Fossil identification method and device, electronic equipment and computer storage medium

Abstract

The present application relates to a kind of vestige identification method, device, electronic equipment and computer storage medium.The method thereof includes: obtaining the original point cloud data and image data in target area, and the original point cloud data is preprocessed, and the corrected point cloud data is obtained;Based on the corrected point cloud data extraction vestige feature information, and based on vestige feature information and image data constructs the DEM data and DOM data of vestige;Sample data set is constructed using DEM data and DOM data;Training set is input into pre-built vestige identification model and is trained until loss function tends to be stable, and after being verified based on verification set, the vestige identification model of training completion is obtained;Test set is input into the vestige identification model of training completion, and the vestige identification result of target area is output.The present application solves the problem of point cloud data disorder, no structure by constructing DEM data;Intelligent identification is carried out to the vestige of target area based on deep learning model, and the identification precision of vestige identification is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing archaeology, and in particular to a ruin identification method and device based on laser radar point cloud, electronic equipment and computer storage medium. BACKGROUND

[0002] In recent years, with the acceleration of urbanization, a large number of ancient city wall ruins have been irreversibly damaged. As a microcosm of history and culture and a witness to the changes of the times, the ancient city wall has high scientific, cultural and economic value. Therefore, in order to protect the ancient city wall ruins faster and better, it is of great significance to investigate and identify them.

[0003] The traditional method of manual field measurement uses total station, global navigation satellite system receiver and other measuring instruments to obtain the discrete points and lines on the surface of the ancient city wall, which is difficult to completely express the three-dimensional spatial information of the ancient city wall, and the work is tedious and inefficient. With the development of photogrammetry and remote sensing technology, images are collected by airborne cameras, unmanned aerial vehicles and other equipment, and then a real three-dimensional model is generated based on three-dimensional reconstruction method, making large-scale ancient city wall investigation and identification possible. However, the three-dimensional reconstruction process of this method requires a lot of computing power, and the three-dimensional model generated often has low precision and quality, which needs time-consuming indoor work such as hole filling and individualization, and is difficult to meet the urgent needs of ancient city wall investigation and protection.

[0004] In recent years, laser radar technology has developed rapidly, and with its high precision, high efficiency and non-contact advantages, it has become an important means of cultural heritage protection. In order to extract the information of ancient city wall ruins from three-dimensional point cloud, the extraction method of laser point cloud needs to be studied. The early extraction method of laser point cloud relies on manual design of features, mainly based on support vector machine, random forest and Markov random field algorithms to realize the extraction task. With the development of deep learning technology, convolutional neural networks have been widely used in computer vision tasks, and have gradually dominated the field of point cloud extraction. Currently, the commonly used methods include multi-view based, voxel based and direct processing of point cloud based methods, which have solved the problem of the application of deep learning in point cloud field caused by the unordered and unstructured nature of three-dimensional point cloud data. However, the current point cloud deep learning method has complex algorithm and low efficiency, which is difficult to adapt to the massive point cloud data today; and the labeling method of laser point cloud is complex and tedious, which leads to weak generalization ability and limited precision of the model. SUMMARY

[0005] Therefore, it is necessary to provide a ruin identification method and device based on laser radar point cloud, electronic equipment and computer storage medium, in order to solve the problem of limited precision of ancient city wall ruin identification.

[0006] To solve the above problems, in a first aspect, the present application provides a trace recognition method, comprising:

[0007] Obtaining original point cloud data and image data in a target area, and preprocessing the original point cloud data to obtain corrected point cloud data;

[0008] Extracting trace feature information based on the corrected point cloud data, and constructing DEM data and DOM data of the trace based on the trace feature information and the image data;

[0009] Constructing a sample data set using the DEM data and the DOM data, wherein the sample data set includes a training set, a validation set, and a test set;

[0010] Inputting the training set into a pre-built trace recognition model for training until the loss function tends to be stable, and obtaining a trained trace recognition model based on the validation set;

[0011] Inputting the test set into the trained trace recognition model, and outputting a trace recognition result of the target area.

[0012] Further, the preprocessing of the original point cloud data comprises:

[0013] Performing point cloud denoising, reflection intensity correction, and position deviation correction operations on the original point cloud data, wherein the processing formula of the reflection intensity correction comprises: I s is the corrected laser intensity, I is the original laser intensity, R and R s are the target distance and the standard distance, respectively, θ and θ s are the incident angle and the standard incident angle, respectively, and η atm is the atmospheric attenuation factor.

[0014] Further, the extraction of the trace feature information based on the corrected point cloud data, and the construction of the DEM data and the DOM data of the trace based on the trace feature information and the image data, comprise:

[0015] Extracting trace feature information from the corrected point cloud data, performing segmentation processing on the corrected point cloud data based on the trace feature information to obtain trace point cloud data;

[0016] Sampling the topographic feature information and the spatial position of the trace point cloud data, and constructing the DEM data of the trace using a preset point cloud interpolation algorithm;

[0017] Calculate the exterior orientation elements of the image data based on the method of aerial triangulation, and perform digital differential rectification on the image data based on the DEM data to obtain the DOM data of the vestige.

[0018] Further, the constructing the sample data set based on the DEM data and the DOM data comprises:

[0019] Taking the DEM data as a spatial reference, combining the DOM data and the vectorized drawing of the actual survey results to draw a vestige layer, and rasterizing the vectorized drawing of the vestige layer as the original sample data;

[0020] Perform augmentation and clipping processing on the original sample data to generate sample data, and divide the sample data into a training set, a validation set and a test set according to a preset distribution ratio.

[0021] Further, the pre-built vestige recognition model comprises a vestige recognition model constructed by using a multi-scale semantic feature extraction method;

[0022] The training of the training set in the pre-built vestige recognition model until the loss function tends to be stable comprises:

[0023] Performing transfer learning training on the training set to establish a feature quantitative relationship between the vestige and the DEM data until the loss function tends to be stable, wherein the loss function comprises a cross-entropy loss function based on a discrete classification problem.

[0024] Further, the inputting of the test set into the trained vestige recognition model to output a vestige recognition result of a target area comprises:

[0025] Based on the established feature quantitative relationship between the vestige and the DEM, an initial vestige recognition result of a target area is extracted in the test set by using a forward propagation method;

[0026] Based on the scale information of the vestige and a region connectivity algorithm, pixel elimination is performed on the initial vestige recognition result to output the vestige recognition result of the target area.

[0027] Further, after outputting the vestige recognition result of the target area, the method further comprises:

[0028] Verifying the vestige recognition result of the target area based on an evaluation index of the accuracy of the ancient city wall extraction.

[0029] In a second aspect, the present application also provides a vestige recognition device, comprising:

[0030] The data preprocessing module is configured to acquire original point cloud data and image data in a target area, and preprocess the original point cloud data to obtain corrected point cloud data.

[0031] The data construction module is configured to extract trace feature information based on the corrected point cloud data, and construct DEM data and DOM data of the trace based on the trace feature information and the image data.

[0032] The sample data set construction module is configured to construct a sample data set using the DEM data and the DOM data, wherein the sample data set includes a training set, a verification set and a test set.

[0033] The model training module is configured to input the training set into a pre-built trace recognition model for training until a loss function tends to be stable, and obtain a trained trace recognition model based on verification of the verification set.

[0034] The trace recognition module is configured to input the test set into the trained trace recognition model, and output a trace recognition result of the target area.

[0035] In a third aspect, the present application further provides an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps in the trace recognition method when executing the computer program.

[0036] In a fourth aspect, the present application further provides a computer storage medium, which stores a computer program, and the computer program is executable on a processor to implement the steps in the trace recognition method.

[0037] The beneficial effects of the above embodiments are as follows:

[0038] The present application uses airborne laser radar scanning technology and deep learning method to scan and photograph the ground surface of the target area, acquires three-dimensional laser point cloud data and two-dimensional image data of the target area, then preprocesses the three-dimensional laser point cloud data, combines the trace feature information in the preprocessed point cloud data with the two-dimensional image data to generate DEM data, solves the problem of unordered and unstructured laser point cloud, and overcomes the problems of limited laser radar point cloud data and weak target features; then based on the deep learning method, a sample library is constructed, a training model is built, and the trace spatial distribution information of the target area is recognized and extracted, which improves the accuracy of trace recognition, meets the current needs of trace investigation and protection, and has high quality and high efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 A flowchart of an embodiment of a trace recognition method provided by the present application is shown in the figure.

[0040] Figure 2 A multi-scale semantic feature extraction structure schematic diagram provided for an embodiment of the present application;

[0041] Figure 3 A trace recognition model training structure schematic diagram provided for an embodiment of the present application;

[0042] Figure 4 A structure schematic diagram of an embodiment of the trace recognition device provided by the present application;

[0043] Figure 5 A structure schematic diagram of an embodiment of the electronic device provided by the present application. DETAILED DESCRIPTION

[0044] The preferred embodiments of the present application will be described in detail below with reference to the accompanying drawings, which form a part of this application. The accompanying drawings, together with the description, illustrate the principles of the present application and are used to explain the principle of the present application, but are not used to limit the scope of the present application.

[0045] In the description of the present application, the terms "first", "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include at least one of the features. In addition, the meaning of "multiple" is two or more, unless otherwise specifically limited. In this document, the reference to "embodiments" means that the specific features, structures or properties described in conjunction with the embodiments can be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily mean the same embodiment, nor is it independent or alternative to other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0046] The present application provides a trace recognition method and device based on laser radar point cloud, electronic equipment and computer storage medium, which uses airborne laser radar scanning technology and deep learning method to identify and extract ancient city walls as the target, scans and shoots the ground surface of the ancient city walls, and obtains three-dimensional laser point cloud and two-dimensional image data. Then, after preprocessing, segmentation processing and other steps, DEM (Digital Elevation Model) data is generated, solving the problem of unordered and unstructured laser point cloud. Finally, based on the deep learning method, a sample library is constructed and a training model is built to identify and extract the spatial distribution information of the ancient city wall traces in the target area, and the identification result is verified and evaluated to meet the current needs of ancient city wall trace investigation and protection.

[0047] Before describing the embodiments, the related terms are explained:

[0048] DEM (Digital Elevation Model, digital elevation model) data is a digital simulation of a terrain surface or a digital representation of a terrain surface form through limited terrain elevation. The DEM data often uses absolute elevation or altitude, and the regular grid DEM has a simple data storage structure, good combination with remote sensing image structure, and better analysis of surface performance.

[0049] DOM (Digital Orthophoto Map, digital orthophoto map) data is image data generated by correcting each pixel and cutting according to the range of the map by using a digital elevation model to scan and process digital aerial photographs and remote sensing images. The information is more intuitive, and has good interpretability and measurability.

[0050] The following will be described in detail with respect to specific embodiments:

[0051] Please refer to Figure 1 , Figure 1 The flowchart of an embodiment of a trace recognition method provided by the present application, a specific embodiment of the present application, discloses a trace recognition method, comprising:

[0052] Step S101: obtaining original point cloud data and image data in a target area, and pre-processing the original point cloud data to obtain corrected point cloud data;

[0053] Step S102: extracting trace feature information based on the corrected point cloud data, and constructing DEM data and DOM data of the trace based on the trace feature information and the image data;

[0054] Step S103: constructing a sample data set using the DEM data and the DOM data, wherein the sample data set includes a training set, a validation set and a test set;

[0055] Step S104: inputting the training set into a pre-built trace recognition model for training until the loss function tends to be stable, and obtaining a trained trace recognition model based on the validation set;

[0056] Step S105: inputting the test set into the trained trace recognition model, and outputting a trace recognition result of the target area.

[0057] Before acquiring the original point cloud data and image data in the target area, the target area, i.e., the ancient city wall relic area, needs to be surveyed in the field, and parameters such as flight route, flight height, flight speed, etc. are set according to the range and terrain of the target area; the model of the airborne platform, the laser radar, the optical camera and other equipment is selected; and the parameters of the camera are calibrated.

[0058] After the preparation work such as field survey, flight plan making, system parameter determination and checking, the ancient city wall site is scanned based on the airborne laser radar system carrying devices such as laser scanner, global navigation satellite system, inertial navigation system and optical sensor, and the original laser point cloud and aerial image data of the target area are acquired.

[0059] Specifically, according to the hardware technical parameters and operation specifications, the target area, i.e., the ancient city wall relic area, is data collected to acquire original point cloud data, global navigation satellite data, inertial navigation data, optical image data, etc. Among them, the original point cloud data includes the angle, distance, intensity, echo, etc. of each point.

[0060] The present application utilizes airborne laser radar scanning technology and deep learning method to scan and shoot the ground surface of the target area, acquires three-dimensional laser point cloud data and two-dimensional image data of the target area, then pre-processes the three-dimensional laser point cloud data, combines the feature information of the relic in the pre-processed point cloud data with the two-dimensional image data to generate DEM data, solves the problem of unordered and unstructured laser point cloud, and overcomes the problems of limited laser radar point cloud data and weak target features; then based on the deep learning method, a sample library is constructed, a training model is built, and the spatial distribution information of the relics in the target area is recognized and extracted, which improves the accuracy of relic recognition and meets the current needs of relic investigation and protection with high quality and high efficiency.

[0061] In one embodiment of the present application, the pre-processing of the original point cloud data includes:

[0062] The original point cloud data is subjected to point cloud denoising, reflection intensity correction and position deviation correction operations, wherein the processing formula of the reflection intensity correction includes: I s is the corrected laser intensity, I is the original laser intensity, R and R s are the target distance and the standard distance respectively, θ and θ s are the incident angle and the standard incident angle respectively, and η atm is the atmospheric attenuation factor.

[0063] It can be understood that the preprocessing of the original laser point cloud data mainly includes point cloud denoising, reflection intensity correction, position deviation correction and the like. Among them, the point cloud denoising processing mainly removes noise data such as height abnormal points and isolated points; the reflection intensity correction processing mainly corrects the incidence angle, distance and atmospheric image factors; and the position deviation correction processing is mainly used to correct the errors generated by the global navigation satellite system and the inertial navigation system.

[0064] Specifically, the absolute coordinates of the laser points are calculated by using the collected global navigation satellite data, inertial navigation data and ranging data, and then the point cloud is processed by denoising, reflection intensity correction, position deviation correction and the like to eliminate system errors and improve the point cloud precision. Among them, the point cloud denoising can distinguish noise points and outliers by the distribution density near the point cloud or the number within the specified radius of the point cloud; the reflection intensity correction needs to be corrected from the incidence angle, distance and atmosphere.

[0065] The processing formula of the reflection intensity correction includes: I is the original laser intensity, R and R S are the target distance and the standard distance, θ and θs are the incidence angle and the standard incidence angle, and η atm is the atmospheric attenuation factor.

[0066] In an embodiment of the present application, the trace feature information is extracted based on the corrected point cloud data, and the DEM data and the DOM data of the trace are constructed based on the trace feature information and the image data, including:

[0067] The trace feature information is extracted in the corrected point cloud data, the corrected point cloud data is segmented based on the trace feature information, and the trace point cloud data is obtained;

[0068] The topographic feature information and the spatial position of the trace point cloud data are sampled, and the DEM data of the trace is constructed by using a preset point cloud interpolation algorithm;

[0069] The exterior orientation elements of the image data are calculated based on the aerial triangulation method, and the image data is digitally differentiated and corrected based on the DEM data, and the DOM data of the trace is obtained.

[0070] It can be understood that, regarding the generation of the DEM data and the DOM data of the trace, the preprocessed point cloud data, i.e. the corrected point cloud data, can be segmented based on the weak feature information of the ancient city wall to retain the ancient city wall trace point cloud, and then the DEM data and the DOM data of the trace are generated.

[0071] Firstly, the pre-processed laser point cloud data is segmented by selecting appropriate space information of the ancient city ruins as a characteristic factor, and the point cloud of the ruins surface terrain is obtained by eliminating the building, vegetation and other object point clouds on the ruins surface. Specifically, according to the point cloud density and the spatial distribution of the ancient city wall ruins, the point cloud is divided into appropriate grids. Considering the profile parameters such as the elevation distribution and the elevation difference of the grid point cloud, the characteristic factor of the point cloud grid is calculated: And the fuzzy clustering is performed with the characteristic factor value of the point cloud grid as the reference, and the labeling segmentation is performed according to the clustering result, so as to segment the ancient city wall ruins from the entire point cloud scene.

[0072] Wherein, GSD is the spatial resolution of the point cloud, which depends on the point cloud density, i and j are the row and column of the grid respectively, and k is the number of point clouds in the current grid; is the average value of the spatial distance between the point clouds in the grid and the center of the grid; and α, β are adjustment factors for the calculation of the characteristics of each grid; ij is the distance inverse power, is the elevation difference of the point clouds in the grid, and and are the maximum and minimum values of the elevations of the point clouds in the grid respectively; and max and min are the maximum and minimum values of the elevations of the entire point cloud scene.

[0073] Then, the ancient city wall ruins point cloud data segmented from the scene is sampled according to the terrain characteristic information and the spatial position of the ruins point cloud data, and a high-precision DEM is constructed by using a suitable point cloud interpolation algorithm and represented in the form of a regular grid. Specifically, the digital elevation model DEM can be fitted and constructed based on the Delaunay triangulation algorithm. When the elevation value of the target point is determined, the DEM is sequentially linearly interpolated according to the triangle in which the target point is located.

[0074] For example, assuming that the coordinates of the triangle vertices are (x1, y1, z1), (x2, y2, z2) and (x3, y3, z3), the calculation formula of the target point elevation is: Wherein, (x 21 ,y 21 ,z 21 ), (x 31 ,y 31 ,z 31 ) are the differences of the corresponding point cloud coordinates respectively.

[0075] Finally, the exterior orientation elements of the image data are recovered using the aerial triangulation method, the image data is digitally differential corrected based on the DEM elevation data, that is, the orthographic correction of the image data is performed by using the cubic convolution method, and finally the occluded area of the image data is detected and compensated, that is, the inlaying and color uniformity processing are performed, so as to obtain the DOM data of the ancient city wall relic area.

[0076] In an embodiment of the present application, the constructing the sample data set based on the DEM data and the DOM data comprises:

[0077] The DEM data is taken as a spatial reference, the DOM data and the actual survey result are combined to vectorize and draw a relic layer, and the vectorized and drawn relic layer is rasterized to serve as original sample data;

[0078] The original sample data is subjected to augmentation and clipping processing to generate sample data, and the sample data is divided into a training set, a verification set and a test set according to a preset allocation ratio.

[0079] It can be understood that, in the constructing the sample data set of the ancient city wall relics by using the DEM data and the DOM data, the spatial distribution of the ancient city wall relics is determined by combining expert knowledge and manual interpretation of the DOM data on the basis of field survey, and the data set is constructed through the steps of vectorization, rasterization, data enhancement, clipping and the like.

[0080] Firstly, the corresponding ancient city wall relic layer is vectorized and drawn by taking the DEM data as a spatial reference and through manual interpretation, to serve as sample data for deep learning. The rasterization is performed on the ancient city wall relic layer, so as to facilitate the reading of the deep learning model. Then, the typical features of the ancient city wall are fully considered, the areas with typical features are selected as training data in the sample, and the remaining part is taken as test data. The ratio of the training data to the test data is 7:3, 20% of the training data is taken as verification data to adjust the parameters of the model and preliminarily evaluate the ability of the model during the training process. Finally, the rotation, translation and the like are performed on the training data to increase the sample data amount. The augmented data is clipped to adapt to the scene with limited graphics card memory, considering the volume of the raster data. The construction of the training set, the verification set and the test set is completed.

[0081] In an embodiment of the present application, the pre-built relic identification model comprises a relic identification model constructed by using a multi-scale semantic feature extraction method;

[0082] The training of the training set in the pre-built relic identification model is performed until the loss function tends to be stable, comprising:

[0083] The training set is subjected to transfer learning training to establish a feature quantitative relationship of the vestige and DEM data until a loss function tends to be stable, wherein the loss function comprises a cross-entropy loss function based on a discrete classification problem.

[0084] It can be understood that the pre-built vestige recognition model comprises a vestige recognition model built by using a multi-scale semantic feature extraction method, and the model is suitable for extracting vestige feature information of ancient city walls. Figure 2 , Figure 2 A multi-scale semantic feature extraction structure diagram is provided for an embodiment of the present application. On this basis, a training model for large-scale DEM and weak feature information of ancient city wall vestiges is built, please refer to Figure 3 , Figure 3 A training structure diagram of a vestige recognition model is provided for an embodiment of the present application, specifically comprising an input layer, a convolution layer, a pooling layer and a full connection layer, then based on a cross-entropy loss function for a discrete classification problem, a training set is used for training a neural network model until the loss function value is stable. Then the validation set is used in the training process to adjust the parameters of the model, preliminarily evaluate the ability of the model, and finally obtain a trained vestige recognition model.

[0085] In an embodiment of the present application, the test set is input into the trained vestige recognition model, and a vestige recognition result of the target area is output, comprising:

[0086] Based on the established feature quantitative relationship of the vestige and DEM, an initial vestige recognition result of the target area is extracted in the test set by using a forward propagation method;

[0087] Based on the scale information of the vestige and the region connection algorithm, pixel elimination is performed on the initial vestige recognition result, and the vestige recognition result of the target area is output.

[0088] It can be understood that by using the trained model, an initial ancient city wall vestige is extracted in the test set, and then the initial ancient city wall vestige is vectorized and simplified to eliminate redundant and obviously incorrect data, and the vestige recognition result of the target area is obtained.

[0089] Specifically, by using the established feature quantitative relationship of the vestige and DEM, a forward propagation method of a deep learning model is used for reasoning to intelligently recognize the near-surface vestige in the region. Since the DEM and the point cloud absolute position correspond to each other, remote sensing interpretation and positioning of the ancient city wall are realized.

[0090] Finally, in view of the problems of "holes" and "noise points" in the extracted ancient city wall results, the scale information of the ancient city wall and the region connection algorithm are used to remove the small pixel patches through two iterations, realize the post-processing of the ancient city wall extraction results, and improve the interpretation and recognition accuracy.

[0091] In an embodiment of the present application, after outputting the trace recognition result of the target region, the above method further comprises:

[0092] The trace recognition result is verified based on the evaluation index of extraction accuracy.

[0093] It can be understood that the intelligent recognition result of the ancient city wall trace is verified by combining the interpretation of the archaeologist and the field investigation, the accuracy of the ancient city wall trace recognition and positioning is evaluated, and support is provided for the popularization and application of the model.

[0094] Specifically, the evaluation index of the ancient city wall extraction accuracy can adopt pixel accuracy rate PA and intersection over union IoU). Among them, the pixel accuracy rate represents the ratio of the number of correctly extracted pixels to the total number of pixels in the image, and the calculation formula is: The intersection over union represents the ratio of the intersection of the predicted graph set and the real labeled graph set to the union of the two sets, and the calculation formula is: In the formula, k represents the number of pixel categories; p ii represents the number of pixels whose actual category is i and whose predicted category is also i; t i represents the total number of pixels of category i; p ji represents the number of pixels whose actual category is i and whose predicted category is j.

[0095] The present application uses an airborne laser radar system to collect point cloud data, generates DEM data through preprocessing, segmentation and other processing steps, then extracts the ancient city wall trace of the target region based on a deep learning method, and finally verifies and evaluates the recognition and extraction result. The problems of weak ancient city wall trace features, scattered and discontinuous spatial distribution, mixed with terrain features, dense vegetation coverage and the like are solved. And considering the scale and distribution characteristics of the ancient city wall site, combining the knowledge in the field of archaeology, the effectiveness and reliability of the laser point cloud segmentation are improved, and then the accuracy of the DEM data describing the spatial information of the ancient city wall is improved.

[0096] In order to better implement the trace recognition method in the embodiments of the present application, on the basis of the trace recognition method, please refer to Figure 4 , Figure 4 An embodiment of the structure of the trace recognition device provided by the present application is shown in the figure. The embodiment of the present application provides a trace recognition device 400, which comprises:

[0097] The data preprocessing module 401 is configured to acquire original point cloud data and image data in a target area, and preprocess the original point cloud data to obtain corrected point cloud data.

[0098] The data construction module 402 is configured to extract trace feature information based on the corrected point cloud data, and construct DEM data and DOM data of the trace based on the trace feature information and the image data.

[0099] The sample data set construction module 403 is configured to construct a sample data set by using the DEM data and the DOM data, wherein the sample data set includes a training set, a verification set and a test set.

[0100] The model training module 404 is configured to input the training set into a pre-built trace recognition model for training until a loss function tends to be stable, and obtain a trained trace recognition model based on verification of the verification set.

[0101] The trace recognition module 405 is configured to input the test set into the trained trace recognition model, and output a trace recognition result of the target area.

[0102] It should be noted that the apparatus 400 provided in the above embodiments can implement the technical solutions described in the above method embodiments, and the principles of the implementation of the above modules or units can be referred to the corresponding content in the above method embodiments, which will not be described here.

[0103] Based on the above trace recognition method, the embodiments of the present application also correspondingly provide an electronic device, which includes a processor and a memory, and a computer program stored in the memory and executable on the processor; the processor implements the steps in the trace recognition method of the above embodiments when executing the computer program.

[0104] Figure 5 FIG. 5 shows a structural schematic diagram of an electronic device 500 suitable for implementing the embodiments of the present application. The electronic device in the embodiments of the present application can include but is not limited to a mobile terminal such as a mobile phone, a notebook computer, a digital broadcast receiver, a PDA (Personal Digital Assistant), a PAD (Tablet Personal Computer), a PMP (Portable Multimedia Player), a vehicle terminal (such as a vehicle navigation terminal), and the like, and a fixed terminal such as a digital TV, a desktop computer, and the like. Figure 5 The electronic device shown is only an example, and should not bring any limitation to the functions and use range of the embodiments of the present application.

[0105] The electronic device includes a memory and a processor, where the processor can be referred to as the processing device 501 below, and the memory can include at least one of a read-only memory (ROM) 502, a random access memory (RAM) 503, and a storage device 508, as shown below:

[0106] As shown in Figure 5 The electronic device 500 can include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 501 that can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 502 or loaded from a storage device 508 into a random access memory (RAM) 503. Various programs and data required for operation of the electronic device 500 are also stored in the RAM 503. The processing device 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0107] In general, the following devices can be connected to the I / O interface 505: input devices 506 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 508 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 509. The communication devices 509 can allow the electronic device 500 to communicate wirelessly or via a wire with other devices to exchange data. Although Figure 5 The electronic device 500 is shown with various devices, but it should be understood that all of the shown devices are not required to be implemented or present. More or fewer devices can alternatively be implemented or present.

[0108] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to embodiments of the present application. For example, embodiments of the present application include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication devices 509, or installed from the storage devices 508, or installed from the ROM 402. When the computer program is executed by the processing device 501, the above-mentioned functions defined in the methods of embodiments of the present application are performed.

[0109] Based on the above-described trace identification method, embodiments of the present application also correspondingly provide a computer-readable storage medium storing one or more programs, which can be executed by one or more processors to implement the steps in the trace identification method of the above-described embodiments.

[0110] Those skilled in the art can understand that all or part of the processes of the above-mentioned embodiment methods can be completed by instructing the relevant hardware by a computer program, and the program can be stored in a computer readable storage medium. The computer readable storage medium is a disk, an optical disk, a read-only memory, a random access memory, etc.

[0111] The above merely provides the preferred but not limiting embodiments of the present application, and any changes or replacements within the technical scope of the present application should be covered within the protection scope of the present application.

Claims

1. A vestige recognition method characterized by comprising: The method comprises the following steps: acquiring original point cloud data and image data in a target area, and pre-processing the original point cloud data to obtain corrected point cloud data; extracting trace feature information based on the corrected point cloud data, and constructing DEM data and DOM data of the trace based on the trace feature information and the image data, comprising: extracting trace feature information in the corrected point cloud data, performing segmentation processing on the corrected point cloud data based on the trace feature information to obtain trace point cloud data; sampling the topographic feature information and spatial position of the trace point cloud data, and constructing DEM data of the trace by using a preset point cloud interpolation algorithm; calculating the exterior orientation elements of the image data based on the aerial triangulation method, and performing digital differential rectification on the image data based on the DEM data to obtain DOM data of the trace; constructing a sample data set by using the DEM data and the DOM data, comprising: taking the DEM data as a spatial reference, combining the DOM data and the actual survey result to vectorize and draw a trace layer, and rasterizing the vectorized and drawn trace layer as original sample data; performing augmentation and clipping processing on the original sample data to generate sample data, and dividing the sample data into a training set, a validation set and a test set according to a preset distribution ratio, wherein the sample data set comprises the training set, the validation set and the test set; inputting the training set into a pre-built trace recognition model for training until the loss function tends to be stable, and obtaining a trained trace recognition model based on the validation set; inputting the test set into the trained trace recognition model to output a trace recognition result of the target area.

2. The vestige recognition method according to claim 1, characterized by, The pre-processing of the original point cloud data comprises: The original point cloud data is subjected to point cloud denoising, reflection intensity correction and position deviation correction operations, wherein the processing formula of the reflection intensity correction comprises: , is the corrected laser intensity, is the original laser intensity, and are the target distance and the standard distance respectively, and are the incident angle and the standard incident angle respectively, is the atmospheric attenuation factor.

3. The vestige recognition method according to claim 1, characterized by, The pre-built trace recognition model comprises a trace recognition model constructed by using a multi-scale semantic feature extraction method; The training of the training set in the pre-built trace recognition model until the loss function tends to be stable comprises: performing transfer learning training on the training set to establish a feature quantitative relationship between the trace and the DEM data until the loss function tends to be stable, wherein the loss function comprises a cross-entropy loss function based on a discrete classification problem.

4. The vestige recognition method according to claim 1, characterized by, The inputting of the test set into the trained trace recognition model to output the trace recognition result of the target area comprises: extracting an initial trace recognition result of the target area in the test set by using a forward propagation method based on the established feature quantitative relationship between the trace and the DEM; performing pixel elimination on the initial trace recognition result based on the scale information of the trace and a region connection algorithm to output the trace recognition result of the target area.

5. The vestige recognition method according to claim 1, characterized by, After outputting the trace recognition result of the target area, the method further comprises: verifying the trace recognition result of the target area based on an evaluation index of extraction accuracy.

6. A vestige recognition apparatus characterized by comprising: The method comprises the following steps: a data pre-processing module for acquiring original point cloud data and image data in a target area, and pre-processing the original point cloud data to obtain corrected point cloud data; The data construction module is configured to extract trace feature information based on the corrected point cloud data, and construct DEM data and DOM data of the trace based on the trace feature information and the image data. The data construction module includes: extracting trace feature information from the corrected point cloud data, performing segmentation processing on the corrected point cloud data based on the trace feature information to obtain trace point cloud data; sampling topographic feature information and spatial positions of the trace point cloud data, and constructing DEM data of the trace by using a preset point cloud interpolation algorithm; calculating exterior orientation elements of the image data based on an aerial triangulation method, and performing digital differential rectification on the image data based on the DEM data to obtain DOM data of the trace. The sample data set construction module is configured to construct a sample data set by using the DEM data and the DOM data. The sample data set construction module includes: taking the DEM data as a spatial reference, combining the DOM data and an actual survey result to vectorize and draw a trace layer, rasterizing the vectorized and drawn trace layer as original sample data, performing augmentation and clipping processing on the original sample data to generate sample data, and dividing the sample data into a training set, a validation set and a test set according to a preset allocation ratio. The sample data set includes the training set, the validation set and the test set. The model training module is configured to input the training set into a pre-built trace recognition model to perform training until a loss function tends to be stable, and obtain a trained trace recognition model based on the validation set. The trace recognition module is configured to input the test set into the trained trace recognition model to output a trace recognition result of a target area.

7. An electronic device, comprising: A device includes a memory and a processor. The memory is configured to store a program. The processor is coupled to the memory and is configured to execute the program stored in the memory to implement steps of the trace recognition method in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, A computer readable program or instruction is stored. When the program or instruction is executed by a processor, steps of the trace recognition method in any one of claims 1 to 5 can be implemented.

Images