A method and device for predicting the scope of a target mine area under high vegetation coverage, and an electronic device

Abstract

The present application provides a high-vegetation-covered area target mining area range prediction method, device and electronic equipment, comprising: acquiring hyperspectral data and laser radar point cloud data of a high-vegetation-covered research area, and registering; performing point cloud data layering on the point cloud data; manufacturing each layer of point cloud data into a height interval mask; performing spatial layering on the hyperspectral data based on the height interval mask, and extracting a hyperspectral image of a vegetation canopy bottom; extracting stressed vegetation according to abnormal spectral information, and manufacturing a sample; dividing all samples into a training set and a verification set according to a proportion; constructing a high-vegetation-covered area target mining area range prediction model based on deep learning DeeplabV3+; performing training and verification of the prediction model through the training set and the verification set respectively; acquiring a hyperspectral image of a target high-vegetation-covered area, and predicting the target mining area range in the target high-vegetation-covered area through the trained prediction model. The present application can efficiently and accurately realize high-vegetation-covered target mining area range prediction.

Description

Technical Field

[0001] This invention relates to the field of target mining area range prediction technology, and in particular to a method, device and electronic equipment for predicting the range of target mining areas under high vegetation cover. Background Technology

[0002] Predicting the extent of target mining areas is an indispensable part of mineral resource exploration. For areas with low to medium vegetation cover, passive hyperspectral remote sensing has achieved significant results by inverting anomalous spectra of mineralization and alteration information. However, extracting alteration information from areas with high vegetation cover is difficult, and the actual extraction results are unsatisfactory. Previous research has failed to develop a suitable method for inverting rock and mineral alteration information in areas with high vegetation cover due to limitations in data acquisition methods and intelligent processing algorithms. Most mineral deposits are located in areas with high vegetation cover, which significantly interferes with mineral resource exploration using remote sensing. Therefore, the composition and distribution of minerals can be indirectly determined by acquiring the anomalous reflectance spectra of stressed vegetation caused by mineralization and alteration. However, any single remote sensing technology has certain limitations in remote sensing mineral exploration research.

[0003] Existing related technologies and patents:

[0004] Chinese patent "A Remote Sensing Identification Method for Uranium Mineralization Alteration Information in Vegetated Areas" (application number CN202111660772.4) provides a method for remote sensing identification of volcanic rock-type uranium mineralization information in vegetated areas. Before extracting alteration information, this patent uses a pixel-based binary model to estimate vegetation cover in remote sensing images, dividing the images into high-vegetation-coverage areas and medium-to-low-vegetation-coverage areas. After removing interference from vegetation, vegetation in medium-to-low-vegetation-coverage areas is suppressed, and vegetation in high-vegetation-coverage areas is masked, resulting in an image of the medium-to-low-vegetation-coverage areas after removing interference factors. Principal component analysis is used to extract uranium mineralization alteration anomaly information, completing the remote sensing identification of uranium mineralization alteration information in vegetated areas. This method solves the technical problem of mitigating the impact of vegetation interference on the extraction of remote sensing alteration information.

[0005] Chinese patent "A Hyperspectral Remote Sensing Method for Oil and Gas Exploration in Sparsely Vegetated Areas" (application number CN201010560792.X) provides a method for hyperspectral remote sensing oil and gas exploration in sparsely vegetated areas. This patent performs K-means clustering-based land cover classification on the original hyperspectral image before extracting alteration information. The classification results are divided into bare surface areas and vegetated surface areas. In bare surface areas, oil and gas information based on alteration minerals is extracted from the hyperspectral image; in vegetated areas, oil and gas information based on vegetation spectral anomalies is extracted. The classification results obtained from the previous operations are then combined to delineate the distribution area of ​​oil and gas reservoirs. This patent utilizes surface alteration mineral anomalies to conduct hyperspectral oil and gas information exploration in bare surface areas and utilizes vegetation anomaly information to extract oil and gas information in vegetated areas, thereby delineating the oil and gas anomaly areas in the region based on a comprehensive analysis of both methods.

[0006] Existing studies only focus on the surface reflectance spectral information of vegetation canopy, without considering the precise extraction of anomalous reflectance spectral information from the bottom of the canopy. For stressed vegetation, the spectral anomalies at the bottom of the canopy are significantly stronger than those in the middle and top, exhibiting distinct differential stratification characteristics. Furthermore, existing studies only target mining areas in areas with low to medium sparse vegetation cover, failing to consider the challenges of high vegetation cover and the low efficiency of manual detection.

[0007] Therefore, there is an urgent need to provide an efficient and accurate method for predicting the extent of target mining areas under high vegetation cover. Summary of the Invention

[0008] The main technical problem to be solved by this invention is to achieve efficient and accurate identification of the target mining area under high vegetation cover.

[0009] To address the aforementioned technical problems, the present invention adopts the following technical solution: Introducing lidar point cloud data to further improve the accuracy of vegetation anomaly spectral information extraction. Lidar pulses can penetrate canopy gaps, possessing the advantage of detecting vertical spatial characteristics of the canopy, and have been effectively applied in the field of vegetation detection. However, relying solely on the vertical structure of the canopy makes it difficult to detect anomalous spectral distributions. Therefore, this invention leverages the rich spectral information of hyperspectral remote sensing and the rich spatial information of lidar, fusing these two different dimensions of data to maximize their respective characteristics and achieve higher-precision vegetation information inversion.

[0010] The introduction of deep learning methods further improves the automation of etch information extraction. The application of high-precision sample training models enables them to extract highly representative features even with a small number of samples. Furthermore, it enhances the model's ability to perform deep semantic mining. Under certain conditions, it addresses the drawback of relying on large datasets for feature learning in etch information extraction, achieving both automated extraction and high recognition accuracy.

[0011] According to a first aspect of the present invention, a method for predicting the extent of a target mining area in a high vegetation cover zone includes the following steps:

[0012] Acquire hyperspectral data and lidar point cloud data of the high vegetation cover study area and perform registration;

[0013] The acquired lidar point cloud data is layered.

[0014] Each layer of point cloud data is made into a mask for each height range;

[0015] The hyperspectral data is spatially layered based on masks for each height range, and hyperspectral images of the bottom of the vegetation canopy are extracted.

[0016] Based on anomalous spectral information, stressed vegetation was extracted from hyperspectral images of the bottom of the vegetation canopy and prepared as samples;

[0017] All samples are divided into training and validation sets according to a preset ratio;

[0018] A prediction model for the target mining area in a high vegetation cover area was constructed based on DeeplabV3+ deep learning.

[0019] The prediction model for the target mining area in high vegetation cover area is trained and validated using training and validation sets respectively until a well-trained prediction model is obtained.

[0020] Acquire hyperspectral images of the target high-vegetation-coverage area, and use a trained prediction model to predict the range of the target mining area within the target high-vegetation-coverage area.

[0021] Further, the steps of acquiring hyperspectral data and lidar point cloud data of the high vegetation cover study area and performing registration include:

[0022] Acquire hyperspectral data and lidar point cloud data of the high vegetation cover study area;

[0023] High-resolution DEM data obtained by TIN interpolation of ground point cloud data were used to perform orthorectification on hyperspectral data of high vegetation cover study area to eliminate image distortion caused by topographic undulation and other factors.

[0024] The lidar point cloud data was processed into a canopy height model and resampled using the nearest neighbor method to achieve the same spatial resolution as the hyperspectral data of the high vegetation cover study area.

[0025] Typical ground features were selected as control points from both hyperspectral data and lidar point cloud data of the high vegetation cover study area, and the two images were registered.

[0026] Furthermore, within the study area, a hyperspectral-lidar integrated UAV platform was used to simultaneously collect hyperspectral data and lidar point cloud data from the high-vegetation-coverage study area.

[0027] Furthermore, the step of performing point cloud data layering on the acquired lidar point cloud data includes:

[0028] Using Cloud Compare software, the lidar point cloud data is layered using intervals of 1 meter.

[0029] Furthermore, the steps of extracting stressed vegetation from hyperspectral images at the base of the vegetation canopy based on anomalous spectral information and preparing samples include:

[0030] Based on the abnormal spectral information, the stressed vegetation was extracted from the hyperspectral image of the bottom of the vegetation canopy. The extracted hyperspectral image of the bottom of the stressed vegetation canopy was cropped into an image of size 512×512 pixels and made into a sample.

[0031] Furthermore, the step of dividing all samples into training and validation sets according to a preset ratio includes:

[0032] All samples were divided into training and validation sets in an 8:2 ratio.

[0033] Furthermore, the step of constructing a target mining area range prediction model based on DeeplabV3+ deep learning includes:

[0034] A target mining area range prediction model was constructed using DeeplabV3+ as the base model and MobileNetV2 as the backbone network in a two-branch structure.

[0035] According to a second aspect of the present invention, the present invention provides a device for predicting the range of a target mining area in a high vegetation cover area for implementing the method, comprising the following modules:

[0036] The acquisition and registration module is used to acquire hyperspectral data and lidar point cloud data of the high vegetation cover study area and perform registration.

[0037] The point cloud layering module is used to layer the acquired lidar point cloud data.

[0038] The interval mask module is used to create interval masks for each height of each layer of point cloud data;

[0039] The canopy bottom image extraction module is used to spatially layer hyperspectral data based on masks of various height intervals and extract hyperspectral images of the vegetation canopy bottom.

[0040] The sample preparation module is used to extract stressed vegetation from hyperspectral images of the bottom of the vegetation canopy based on anomalous spectral information and prepare samples.

[0041] The sample partitioning module is used to divide all samples into training and validation sets according to a preset ratio.

[0042] The model building module is used to build a prediction model of the target mining area range in high vegetation cover areas based on DeeplabV3+ deep learning;

[0043] The model training module is used to train and validate the prediction model of the target mining area in the high vegetation cover area using the training set and the validation set, respectively, until a well-trained prediction model is obtained.

[0044] The target mining area prediction module is used to acquire hyperspectral images of the target high vegetation cover area and predict the range of the target mining area within the target high vegetation cover area using a trained prediction model.

[0045] According to a third aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method for predicting the range of target mining areas in high vegetation cover areas.

[0046] According to a fourth aspect of the present invention, a storage medium is provided thereon storing a computer program that, when executed by a processor, implements the steps of the method for predicting the range of target mining areas in high vegetation cover areas.

[0047] The technical solution provided by this invention has the following beneficial effects:

[0048] Compared with existing technologies, the technical solution proposed in this invention focuses on the significant difference in spectral information anomalies observed at the bottom of the canopy of stressed vegetation compared to the middle and top sections, highlighting a distinct stratification characteristic. Furthermore, it leverages the rich spectral information from hyperspectral remote sensing and the abundant spatial information from lidar, fusing these two datasets to achieve a vertically stratified distribution of vegetation canopy spectral information in a three-dimensional structure, eliminating interference from the top of the canopy and redundant mixed spectra. The filtered anomalous vegetation canopy sections are then used to create high-precision samples. These high-precision samples are used to train the DeeplabV3+ semantic segmentation model, enabling the model to extract more accurate and discriminative features, thus improving the ability to mine deep feature semantics. This addresses the problem of traditional classification methods lacking accurate samples and self-learning capabilities, which are unable to handle the extraction of stress information from mineral alteration vegetation in high-vegetation areas. A method for extracting indicative information of stressed vegetation in high-vegetation areas is established, improving the efficiency of target mining area location. Attached Figure Description

[0049] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0050] Figure 1 This is a flowchart illustrating the overall process of a method for predicting the range of a target mining area in a high vegetation cover region according to the present invention.

[0051] Figure 2 This is a flowchart of the hyperspectral image extraction process for the bottom of the vegetation canopy in this invention;

[0052] Figure 3 This is a schematic diagram of the structure of the target mining area range prediction model built by DeeplabV3+ based on deep learning in this invention;

[0053] Figure 4 This is a schematic diagram of the structure of a target mining area range prediction device in a high vegetation cover area according to the present invention.

[0054] Figure 5 This is a schematic diagram of the structure of an electronic device according to the present invention. Detailed Implementation

[0055] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0056] refer to Figure 1 , Figure 1 This is a flowchart illustrating the overall process of a method for predicting the range of a target mining area in a high vegetation cover region according to the present invention. The method specifically includes the following steps:

[0057] S1: Acquire hyperspectral data and lidar point cloud data of the high vegetation cover study area and perform registration;

[0058] S2: Perform point cloud data layering on the acquired lidar point cloud data;

[0059] S3: Create masks for each height range from the point cloud data of each layer;

[0060] S4: Spatial layering of hyperspectral data based on masks for each height interval, and extraction of hyperspectral images of the bottom of the vegetation canopy;

[0061] S5: Extract stressed vegetation from hyperspectral images at the bottom of the vegetation canopy based on anomalous spectral information and prepare samples;

[0062] S6: Divide all samples into training set and validation set according to a preset ratio;

[0063] S7: Constructing a target mining area range prediction model based on DeeplabV3+ deep learning;

[0064] S8: Train and validate the prediction model for the target mining area in the high vegetation cover area using the training set and validation set respectively, until a well-trained prediction model is obtained.

[0065] S9: Acquire hyperspectral images of the target high vegetation cover area and predict the target mining area range within the target high vegetation cover area using a trained prediction model.

[0066] Based on, but not limited to, the specific implementation process of step S1 is as follows: Figure 2 As shown:

[0067] S1.1: Within the study area, a hyperspectral-lidar integrated UAV platform is used to simultaneously acquire hyperspectral data and lidar point cloud data of the high vegetation cover study area.

[0068] S1.2: High-resolution DEM (Digital Elevation Model) data obtained by TIN interpolation of ground point cloud data is used to perform orthorectification on hyperspectral data of high vegetation cover study areas to eliminate image distortion caused by topographic relief and other reasons.

[0069] S1.3: The lidar point cloud data was processed into a canopy height model (CHM), and the nearest neighbor resampling function in ArcGIS software was used to resample the data to the same spatial resolution as the hyperspectral data of the high vegetation cover study area.

[0070] S1.4: Using ArcGIS software, select typical land features as control points (such as the boundary of the high vegetation cover area, surrounding roads, houses, and other obvious land features) on both hyperspectral data and lidar point cloud data of the high vegetation cover study area, and accurately register the two images.

[0071] It should be noted that a hyperspectral-LiDAR integrated UAV platform was used to acquire two types of vegetation canopy data with high matching degree, and the two types of data were registered with high precision. The three-dimensional data and two-dimensional data were fused to achieve the purpose of vertical structure layering of hyperspectral data, so that the vegetation interference spectral information was suppressed to the greatest extent and the accuracy of automated extraction by deep learning was greatly improved.

[0072] Based on, but not limited to, the above methods, the specific implementation process of step S2 is as follows:

[0073] To layer the lidar point cloud data, Cloud Compare software was used to layer the lidar point cloud data using intervals of 1 meter.

[0074] Based on, but not limited to, the above methods, the specific implementation process of step S5 is as follows:

[0075] Based on the abnormal spectral information, the stressed vegetation was extracted from the hyperspectral image of the bottom of the vegetation canopy. The extracted hyperspectral image of the bottom of the stressed vegetation canopy was cropped into an image of size 512×512 pixels and made into a sample.

[0076] Based on, but not limited to, the above methods, in step S6, all samples are divided into training set and validation set in a ratio of 8:2 for subsequent model training.

[0077] Based on, but not limited to, the above methods, step S7, which involves constructing a prediction model for the target mining area range in high vegetation cover areas using DeeplabV3+ deep learning, includes:

[0078] A prediction model for the target mining area range in high vegetation cover areas was constructed using DeeplabV3+ as the base model and MobileNetV2 as the backbone network in a two-branch structure. For example... Figure 3As shown, the Encoder section mainly consists of two parts: the backbone and the ASPP module. The backbone module adopts a two-branch structure with MobileNet V2 as the backbone network, which reduces the number of parameters while enabling the model to simultaneously extract features from both hyperspectral data and lidar point cloud data of the high vegetation cover research area. The ASPP module takes the first part of the backbone output as input and uses four dilated convolutional blocks with different dilation rates (including convolution, batch normalization, and activation layers) and a global average pooling block (including pooling, convolution, batch normalization, and activation layers) to obtain a total of five sets of feature maps. These maps are concatenated using concat and then passed through a 1*1 convolutional block (including convolution, batch normalization, activation, and dropout layers) before being fed into the Decoder module. The Decoder module receives the output from the MobileNetV2 module and the output from the ASPP module as input. First, the low-level feature maps output by the MobileNetV2 module are subjected to channel dimensionality reduction using 1*1 convolutions, from 256 to 48. Then, the feature maps from ASPP are interpolated and upsampled to obtain feature maps of the same size as the low-level feature maps. Next, the low-level feature maps with channel dimensionality reduction and the feature maps obtained by linear interpolation and upsampling are concatenated using concat and fed into a set of 3*3 convolutional blocks for processing. Finally, linear interpolation and upsampling are performed again to obtain a prediction map with the same resolution as the original image.

[0079] The key points in the implementation of this invention are mainly as follows:

[0080] Key Point 1: This invention utilizes the approach of feature extraction and fusion of hyperspectral data and lidar point cloud data. It precisely registers lidar point cloud data with hyperspectral data, combining three-dimensional vertical structural information with rich spectral information, allowing vegetation spectral information to be distributed across the three-dimensional vertical structure. This effectively eliminates interference from vegetation canopy surface spectra and redundant mixed spectral information, resulting in more accurate extraction of anomalous spectral information from stressed vegetation in high-vegetation areas. Compared to extracting stressed vegetation information using only hyperspectral data, the extraction accuracy and subsequent sample accuracy are significantly improved, providing greater support for automatic extraction using deep learning.

[0081] Key Point 2: In real-world environments, due to the high canopy density in areas with high vegetation cover, extracting stressed vegetation using traditional deep learning methods is extremely difficult when precise samples are lacking. This invention constructs a multimodal deep learning-based model for predicting the extent of target mining areas in areas with high vegetation cover by using DeeplabV3+ as the base model and MobileNetV2 as the backbone network. This model allows for the extraction of more representative features while simultaneously providing high-precision input samples, thus improving its ability to mine deep feature semantics. Furthermore, it significantly improves the accuracy of extracting weak information such as the spectral information of stressed vegetation in areas with high vegetation cover.

[0082] The beneficial effects of implementing this invention are as follows:

[0083] Compared with existing technologies, the technical solution proposed in this invention focuses on the significant difference in spectral information anomalies observed at the bottom of the canopy of stressed vegetation compared to the middle and top sections, highlighting a distinct stratification characteristic. Furthermore, it leverages the rich spectral information from hyperspectral remote sensing and the abundant spatial information from lidar, fusing these two datasets to achieve a vertically stratified distribution of vegetation canopy spectral information in a three-dimensional structure, eliminating interference from the top of the canopy and redundant mixed spectra. The selected anomalous vegetation canopy sections are then used to create high-precision samples. These high-precision samples are used to train the DeeplabV3+ semantic segmentation model, enabling the model to extract more accurate and discriminative features, thus improving the ability to mine deep feature semantics. This solves the problem of traditional classification methods lacking accurate samples and self-learning capabilities, which are unable to handle the extraction of mineral alteration vegetation stress information in high-vegetation areas. A method for extracting rock and mineral alteration information in high-vegetation areas is established, improving the efficiency of target mining area location.

[0084] The following describes a device for predicting the range of a target mining area in a high vegetation cover area provided by the present invention. The device for predicting the range of a target mining area in a high vegetation cover area described below can be referred to in correspondence with the method for predicting the range of a target mining area in a high vegetation cover area described above.

[0085] like Figure 4 As shown, a device for predicting the range of a target mining area in a high vegetation cover area includes the following modules:

[0086] The acquisition and registration module 001 is used to acquire hyperspectral data and lidar point cloud data of the high vegetation cover study area and perform registration.

[0087] Point cloud layering module 002 is used to layer the acquired lidar point cloud data.

[0088] The interval mask module 003 is used to create interval masks of each height from each layer of point cloud data.

[0089] The canopy bottom image extraction module 004 is used to spatially layer hyperspectral data based on masks of various height intervals and extract hyperspectral images of the vegetation canopy bottom.

[0090] The sample preparation module 005 is used to extract stressed vegetation from hyperspectral images at the bottom of the vegetation canopy based on abnormal spectral information and prepare samples.

[0091] The sample partitioning module 006 is used to divide all samples into training set and validation set according to a preset ratio;

[0092] Model building module 007 is used to build a target mining area range prediction model based on DeeplabV3+ deep learning;

[0093] Model training module 008 is used to train and validate the prediction model of the target mining area in high vegetation cover area using training set and validation set respectively, until a well-trained prediction model is obtained.

[0094] The target mining area prediction module 009 is used to acquire hyperspectral images of the target high vegetation cover area and predict the range of the target mining area within the target high vegetation cover area using a trained prediction model.

[0095] like Figure 4As shown, a schematic diagram of the physical structure of an electronic device is illustrated. The electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call logic instructions in the memory 630 to execute the aforementioned method for predicting the range of a target mining area in a high vegetation cover region, including: acquiring hyperspectral data and lidar point cloud data of the high vegetation cover research area and performing registration; performing point cloud data layering on the acquired lidar point cloud data; creating masks for each height interval of the point cloud data layer; performing spatial layering on the hyperspectral data based on the masks for each height interval and extracting hyperspectral images of the bottom of the vegetation canopy; extracting stressed vegetation from the hyperspectral images of the bottom of the vegetation canopy based on anomalous spectral information and creating samples; dividing all samples into training and validation sets according to a preset ratio; constructing a target mining area range prediction model under high vegetation cover region based on deep learning DeeplabV3+; training and validating the target mining area range prediction model under high vegetation cover region through the training and validation sets respectively until a trained prediction model is obtained; acquiring hyperspectral images of the target high vegetation cover region and predicting the target mining area range within the target high vegetation cover region using the trained prediction model.

[0096] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0097] In another aspect, embodiments of the present invention also provide a storage medium storing a computer program. When executed by a processor, this computer program implements the aforementioned method for predicting the range of a target mining area in a high vegetation cover region, comprising: acquiring hyperspectral data and lidar point cloud data of a high vegetation cover research area and performing registration; performing point cloud data layering on the acquired lidar point cloud data; creating masks for each height interval of the point cloud data layer; performing spatial layering of the hyperspectral data based on the masks for each height interval and extracting hyperspectral images of the bottom of the vegetation canopy; extracting stressed vegetation from the hyperspectral images of the bottom of the vegetation canopy based on abnormal spectral information and creating samples; dividing all samples into training and validation sets according to a preset ratio; constructing a target mining area range prediction model under high vegetation cover region based on deep learning DeeplabV3+; training and validating the target mining area range prediction model under high vegetation cover region through the training and validation sets respectively until a trained prediction model is obtained; acquiring hyperspectral images of the target high vegetation cover region and predicting the target mining area range within the target high vegetation cover region using the trained prediction model.

[0098] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0099] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In the unit claims listing several devices, several of these devices may be embodied by the same hardware item. The use of the terms first, second, and third, etc., does not indicate any order and can be interpreted as identifiers.

[0100] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for predicting the extent of a target mining area in a high-vegetation-covered area, characterized by, The method comprises the following steps: obtaining hyperspectral data and laser radar point cloud data of a high-vegetation-covered research area and performing registration; performing point cloud data layering on the obtained laser radar point cloud data; manufacturing each layer of point cloud data into a mask of each height interval; performing spatial layering on the hyperspectral data based on the mask of each height interval, and extracting a hyperspectral image of a bottom of a vegetation canopy; extracting stressed vegetation from the hyperspectral image of the bottom of the vegetation canopy according to abnormal spectral information, and manufacturing samples; dividing all the samples into a training set and a verification set according to a preset ratio; constructing a target mining area range prediction model under a high-vegetation-covered area based on deep learning DeeplabV3+; training and verifying the target mining area range prediction model under the high-vegetation-covered area through the training set and the verification set respectively until a trained prediction model is obtained; obtaining a hyperspectral image of a target high-vegetation-covered area, and predicting a target mining area range in the target high-vegetation-covered area through the trained prediction model; the step of extracting stressed vegetation from the hyperspectral image of the bottom of the vegetation canopy according to abnormal spectral information, and manufacturing samples comprises: extracting stressed vegetation from the hyperspectral image of the bottom of the vegetation canopy according to abnormal spectral information, cutting the extracted stressed vegetation canopy bottom hyperspectral image into an image with a size of 512*512 pixels, and manufacturing samples.

2. The method of claim 1, wherein the high-vegetation-covered area target mining area range prediction method is characterized by, The step of obtaining hyperspectral data and laser radar point cloud data of a high-vegetation-covered research area and performing registration comprises: obtaining hyperspectral data and laser radar point cloud data of a high-vegetation-covered research area; performing orthographic correction on the hyperspectral data of the high-vegetation-covered research area by using high-resolution DEM data obtained by TIN interpolation of ground point cloud data, and eliminating image deformation caused by terrain undulations and the like; processing the laser radar point cloud data into a canopy height model, and resampling the laser radar point cloud data into the same spatial resolution as the hyperspectral data of the high-vegetation-covered research area by using a nearest neighbor method; selecting typical ground objects as control points on the hyperspectral data and the laser radar point cloud data of the high-vegetation-covered research area, and performing registration on the two images.

3. The method of claim 1, wherein the high-vegetation-covered area target mining area range prediction method is characterized by, In the research area, hyperspectral data and laser radar point cloud data of a high-vegetation-covered research area are simultaneously obtained by using a hyperspectral-laser radar integrated unmanned aerial vehicle platform.

4. The method of claim 1, wherein the high-vegetation-covered area target mining area range prediction method is characterized by, The step of performing point cloud data layering on the obtained laser radar point cloud data comprises: performing layering on the laser radar point cloud data by using an interval section with an interval of 1 m by using Cloud Compare software.

5. The method of claim 1, wherein the method further comprises: determining a target area of interest based on the high-vegetation coverage area. The step of dividing all the samples into a training set and a verification set according to a preset ratio comprises: dividing all the samples into a training set and a verification set according to a ratio of 8:

2.

6. The method of claim 1, wherein the method further comprises: determining a target area of interest based on the high-vegetation coverage area. The step of constructing a target mining area range prediction model under a high-vegetation-covered area based on deep learning DeeplabV3+ comprises: using deep learning DeeplabV3+ as a basic model, and using a MobileNetV2 as a two-branch structure of a main network to construct a target mining area range prediction model under a high-vegetation-covered area.

7. An apparatus for predicting the extent of a target mining area in a high-vegetation-covered area, which implements the method according to any one of claims 1 to 6, characterized by, The method comprises the following modules: an obtaining and registration module, used for obtaining hyperspectral data and laser radar point cloud data of a high-vegetation-covered research area and performing registration; The point cloud layering module is configured to perform point cloud data layering on the obtained laser radar point cloud data. The interval mask module is configured to manufacture each layer of point cloud data into a height interval mask. The canopy bottom image extraction module is configured to perform spatial layering on hyperspectral data based on the height interval mask, and extract a hyperspectral image of a canopy bottom of vegetation. The sample manufacturing module is configured to extract stressed vegetation from the hyperspectral image of the canopy bottom of vegetation according to abnormal spectral information, and manufacture the stressed vegetation into a sample. The sample division module is configured to divide all samples into a training set and a validation set according to a preset proportion. The model construction module is configured to construct a target mining area range prediction model under high vegetation coverage based on a deep learning DeeplabV3+. The model training module is configured to train and verify the target mining area range prediction model under high vegetation coverage by using the training set and the validation set, respectively, until a trained prediction model is obtained. The target mining area prediction module is configured to obtain a hyperspectral image of a target high vegetation coverage area, and predict a target mining area range in the target high vegetation coverage area by using the trained prediction model. The step of extracting stressed vegetation from the hyperspectral image of the canopy bottom of vegetation according to abnormal spectral information, and manufacturing the stressed vegetation into a sample, includes: extracting stressed vegetation from the hyperspectral image of the canopy bottom of vegetation according to abnormal spectral information, cutting the extracted stressed vegetation canopy bottom hyperspectral image into an image with a size of 512x512 pixels, and manufacturing the image into a sample.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the high vegetation coverage area target mining area range prediction method according to any one of claims 1-6 when executing the program.

9. A storage medium having stored thereon a computer program, characterized in that The computer program implements the steps of the high vegetation coverage area target mining area range prediction method according to any one of claims 1-6 when executed by the processor.

Images