the stability to the personnel safety, low data precision and accuracy, etc. Therefore, a simpler and more effective method for safety detection of the construction site is always being sought on the project. These technologies on the project not only can monitor the construction site in real time, quickly find and handle safety hazards, but also can reduce the dependence on personnel and reduce the risk of high-altitude operation. At the same time, through the use of digital platform and intelligent equipment, the effective integration and analysis of data can be realized, and strong data support can be provided, so as to provide a scientific basis for the safety management of the construction site.

[0003] The unmanned aerial vehicle is equipped with a camera and a thermal imaging device, which can quickly and efficiently patrol the construction site. With the continuous progress of science and technology, the application of unmanned aerial vehicle technology in various fields is becoming more and more widespread. The application of unmanned aerial vehicles on the construction site is very wide, which can significantly improve the construction efficiency, safety and management level. The emergence of unmanned aerial vehicle + AI automatic inspection technology has brought revolutionary changes to the traditional inspection work. SUMMARY

[0004] In view of the above problems, the present application is proposed.

[0005] Therefore, the technical problem solved by the present application is that the existing engineering inspection and safety supervision method has the problems of low inspection efficiency, limited coverage, high labor cost, insufficient intelligentization of data collection and analysis, and how to use unmanned aerial vehicles combined with AI technology to realize automatic and high-precision construction site inspection and improve the accuracy of safety hazard identification and the efficiency of supervision.

[0006] To solve the above technical problems, the application provides the following technical solutions: an engineering supervision method based on unmanned aerial vehicle AI inspection, comprising collecting multi-source visual information and thermal imaging data and preprocessing, and preliminarily analyzing the preprocessed data.

[0007] The feature recognition engine is constructed through scene skeleton modeling and a depth optimization algorithm, the analysis result is input into the feature recognition engine, and the feature recognition engine outputs abnormal structures and safety hazard areas.

[0008] The abnormal structure and safety hazard area data are encapsulated into a package and sent, and the unmanned aerial vehicle inspection route is adjusted according to the abnormal structure and safety hazard area data.

[0009] The feature recognition engine is constructed, including extracting structural information of the construction site, constructing a three-dimensional space framework of the construction site through three-dimensional modeling, constructing key position points of the construction site scene skeleton according to actual engineering important positions, constructing a scene topology graph according to the spatial connection relationship between the skeleton points, forming a semantic skeleton structure, and constructing a scene skeleton model.

[0010] The output abnormal structure and safety hazard area include: based on the input image recognition result and historical data sample, a feature vector matrix is constructed, and the key structure features are extracted through a depth optimization algorithm and matched and compared with a preset abnormal feature template.

[0011] When a region with a similarity higher than a set threshold to an abnormal template is detected, it is automatically determined as an abnormal structure or a safety hazard area, and the structured data form containing position coordinates, risk type and severity label is output.

[0012] As a preferred scheme of the engineering supervision method based on unmanned aerial vehicle AI inspection, the multi-source visual information acquisition includes: using an unmanned aerial vehicle device with infrared thermal sensing and visible light camera dual channels to fly in a preset path in the construction area, simultaneously collecting infrared thermal imaging and visible light image data during the flight, and aligning and fusing the two types of images through a time stamp synchronization and space registration algorithm.

[0013] The preprocessing includes: performing Gaussian filter denoising, brightness normalization, edge enhancement and contour extraction processing on the collected image data.

[0014] As a preferred scheme of the engineering supervision method based on unmanned aerial vehicle AI inspection, the preliminary analysis includes: performing target region division and structure risk feature recognition processing on the preprocessed visible light image and thermal imaging image, labeling the risk area label according to the region division and recognition processing result, and arranging the labeled image and the corresponding region position index into an analysis result.

[0015] The target region division comprises using an image recognition algorithm to segment and label the building boundary, beam-column structure, operation equipment and personnel activity region in the image.

[0016] The structural risk feature recognition processing comprises detecting differences between images of the same position taken at different time periods through image sequence change analysis, and identifying construction progress abnormalities and equipment position drift conditions.

[0017] As a preferred scheme of the engineering supervision method based on the unmanned aerial vehicle AI inspection, the construction feature recognition engine comprises extracting structural parameters of main buildings through structural drawings of the construction site and field surveying data, using a three-dimensional modeling tool to perform point cloud modeling and structural rendering on the spatial layout of the construction site, and generating a three-dimensional scene model of the construction site with coordinate information and structural semantic labels.

[0018] The three-dimensional scene model of the construction site comprises spatial positioning of key construction nodes, equipment installation points and high-risk operation regions, labeling a plurality of skeleton key points, constructing a spatial topology structure comprising node connectivity, directionality and hierarchical relationship according to the physical connection relationship and construction logic relationship between the skeleton key points, and generating a semantic skeleton graph of the construction site.

[0019] The deep optimization algorithm comprises inputting the semantic skeleton structure into a deep feature recognition model, constructing a feature classifier based on template matching, introducing large model pre-training parameters and image samples in a specific scene for fine-tuning training, and generating an image discrimination network.

[0020] As a preferred scheme of the engineering supervision method based on the unmanned aerial vehicle AI inspection, the output abnormal structure and safety hazard region comprises establishing a spatial correspondence relationship according to the preliminary analysis result and the three-dimensional skeleton model, extracting a structural feature vector from the input image features through a deep convolutional neural network, constructing a feature matching model in combination with historical image samples and manually labeled abnormal templates, and outputting abnormal structure data.

[0021] The feature matching model comprises using an image similarity evaluation algorithm to calculate the matching degree between the current image structure region and the template, and determining the region as an abnormal structure region when the matching degree is higher than a threshold set by the system.

[0022] The two-dimensional image coordinates of the abnormal structure region and the semantic position of the three-dimensional skeleton model are mapped, and a spatial index number in the building scene is output, and according to the abnormal template matching result, a corresponding risk type code, risk level score and recommended disposal priority are automatically attached.

[0023] As a preferred scheme of the engineering supervision method based on unmanned aerial vehicle AI inspection provided in the application, the data packaging and sending comprises packaging the abnormal area image screenshot, the spatial position coordinates, the risk level label, the abnormal type description, the identification timestamp and the processing suggestion text through the JSON format, and sending the data to the engineering supervision platform, the dispatch center terminal and the project management personnel mobile terminal through network transmission.

[0024] As a preferred scheme of the engineering supervision method based on unmanned aerial vehicle AI inspection provided in the application, the adjustment of the unmanned aerial vehicle inspection route comprises analyzing the risk level and the position of the abnormal area according to the received data packet, calling a target function to calculate the comprehensive cost of each area, and generating new inspection path instructions according to the cost ranking.

[0025] Another object of the application is to provide an engineering supervision system based on unmanned aerial vehicle AI inspection, which can realize intelligent detection of structural abnormalities and safety hazards in the construction site by combining three-dimensional scene skeleton modeling and a feature recognition engine of a deep optimization algorithm, and solve the problems of current traditional engineering inspection technology, such as dependence on manual work, fragmented data collection and lagging risk identification.

[0026] As a preferred scheme of the engineering supervision system based on unmanned aerial vehicle AI inspection provided in the application, the system comprises a data collection and preprocessing analysis module, a feature recognition and abnormality discrimination module, and a data packaging and transmission and dynamic response module.

[0027] The data collection and preprocessing analysis module is used for collecting multi-source visual information and thermal imaging data and performing preprocessing, and performing preliminary analysis on the preprocessed data.

[0028] The feature recognition and abnormality discrimination module is used for constructing a feature recognition engine through scene skeleton modeling and a deep optimization algorithm, inputting the analysis result into the feature recognition engine, and outputting abnormal structures and safety hazard areas by the feature recognition engine.

[0029] The data packaging and transmission and dynamic response module is used for packaging and sending the abnormal structure and safety hazard area data, and adjusting the unmanned aerial vehicle inspection route according to the abnormal structure and safety hazard area data.

[0030] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the steps of the engineering supervision method based on unmanned aerial vehicle AI inspection.

[0031] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the steps of the engineering supervision method based on unmanned aerial vehicle AI inspection.

[0032] The application provides an engineering supervision method based on unmanned aerial vehicle AI inspection, which constructs a feature recognition engine through three-dimensional scene skeleton modeling and a depth optimization algorithm, improves the recognition accuracy of abnormal structures in a construction site, realizes efficient adjustment and automatic response of the inspection task through data encapsulation and intelligent scheduling strategy optimization, improves the intelligent level of engineering progress monitoring through multi-source data fusion and change detection, and effectively improves the inspection automation degree and risk early warning capability. The application achieves better effects in terms of inspection efficiency, recognition accuracy and engineering safety management intelligence. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0034] Figure 1 The overall flowchart of the engineering supervision method based on unmanned aerial vehicle AI inspection provided by the first embodiment of the application. DETAILED DESCRIPTION

[0035] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the specific embodiments of the application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are only a part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the application.

[0036] Embodiment 1, refer to Figure 1 For an embodiment of the application, an engineering supervision method based on unmanned aerial vehicle AI inspection is provided, which comprises:

[0037] S1: Collecting multi-source visual information and thermal imaging data and performing preprocessing, and performing preliminary analysis on the preprocessed data.

[0038] An unmanned aerial vehicle device with infrared thermal sensing and visible light camera dual channels is used to fly along a preset path in a construction area. Infrared thermal imaging and visible light image data are collected simultaneously during the flight. Time stamp synchronization and space registration algorithm is used to align and fuse the two types of images.

[0039] The collected image data is subjected to Gaussian filter denoising, brightness normalization, edge enhancement and contour extraction processing.

[0040] One preferred scheme of Gaussian filter denoising is:

[0041]

[0042] wherein G(x, y) represents the weight value of the Gaussian filter kernel at coordinates (x, y), σ represents the standard deviation, x represents the x-coordinate offset of the pixel point relative to the center of the filter kernel, and y represents the y-coordinate offset of the pixel point relative to the center of the filter kernel

[0043] A preferred scheme of brightness normalization is:

[0044]

[0045] wherein I(x, y) represents the pixel brightness value of the original image at coordinates (x, y), I min , I max represent the minimum and maximum brightness values of the image, I norm (x, y) represents the normalized brightness value.

[0046] A preferred scheme of edge enhancement is:

[0047]

[0048] wherein G x , G y represent the gradient values of the image in the horizontal and vertical directions, * represents the convolution operator, G represents the edge intensity, and I represents the input image matrix.

[0049] The preprocessed visible light image and the thermal imaging image are subjected to target region division and structural risk feature identification processing, and according to the results of the region division and identification processing, the risk region label is marked, and the marked image and the corresponding region position index are sorted into an analysis result.

[0050] A preferred scheme of target region division and structural risk feature identification processing is:

[0051]

[0052] ΔI t =||I t (x, y)-I t-1 (x, y)||2

[0053] wherein T threshold represents a segmentation threshold, Segment(x, y) represents a binary result, I t (x, y), I t-1 (x, y) represent the pixel values of the same position image at time points t and t-1, ||·||2 represents the Euclidean norm, and ΔI t represents the difference value of the two frames of images.

[0054] The target region division includes segmenting and labeling the structure boundary, beam column structure, operation equipment and personnel activity region in the image by using an image recognition algorithm.

[0055] The structural risk feature recognition processing includes detecting differences of images of the same position taken at different time periods by image sequence change analysis, and identifying construction progress abnormalities and equipment position drift conditions.

[0056] S2: Constructing a feature recognition engine through scene skeleton modeling and deep optimization algorithm, inputting the analysis result into the feature recognition engine, and outputting abnormal structures and safety hazard areas by the feature recognition engine.

[0057] The structural parameters of the main structures are extracted from the structural drawings and field surveying data of the construction site, point cloud modeling and structure rendering of the site space layout are performed by using a three-dimensional modeling tool, and a site three-dimensional scene model with coordinate information and structure semantic labels is generated.

[0058] One preferred scheme of the point cloud modeling is:

[0059]

[0060] Wherein, R represents a rotation matrix, t represents a translation vector, p i , q i represent the i-th corresponding point of the point cloud to be registered and the target point cloud, and N represents the number of point cloud pairs.

[0061] The site three-dimensional scene model includes spatial positioning of key construction nodes, equipment installation points and high-risk operation areas, calibration of multiple skeleton key points, construction of a spatial topology structure including node connectivity, directionality and hierarchical relationship according to the physical connection relationship and construction logic relationship between the skeleton key points, and generation of a semantic skeleton graph of the site.

[0062] The deep optimization algorithm includes inputting the semantic skeleton structure into a deep feature recognition model, constructing a feature classifier based on template matching, introducing large model pre-training parameters and image samples in specific scenes for fine-tuning training, and generating an image discrimination network.

[0063] It should be noted that the image discrimination network is trained by transfer learning, the training data is a labeled sample set composed of structure labels and image corresponding relationships, an image-structure alignment mechanism is used as auxiliary input, a cross-entropy loss function is used for supervised optimization, and finally the classification judgment ability of the abnormal area in the input image is obtained.

[0064] One preferred scheme of constructing a feature classifier based on template matching is:

[0065]

[0066] wherein W(i,j) represents the weight of the convolution kernel at the (i,j) position, b represents the bias term, k represents the size of the convolution kernel, and I(x+i,y+j) represents the pixel value of the input image within the local window.

[0067] According to the preliminary analysis result and the three-dimensional skeleton model, a spatial correspondence relationship is established, input image features are extracted through a deep convolutional neural network, a feature matching model is constructed by combining historical image samples and manually labeled abnormal templates, and abnormal structure data is output.

[0068] The feature matching model includes an image similarity evaluation algorithm, which calculates the matching degree between the current image structure region and the template, and when the matching degree is higher than the threshold set by the system, the region is determined as an abnormal structure region.

[0069] One preferred scheme of feature matching is:

[0070]

[0071] wherein V current , V template represent the feature vectors of the current image region and the preset template, · represents the vector dot product, and ||·|| represents the L2 norm of the vector.

[0072] The two-dimensional image coordinates of the abnormal structure region are mapped to the semantic position of the three-dimensional skeleton model, and the spatial index number in the building scene is output, and according to the abnormal template matching result, the corresponding risk type code, risk level score, and recommended disposal priority are automatically attached.

[0073] One preferred scheme of mapping the two-dimensional image coordinates to the semantic position of the three-dimensional skeleton model is:

[0074]

[0075] wherein (u,v) represents the two-dimensional image coordinates, (X,Y,Z) represents the three-dimensional skeleton model coordinates, k u , k v , b x , b y represent calibration parameters, Depth(u,v) represents depth information, and a represents a scale factor.

[0076] S3: encapsulate the abnormal structure and the security hazard region data into a package and send it, and adjust the unmanned aerial vehicle inspection route according to the abnormal structure and the security hazard region data.

[0077] The abnormal area image screenshot, spatial position coordinates, risk level label, abnormal type description, identification timestamp and processing suggestion text are encapsulated through the JSON format and transmitted to the engineering supervision platform, dispatch center terminal and project management personnel mobile terminal through network transmission.

[0078] According to the received data packet, the risk level and position of the abnormal area are analyzed, a target function is called to calculate the comprehensive cost of each area, the new inspection path instruction is generated according to the cost ranking.

[0079] The maximum value normalization processing is used to unify the dimension and numerical level.

[0080] A preferred scheme of calling the target function for calculation is as follows:

[0081]

[0082] Wherein, d t represents the length of the tth path, r t represents the risk level of the region corresponding to the tth path, w1 and w2 represent weight coefficients, d max represents the maximum value of the path, r max represents the maximum value of the risk level.

[0083] According to the optimized inspection priority list, the A algorithm is called to plan the path in turn, the planning result is converted into the flight control instruction of the unmanned aerial vehicle, and the environment change is continuously monitored during the flight to trigger the dynamic path re-planning.

[0084] A preferred scheme of the A algorithm for planning the path is as follows:

[0085] f(n)=g(n)+h(n)

[0086] Wherein, g(n) represents the actual cost from the starting point to the node n, h(n) represents the estimated cost from the node n to the end point, and f(n) represents the total cost of the node n.

[0087] The data and adjustment results are recorded and stored as subsequent optimization training data of the model.

[0088] Embodiment 2 is an embodiment of the present application, which provides an engineering supervision system based on unmanned aerial vehicle AI inspection, including a data acquisition and preprocessing analysis module, a feature recognition and abnormality discrimination module, a data encapsulation transmission and dynamic response module.

[0089] The data acquisition and preprocessing analysis module is used for acquiring multi-source visual information and thermal imaging data and performing preprocessing, and performing preliminary analysis on the preprocessed data.

[0090] It should also be noted that the data acquisition and preprocessing analysis module not only undertakes the image acquisition and enhancement task, but also provides standardized and high-credibility data input for the feature recognition and anomaly discrimination module to construct deep recognition. The output content includes multi-modal images, boundary segmentation mask, change analysis label and image space coordinate index, etc., which is the key pre-process link to realize high-precision recognition of downstream recognition engine.

[0091] The feature recognition and anomaly discrimination module is used to construct a feature recognition engine through scene skeleton modeling and deep optimization algorithm, and the analysis result is input into the feature recognition engine, and the feature recognition engine outputs abnormal structure and safety hazard area.

[0092] It should also be noted that the feature recognition and anomaly discrimination module not only has static image recognition capability in the implementation process, but also conducts time dynamic detection combined with image sequence difference information, comprehensively considers two dimensions of spatial structure and time change, realizes high-robustness recognition of abnormal structure through feature vector similarity and structure mapping, and standardizes the recognition result as structured output containing position coordinates, risk level and abnormal type, and transmits it to the data packaging transmission and dynamic response module for instruction decision.

[0093] The data packaging transmission and dynamic response module is used to package and send the abnormal structure and safety hazard area data, and adjust the unmanned aerial vehicle inspection route according to the abnormal structure and safety hazard area data.

[0094] It should also be noted that the data packaging transmission and dynamic response module not only undertakes the instruction execution and data communication responsibility, but also is the core relay of the "recognition-response-feedback" closed loop. The flight log and feedback information after each dynamic path planning will be automatically stored by the system as training samples for future model optimization, realizing intelligent upgrading of the whole chain from data acquisition, structure judgment, task scheduling and data accumulation.

Claims

1. An engineering supervision method based on unmanned aerial vehicle AI inspection, characterized in that, The method comprises the following steps: Collecting multi-source visual information and thermal imaging data and preprocessing, preliminary analysis of the preprocessed data; Through scene skeleton modeling and deep optimization algorithm to construct feature recognition engine, input the analysis result into the feature recognition engine, the feature recognition engine outputs abnormal structure and safety hidden danger area; Abnormal structure and safety hidden danger area data encapsulation and sending, adjusting the unmanned aerial vehicle inspection route according to the abnormal structure and safety hidden danger area data; The construction of feature recognition engine includes extracting the structural information of construction site, constructing the three-dimensional space framework of construction site through three-dimensional modeling, calibrating key position points according to the important position of actual engineering, constructing scene topology graph according to the space connection relationship between skeleton points, forming semantic skeleton structure, and constructing scene skeleton model; The output of abnormal structure and safety hidden danger area includes constructing feature vector matrix based on input image recognition result and historical data sample, extracting key structure features through deep optimization algorithm and matching comparison with preset abnormal feature template; When the similarity of the detected area to the abnormal template is higher than the set threshold, it is automatically determined as abnormal structure or safety hidden danger area, and the structured data form containing position coordinates, risk type and severity label is output. 2.The engineering supervision method based on UAV AI inspection of claim 1, wherein: The collection of multi-source visual information includes, The unmanned aerial vehicle equipment with infrared thermal sensing and visible light camera dual channel is used to fly in the preset path of the construction area, and the infrared thermal imaging and visible light image data are collected at the same time during the flight. The two kinds of images are aligned and fused through timestamp synchronization and space registration algorithm; The preprocessing includes Gaussian filter denoising, brightness normalization, edge enhancement and contour extraction processing of the collected image data. 3.The engineering supervision method based on UAV AI inspection of claim 2, wherein: The preliminary analysis includes, The target area division and structure risk feature recognition processing are carried out on the preprocessed visible light image and thermal imaging image, the risk area label is labeled according to the division and recognition processing result, and the labeled image and corresponding area position index are sorted into analysis result; The target area division includes using image recognition algorithm to segment and label the structure boundary, beam column structure, operation equipment and personnel activity area in the image; The structure risk feature recognition processing includes difference detection of images of the same position taken at different times through image sequence change analysis, identification of construction progress abnormality and equipment position drift. 4.The engineering supervision method based on UAV AI inspection of claim 3, wherein: The construction of feature recognition engine includes, Through the structural drawing and field surveying data of construction site, the structural parameters of main structures are extracted, the three-dimensional modeling tool is used for point cloud modeling and structure rendering of the space layout of construction site, and the three-dimensional scene model of construction site with coordinate information and structure semantic label is generated; The three-dimensional scene model of construction site includes spatial positioning of key construction nodes, equipment installation points and high-risk operation areas, calibration of multiple skeleton key points, construction of space topology structure including node connectivity, directionality and hierarchical relationship according to the physical connection relationship and construction logic relationship between skeleton key points, and generation of semantic skeleton graph of construction site. The deep optimization algorithm comprises inputting the semantic skeleton structure into a deep feature recognition model, constructing a feature classifier based on template matching, introducing large model pre-training parameters and fine-tuning training with image samples in a specific scene to generate an image discrimination network. 5.The engineering supervision method based on UAV AI inspection of claim 4, wherein: The output abnormal structure and safety hazard area comprises, According to the preliminary analysis result and the three-dimensional skeleton model, a spatial correspondence relationship is established, the input image features are extracted through a deep convolutional neural network to obtain a structure feature vector, a feature matching model is constructed by combining historical image samples and manually labeled abnormal templates, and abnormal structure data is output; The feature matching model comprises an image similarity evaluation algorithm, which is used to calculate the matching degree between the current image structure area and the template, and when the matching degree is higher than the threshold set by the system, the area is determined as an abnormal structure area; The two-dimensional image coordinates of the abnormal structure area are mapped to the semantic position of the three-dimensional skeleton model, and the spatial index number in the building scene is output, and according to the abnormal template matching result, the corresponding risk type code, risk level score and recommended treatment priority are automatically added. 6.The engineering supervision method based on UAV AI inspection of claim 5, wherein: The data packaging and sending comprises, The abnormal area image screenshot, spatial position coordinates, risk level label, abnormal type description, identification timestamp and processing suggestion text are packaged through the JSON format, and are sent to the engineering supervision platform, the dispatching center terminal and the project management personnel mobile terminal through network transmission. 7.The engineering supervision method based on UAV AI inspection of claim 6, wherein: The adjustment of the unmanned aerial vehicle inspection route comprises, According to the received data packet, the risk level and position of the abnormal area are analyzed, a target function is called to calculate the comprehensive cost of each area, and a new inspection path instruction is generated according to the cost sorting.

8. A system employing the engineering supervision method based on UAV AI inspection according to any one of claims 1-7, characterized in that: The method comprises a data acquisition and preprocessing analysis module, a feature recognition and abnormality discrimination module, and a data packaging and transmission and dynamic response module; The data acquisition and preprocessing analysis module is used for collecting multi-source visual information and thermal imaging data and performing preprocessing, and performing preliminary analysis on the preprocessed data; The feature recognition and abnormality discrimination module is used for constructing a feature recognition engine through scene skeleton modeling and deep optimization algorithm, inputting the analysis result into the feature recognition engine, and outputting the abnormal structure and safety hazard area by the feature recognition engine; The data packaging and transmission and dynamic response module is used for packaging and sending the abnormal structure and safety hazard area data, and adjusting the unmanned aerial vehicle inspection route according to the abnormal structure and safety hazard area data. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to realize the steps of the engineering supervision method based on unmanned aerial vehicle AI inspection according to any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the engineering supervision method based on unmanned aerial vehicle AI inspection according to any one of claims 1 to 7.

Images