Hyperspectral image lake DLG extraction method and device based on GPU acceleration

Abstract

The invention provides a hyperspectral image lake DLG extraction method and device based on GPU acceleration. According to the hyperspectral image lake DLG extraction method and device based on GPU acceleration, rough extraction of water body pixels is carried out on hyperspectral image data in a CPU, a water body vector set is constructed, the data transmission amount between the CPU and the GPU is reduced, the utilization rate of the GPU is improved, and the processing efficiency is further improved; when the pixels of the lake water system are accurately judged, a parallel architecture (NVIDIA CUDA) of a GPU is used for synchronously processing tasks such as preprocessing of pixel points of tens of thousands of hyperspectral images containing hundreds of wave bands, lake pixel extraction and lake DLG extraction, so that manual participation in calculation is avoided, and the feasibility of large-scale application is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of hyperspectral image processing, and more specifically, to a method and device for extracting lake DLG from hyperspectral images based on GPU acceleration. Background Art

[0002] With the continuous development and integration of computer technology and spatial information technology, surveyors are no longer satisfied with the early method of using surveying and mapping technology to collect large-scale topographic maps in the field for lake DLG collection, because this method is not only greatly affected by the environment but also has very low efficiency.

[0003] Remote sensing technology is the ability to detect and identify targets from a distance by sensing electromagnetic waves, visible light, and infrared radiation reflected or radiated by them. Hyperspectral remote sensing is a new Earth observation technology that emerged in the 1980s. It acquires image data of very narrow, spectrally continuous wavelengths in the ultraviolet, visible, near-infrared, and mid-infrared regions of the electromagnetic spectrum. A hyperspectral imaging spectrometer provides tens to hundreds of narrow-band spectral information for each pixel, generating a spectral curve that maps spectral bands to spectral values. Therefore, hyperspectral images contain not only two-dimensional spatial information but also a wealth of spectral information. Different substances exhibit different spectral signals in different bands. Based on these differences in spectral curves, we can perform DLG acquisition on lakes in hyperspectral images.

[0004] However, the above method of collecting lake DLG through hyperspectral images still requires a lot of computing time, which is difficult to meet the processing time requirements of surveyors. Summary of the Invention

[0005] In view of the problem of low efficiency in hyperspectral image data processing in the prior art, the present invention provides a method and device for extracting lake DLG from hyperspectral images based on GPU acceleration.

[0006] According to a first aspect of the present invention, a method for extracting lake DLG from hyperspectral images based on GPU acceleration is provided, comprising:

[0007] Preprocessing the hyperspectral image data in the CPU, performing rough extraction on the water body area in the hyperspectral image data to generate a rough extracted water body vector set, and transmitting the rough extracted water body vector set to the GPU, wherein the rough extracted water body vector set records the spectral data of each roughly extracted water body pixel;

[0008] Deploy multiple CUDA cores in the GPU and assign each CUDA core a water pixel matching task. The matching task is to match the spectral data of the current water pixel with each spectrum in the spectral library and determine whether the current water pixel is a lake water system pixel based on the matching results.

[0009] Generate a result matrix based on the determination results of each water pixel by multiple CUDA cores, and transmit the result matrix to the CPU;

[0010] In the CPU, a binary image is constructed according to the result matrix, the binary image is sent to the GPU, and an opening operation is performed on the binary image in the GPU to generate a lake pixel image;

[0011] A lake DLG is extracted from the lake pixel image.

[0012] According to a second aspect of the present invention, a hyperspectral image lake DLG extraction device based on GPU acceleration is provided, comprising a preprocessing module, a lake pixel extraction module, a binary image construction module and a lake DLG extraction module, wherein the preprocessing module and the binary image construction module are located in the CPU, and the lake pixel extraction module and the lake DLG extraction module are located in the GPU;

[0013] The preprocessing module is used to preprocess the hyperspectral image data, perform rough extraction on the water body area in the hyperspectral image data, generate a rough extracted water body vector set, and transmit the rough extracted water body vector set to the GPU, wherein the rough extracted water body vector set records the spectral data of each roughly extracted water body pixel;

[0014] The lake pixel extraction module completes the matching task of all water pixels based on multiple CUDA cores deployed in the GPU, wherein one CUDA core is assigned a water pixel matching task, wherein the matching task is to match the current water pixel with each spectrum in the spectral library, and based on the matching results, determine whether the current water pixel is a lake water system pixel; and based on the determination results of each water pixel by the multiple CUDA cores, generate a result matrix, and transmit the result matrix to the CPU;

[0015] The binary image construction module is used to construct a binary image according to the result matrix and send the binary image to the GPU;

[0016] The lake DLG extraction module is used to perform an opening operation on the binary image to generate a lake pixel image; and to extract the lake DLG from the binary image.

[0017] The present invention provides a GPU-accelerated hyperspectral imagery lake DLG extraction method and device. This method uses the CPU to perform a coarse extraction of water pixels from hyperspectral image data and construct a water vector set. This reduces data transmission between the CPU and GPU, improves GPU utilization, and further enhances processing efficiency. When accurately determining lake water system pixels, the GPU's parallel architecture (NVIDIA CUDA) is used to simultaneously process tasks such as preprocessing tens of thousands of hyperspectral image pixels encompassing hundreds of bands, extracting lake pixels, and extracting lake DLGs. This eliminates manual computational effort and enhances the feasibility of large-scale applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flow chart of the GPU-accelerated hyperspectral imagery lake DLG extraction method provided by the present invention;

[0019] Figure 2 This is a flowchart of preprocessing hyperspectral image data according to an embodiment of the present invention;

[0020] Figure 3 A schematic diagram of the process of extracting lake water system pixels according to an embodiment of the present invention;

[0021] Figure 4 This is an overall flow chart of lake water system pixel extraction according to an embodiment of the present invention;

[0022] Figure 5 This is a structural block diagram of a GPU-accelerated hyperspectral imagery lake DLG extraction device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, the technical features in the various embodiments or single embodiments provided by the present invention can be arbitrarily combined with each other to form a feasible technical solution. This combination is not restricted by the sequence of steps and / or structural composition mode, but must be based on the ability of ordinary technicians in this field to implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0024] Existing methods for extracting lake DLGs from hyperspectral imagery suffer from two main issues: 1. Low automation. Existing methods require manual adjustment of threshold parameters or rely on prior knowledge to set feature extraction rules, which limits their generalization capabilities when applied across regions. 2. Due to hardware bottlenecks, the data transfer rate between the CPU and GPU cannot match the GPU's computing speed. Furthermore, due to the large amount of data inherent in hyperspectral imagery, most processing time is spent on data transfer, preventing the GPU from fully utilizing its processing capabilities.

[0025] This invention addresses the shortcomings of existing solutions by utilizing the full spectral information of hyperspectral images for matching operations, fully utilizing the image's complete information. This design significantly increases the amount of computation required. Therefore, by introducing GPU acceleration technology, the computational process is optimized, significantly improving computational efficiency. Furthermore, since hyperspectral images are large, transferring them between the CPU and GPU consumes significant time. Therefore, this invention reduces data transfer between the CPU and GPU by constructing a set of coarsely extracted water body vectors, thereby improving efficiency.

[0026] Figure 1 The present invention provides a flow chart of a GPU-accelerated hyperspectral image lake DLG extraction method, such as Figure 1 As shown, the method includes:

[0027] Step 1: Preprocess the hyperspectral image data in the CPU, perform coarse extraction on the water area in the hyperspectral image data, generate a coarsely extracted water body vector set, and transmit the coarsely extracted water body vector set to the GPU, wherein the coarsely extracted water body vector set records the spectral data of each coarsely extracted water body pixel.

[0028] See also Figure 2 The hyperspectral image data is preprocessed in the CPU. The preprocessing of the hyperspectral image data mainly includes radiometric calibration, atmospheric correction, orthorectification processing, and rough extraction of water body areas to obtain a water body vector set.

[0029] The preprocessing of the hyperspectral image data in the CPU, performing rough extraction of the water body area in the hyperspectral image data, and generating a rough extracted water body vector set include:

[0030] Step 11, inputting an RPC file, hyperspectral image data, and hyperspectral image metadata into the CPU, wherein the RPC file includes geometric correction parameters;

[0031] Step 12, performing radiometric calibration on the hyperspectral image data according to the input hyperspectral image data, the hyperspectral image metadata, and radiometric calibration parameters;

[0032] Step 13, performing atmospheric correction on the radiometrically calibrated hyperspectral image data, and performing geometric correction based on the geometric correction parameters;

[0033] Step 14: performing rough extraction of water body regions on the geometrically corrected hyperspectral image data to generate a rough extracted water body vector set.

[0034] Specifically, due to the large amount of hyperspectral image data, the complete transmission between the CPU and GPU will consume a lot of time, and most of the pixels in a single image are non-water pixels, which increases the computational workload of the GPU. Therefore, a rough extraction of water areas is performed after radiometric calibration, atmospheric correction, and geometric correction to preliminarily screen water pixels.

[0035] Because the reflectivity of the near-infrared band has better discrimination performance for water bodies and most non-water objects on the land surface than the reflectivity of other bands (such as blue light band, green light band and red light band), in the present invention, the reflectivity threshold of the near-infrared band can be used to distinguish most non-water objects from water bodies in the target hyperspectral image, so as to extract the coarse water body area from the target hyperspectral image.

[0036] Therefore, step 41, performing rough extraction of water body regions on the geometrically corrected hyperspectral image data to generate a rough extracted water body vector set, includes:

[0037] Obtaining whether the reflectance of the near-infrared band of each pixel in the geometrically corrected hyperspectral image data is less than a set threshold; if so, the pixel is preliminarily determined to be a water pixel; otherwise, the pixel is a non-water pixel;

[0038] Traversing each pixel in the hyperspectral image data, and roughly extracting all water pixels from the hyperspectral image data;

[0039] A coarsely extracted water body vector set is constructed based on the position information of each water body pixel and the spectral value of each band, wherein the coarsely extracted water body vector set includes multiple vector data, one vector data includes the position information of a water body pixel and the spectral value of each pixel band, the position information of the water body pixel is represented by the row and column where the water body pixel is located, and the representation form of the coarsely extracted water body vector set can be seen in Table 1 below.

[0040] Table 1 Roughly extracted water body vector set

[0041]

[0042] Table 1 records the relevant data of each roughly extracted water pixel, including the row and column position of the pixel and the spectral value of each pixel band. A vector data represents the relevant information of a water pixel.

[0043] Step 2: Deploy multiple CUDA cores in the GPU and assign a water pixel matching task to each CUDA core. The matching task is to match the spectral data of the current water pixel with each spectrum in the spectral library. Based on the matching results, determine whether the current water pixel is a lake water system pixel.

[0044] It's understandable that step 1, which roughly extracts water pixels from the hyperspectral image data, isn't very accurate, as it misidentifies some non-water pixels as water. Therefore, further precise determination of whether each water pixel represents a lake is necessary. This precise determination is performed on the GPU, so step 1 performs a coarse extraction of water pixels, which reduces the amount of data subsequently transmitted between the CPU and GPU, shortening data transfer time.

[0045] See Figure 3 The present invention uses spectral angle matching to extract water system pixels by matching the spectrum of each water body pixel in the rough extracted water body vector set output by preprocessing with the spectrum in the spectral library, and then performs an opening operation on the water system pixels to remove non-lake water system pixels and output lake water system pixels.

[0046] Since the core mathematical operations for spectral angle matching involve performing vector dot products, modulus calculations, and arccosine operations on each roughly extracted water vector, the water vector set is loaded into GPU memory and batched using CUDA kernel functions. OpenCV already provides the morphologyEx() interface for GPU-accelerated morphological operations through precompiled CUDA code.

[0047] Since the amount of data to be processed is relatively large, multiple CUDA cores are deployed in the GPU, and each CUDA core is responsible for processing the matching judgment task of a water body pixel. In one embodiment of the present invention, step 2 specifically includes:

[0048] Step 21: In each CUDA core, traverse each spectrum in the spectral library, calculate the spectral angle between each spectrum and the spectrum of the water pixel in the current vector data, and obtain m spectral angles corresponding to the current water pixel, where m is the number of spectra in the spectral library;

[0049] Step 22: If one or more spectral angles among the m spectral angles are smaller than the spectral angle threshold, the current water pixel is determined to be a lake water pixel;

[0050] Step 23: If all the spectral angles in the m spectral angles are greater than the spectral angle threshold, it is determined that the current water body pixel is a non-lake water system pixel.

[0051] Specifically, the GPU accelerated extraction process for lake water pixels is as follows: Figure 4 As shown in the figure, the CPU first sends the preprocessed, roughly extracted water vector set and the spectral library to the GPU. Then, each CUDA core in the GPU is responsible for processing a water vector data set. Each CUDA core in the GPU traverses each spectrum in the spectral library and calculates the spectral angle between each spectrum and the spectrum of the water pixel in the current water vector. If the spectral angle between a spectrum in the spectral library and the spectrum of the current water pixel is less than 1.0°, the water pixel is indeed a lake pixel. If the spectral angle between each spectrum in the spectral library and the current water pixel is greater than 1.0°, the water pixel is not a lake pixel.

[0052] Through multiple CUDA cores in the GPU, each water pixel in the coarsely extracted water vector set is accurately judged to determine whether each water pixel is a lake water system pixel.

[0053] If the number of water pixels in the roughly extracted water vector set is large, but the number of CUDA cores in the GPU is limited, all vector data can be divided into n batches, where n ≥ 2 and n is a positive integer. For each batch of vector data, the matching task is executed in parallel using multiple CUDA cores. After processing one batch of water pixel data using multiple CUDA cores, another batch of water pixel data is processed.

[0054] Step 3: Based on the determination results of each water pixel by multiple CUDA cores, a result matrix is ​​generated, and the result matrix is ​​transmitted to the CPU.

[0055] It can be understood that after step 2 accurately determines whether each water body pixel in the coarsely extracted water body vector set is a lake water system pixel, the determination result of whether each water body pixel is a lake water system pixel is recorded in the result matrix, and the result matrix includes the row and column position information and identification information of each water body pixel. The identification information represents whether the water body pixel is a lake water system pixel, and the constructed result matrix is ​​transmitted to the CPU.

[0056] Step 4: In the CPU, a binary image is constructed according to the result matrix, and the binary image is sent to the GPU. An opening operation is performed on the binary image in the GPU to generate a lake pixel image.

[0057] Step 5: extract the lake DLG from the binary image.

[0058] It is understandable that the lake DLG is extracted from the lake pixel image generated in step 4. A DLG (Digital Line Graphic) is a basic dataset that stores geographic features in vector form, containing both spatial coordinates (such as roads, water systems, and settlement boundaries) and attribute information (such as name and type).

[0059] In one embodiment, step 5 specifically includes:

[0060] Based on the boundary tracing method, lake DLG is extracted from lake pixel images to generate surface features with topological relationships.

[0061] The step of extracting the lake DLG from the binary image based on the boundary tracing method includes:

[0062] Based on the boundary tracing method, the boundary pixel points of each lake water system area are scanned pixel by pixel to extract the boundary outline of each lake water system area;

[0063] Among them, multiple lake and water system areas are allocated to multiple GPU thread blocks, and the extraction of boundary contours of multiple lake and water system areas is processed in parallel based on multiple GPU thread blocks.

[0064] Specifically, lake DLG extraction involves extracting boundary outline pixels from the lake water system. In one embodiment of the present invention, a boundary tracing method is used to extract lake DLG. If there are a large number of lake water systems, or if the lake water system is large, resulting in a large number of boundary pixels, multiple GPU thread blocks can be considered, with each thread responsible for tracking a single boundary segment and managing the global path state through atomic operations, significantly improving efficiency.

[0065] See also Figure 5 , the present invention provides a hyperspectral image lake DLG extraction device based on GPU acceleration, which includes a preprocessing module 501, a lake pixel extraction module 502, a binary image construction module 503 and a lake DLG extraction module 504, wherein the preprocessing module 501 and the binary image construction module 503 are located in the CPU, and the lake pixel extraction module 502 and the lake DLG extraction module 504 are located in the GPU.

[0066] A preprocessing module 501 is configured to preprocess the hyperspectral image data, perform a rough extraction of the water region in the hyperspectral image data, generate a rough extracted water vector set, and transmit the rough extracted water vector set to the GPU, wherein the rough extracted water vector set records the spectral data of each roughly extracted water pixel;

[0067] Lake pixel extraction module 502 completes the matching task of all water pixels based on multiple CUDA cores deployed in the GPU, wherein one CUDA core is assigned a water pixel matching task, wherein the matching task is to match the current water pixel with each spectrum in the spectral library, and based on the matching results, determine whether the current water pixel is a lake water system pixel; and based on the determination results of each water pixel by the multiple CUDA cores, generate a result matrix, and transmit the result matrix to the CPU;

[0068] A binary image construction module 503 is configured to construct a binary image according to the result matrix and send the binary image to the GPU;

[0069] The lake DLG extraction module 504 is configured to perform an opening operation on the binary image to generate a lake pixel image; and extract the lake DLG from the binary image.

[0070] The preprocessing module 501 is used to preprocess the hyperspectral image data, perform rough extraction on the water body area in the hyperspectral image data, generate a rough extracted water body vector set, and transmit the rough extracted water body vector set to the GPU, including:

[0071] Performing radiometric calibration, atmospheric correction, and geometric correction on the hyperspectral image data;

[0072] Determine whether the reflectance of the near-infrared band of each pixel in the hyperspectral image data after geometric correction is less than a set threshold; if so, preliminarily determine that the pixel is a water pixel;

[0073] Traversing each pixel in the hyperspectral image data, and roughly extracting all water pixels therefrom;

[0074] A coarsely extracted water body vector set is constructed based on the position information of each water body pixel and the spectral value of each band, wherein the coarsely extracted water body vector set includes multiple vector data, one vector data includes the position information of a water body pixel and the spectral value of each pixel band, and the position information of the water body pixel is represented by the row and column where the water body pixel is located.

[0075] The lake pixel extraction module 502 completes the matching task of all water body pixels based on multiple CUDA cores deployed in the GPU, including:

[0076] Assign a vector data matching task to each CUDA core, traverse each spectrum in the spectral library, calculate the spectral angle between each spectrum and the spectrum of the water pixel in the current vector data, and obtain m spectral angles corresponding to the current water pixel, where m is the number of spectra in the spectral library;

[0077] If one or more of the m spectral angles is smaller than the spectral angle threshold, the current water pixel is determined to be a lake water pixel;

[0078] If all the spectral angles among the m spectral angles are greater than the spectral angle threshold, the current water body pixel is determined to be a non-lake water system pixel.

[0079] It can be understood that the hyperspectral image lake DLG extraction system based on GPU acceleration provided by the present invention corresponds to the hyperspectral image lake DLG extraction method based on GPU acceleration provided by the aforementioned embodiments. The relevant technical features of the hyperspectral image lake DLG extraction system based on GPU acceleration can refer to the relevant technical features of the hyperspectral image lake DLG extraction method based on GPU acceleration, which will not be repeated here.

[0080] The embodiments of the present invention provide a method and device for extracting lake DLG from hyperspectral images based on GPU acceleration, which has the following advantages:

[0081] (1) Using the parallel architecture of GPU (NVIDIA CUDA) to simultaneously process tasks such as preprocessing of tens of thousands of hyperspectral image pixels containing hundreds of bands, lake pixel extraction, and lake DLG extraction, the calculations can be performed without manual intervention, thus enhancing the feasibility of large-scale applications.

[0082] (2) By constructing a coarsely extracted water body vector set, the amount of data transmission between the CPU and GPU is reduced, the utilization rate of the GPU is improved, and the processing efficiency is further improved.

[0083] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0084] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0085] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0086] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0087] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0088] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0089] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A GPU-accelerated hyperspectral image lake DLG extraction method, characterized in that: include: Preprocessing the hyperspectral image data in the CPU, performing rough extraction on the water body area in the hyperspectral image data to generate a rough extracted water body vector set, and transmitting the rough extracted water body vector set to the GPU, wherein the rough extracted water body vector set records the spectral data of each roughly extracted water body pixel; Deploy multiple CUDA cores in the GPU and assign each CUDA core a water pixel matching task. The matching task is to match the spectral data of the current water pixel with each spectrum in the spectral library and determine whether the current water pixel is a lake water system pixel based on the matching results. Generate a result matrix based on the determination results of each water pixel by multiple CUDA cores, and transmit the result matrix to the CPU; In the CPU, a binary image is constructed according to the result matrix, the binary image is sent to the GPU, and an opening operation is performed on the binary image in the GPU to generate a lake pixel image; A lake DLG is extracted from the lake pixel image.

2. The hyperspectral image lake DLG extraction method according to claim 1 is characterized in that: The method of preprocessing the hyperspectral image data in the CPU, performing rough extraction on the water body area in the hyperspectral image data, and generating a rough extracted water body vector set includes: Inputting an RPC file, hyperspectral image data, and hyperspectral image metadata into the CPU, wherein the RPC file includes geometric correction parameters; Performing radiometric calibration on the hyperspectral image data according to the input hyperspectral image data, the hyperspectral image metadata and radiometric calibration parameters; Performing atmospheric correction on the radiometrically calibrated hyperspectral image data, and performing geometric correction based on the geometric correction parameters; The water body region is roughly extracted from the geometrically corrected hyperspectral image data to generate a set of roughly extracted water body vectors.

3. The hyperspectral image lake DLG extraction method according to claim 1 or 2, characterized in that: The step of performing rough extraction of water body regions on the geometrically corrected hyperspectral image data to generate a rough extracted water body vector set includes: Obtaining whether the reflectance of the near-infrared band of each pixel in the geometrically corrected hyperspectral image data is less than a set threshold; if so, the pixel is preliminarily determined to be a water pixel; otherwise, the pixel is determined to be a non-water pixel; Traversing each pixel in the hyperspectral image data, and roughly extracting all water pixels from the hyperspectral image data; A coarsely extracted water body vector set is constructed based on the position information of each water body pixel and the spectral value of each band, wherein the coarsely extracted water body vector set includes multiple vector data, one vector data includes the position information of a water body pixel and the spectral value of each pixel band, and the position information of the water body pixel is represented by the row and column where the water body pixel is located.

4. The hyperspectral image lake DLG extraction method according to claim 1 is characterized in that: The method deploys multiple CUDA cores in the GPU and assigns a water pixel matching task to each CUDA core. The matching task is to match the spectral data of the current water pixel with each spectrum in the spectral library, and based on the matching results, determine whether the current water pixel is a lake water system pixel, including: In each CUDA core, each spectrum in the spectral library is traversed, and the spectral angle between each spectrum and the spectrum of the water pixel in the current vector data is calculated to obtain m spectral angles corresponding to the current water pixel, where m is the number of spectra in the spectral library; If one or more of the m spectral angles is smaller than the spectral angle threshold, the current water pixel is determined to be a lake water pixel; If all the spectral angles among the m spectral angles are greater than the spectral angle threshold, the current water body pixel is determined to be a non-lake water system pixel.

5. The hyperspectral image lake DLG extraction method according to claim 4 is characterized in that: When the number of CUDA cores is less than the number of vector data, all vector data are divided into n batches, where n≥2 and n is a positive integer. For each batch of vector data, the matching task is executed in parallel based on multiple CUDA cores.

6. The hyperspectral image lake DLG extraction method according to claim 1, characterized in that: The method of generating a result matrix based on the determination result of each water body pixel by multiple CUDA cores and transmitting the result matrix to the CPU includes: The determination result of whether each water body pixel is a lake water system pixel is recorded in a result matrix, which includes row and column position information and identification information of each water body pixel, and the identification information represents whether the water body pixel is a lake water system pixel.

7. The method for extracting lake DLG from hyperspectral imagery according to claim 1, characterized in that: Extracting the lake DLG from the binary image includes: Extracting the lake DLG from the binary image based on the boundary tracing method to generate surface features with topological relationships; The step of extracting the lake DLG from the binary image based on the boundary tracing method includes: Based on the boundary tracing method, the boundary pixel points of each lake water system area are scanned pixel by pixel to extract the boundary outline of each lake water system area; Among them, multiple lake and water system areas are allocated to multiple GPU thread blocks, and the extraction of boundary contours of multiple lake and water system areas is processed in parallel based on multiple GPU thread blocks.

8. A GPU-accelerated hyperspectral image lake DLG extraction device, characterized in that: It includes a pre-processing module, a lake pixel extraction module, a binary image construction module and a lake DLG extraction module, wherein the pre-processing module and the binary image construction module are located in the CPU, and the lake pixel extraction module and the lake DLG extraction module are located in the GPU; The preprocessing module is used to preprocess the hyperspectral image data, perform rough extraction on the water body area in the hyperspectral image data, generate a rough extracted water body vector set, and transmit the rough extracted water body vector set to the GPU, wherein the rough extracted water body vector set records the spectral data of each roughly extracted water body pixel; The lake pixel extraction module completes the matching task of all water pixels based on multiple CUDA cores deployed in the GPU, wherein one CUDA core is assigned a water pixel matching task, wherein the matching task is to match the current water pixel with each spectrum in the spectral library, and based on the matching results, determine whether the current water pixel is a lake water system pixel; and based on the determination results of each water pixel by the multiple CUDA cores, generate a result matrix, and transmit the result matrix to the CPU; The binary image construction module is used to construct a binary image according to the result matrix and send the binary image to the GPU; The lake DLG extraction module is used to perform an opening operation on the binary image to generate a lake pixel image; and to extract the lake DLG from the binary image.

9. The hyperspectral image lake DLG extraction device according to claim 8, characterized in that: The preprocessing module is used to preprocess the hyperspectral image data, perform rough extraction on the water body area in the hyperspectral image data, generate a rough extracted water body vector set, and transmit the rough extracted water body vector set to the GPU, including: Performing radiometric calibration, atmospheric correction, and geometric correction on the hyperspectral image data; Determine whether the reflectance of the near-infrared band of each pixel in the hyperspectral image data after geometric correction is less than a set threshold; if so, preliminarily determine that the pixel is a water pixel; Traversing each pixel in the hyperspectral image data, and roughly extracting all water pixels therefrom; A coarsely extracted water body vector set is constructed based on the position information of each water body pixel and the spectral value of each band, wherein the coarsely extracted water body vector set includes multiple vector data, one vector data includes the position information of a water body pixel and the spectral value of each pixel band, and the position information of the water body pixel is represented by the row and column where the water body pixel is located.

10. The hyperspectral image lake DLG extraction device according to claim 9, characterized in that: The lake pixel extraction module completes the task of matching all water pixels based on multiple CUDA cores deployed in the GPU, including: Assign a vector data matching task to each CUDA core, traverse each spectrum in the spectral library, calculate the spectral angle between each spectrum and the spectrum of the water pixel in the current vector data, and obtain m spectral angles corresponding to the current water pixel, where m is the number of spectra in the spectral library; If one or more of the m spectral angles is smaller than the spectral angle threshold, the current water pixel is determined to be a lake water pixel; If all the spectral angles among the m spectral angles are greater than the spectral angle threshold, the current water body pixel is determined to be a non-lake water system pixel.

Images