Construction method and system of low-altitude hyperspectral remote sensing basic large model
By constructing a basic large model of low-altitude hyperspectral remote sensing, the problem of low-altitude spectral remote sensing change detection accuracy and efficiency in the existing technology is solved, and efficient and accurate change detection is achieved, which is suitable for a variety of application scenarios.
Patent Information
- Application Number
- CN202510075616.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-17
- Publication Date
- 2025-05-16
AI Technical Summary
The prior art has low accuracy and efficiency in low-altitude spectral remote sensing variation detection, making it difficult to effectively process complex scenarios and large amounts of hyperspectral data.
A basic model of low-altitude hyperspectral remote sensing was constructed. By obtaining low-altitude hyperspectral remote sensing images at different time points in the same geographical area, detailed registration and correction, preprocessing and spectral vector extraction were carried out. Combined with the deep learning framework, a dual-time phase input layer, a time differential coding and fusion layer, a multi-scale convolutional layer, a spatiotemporal feature fusion layer and an output layer were constructed to achieve change detection.
Through detailed registration and correction and the use of deep learning models, the accuracy and efficiency of change detection are improved, and large amounts of hyperspectral data can be quickly processed, suitable for multi-category classification or regression analysis.
Smart Images
Figure CN120014490A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of remote sensing technology, and in particular relates to a method and system for constructing a low-altitude hyperspectral remote sensing basic large model. Background Art
[0002] With the development of drone technology and sensor technology, low-altitude hyperspectral remote sensing has become an important means of obtaining detailed surface information. Compared with traditional satellite remote sensing, low-altitude hyperspectral remote sensing can provide higher spatial resolution and richer spectral information, which is of great significance to urban planning, environmental protection, agricultural monitoring and other fields. However, due to the low flight altitude of low-altitude platforms, they are easily affected by factors such as terrain undulations and atmospheric conditions, resulting in geometric distortion and radiation errors in the acquired data, which poses a challenge to subsequent change detection tasks.
[0003] Most existing change detection methods rely on manual feature extraction or simple machine learning algorithms, which perform poorly when dealing with complex scenes. In recent years, deep learning technology has achieved great success in image recognition, but its application in low-altitude hyperspectral remote sensing change detection still faces many challenges, such as large data volumes and high computing resource consumption. Therefore, there is an urgent need for an efficient and accurate low-altitude hyperspectral remote sensing basic large model for change detection in spectral remote sensing. Summary of the invention
[0004] In order to solve the above problems in the prior art, that is, to solve the problem of low accuracy and efficiency in the prior art when performing change detection of low-altitude spectral remote sensing, the first aspect of the present invention proposes a method for constructing a low-altitude hyperspectral remote sensing basic large model. The constructed low-altitude hyperspectral remote sensing basic large model is used for change detection. The method comprises:
[0005] S1, obtain low-altitude hyperspectral remote sensing images of the same geographical area at different time points as training images, and construct a training data set;
[0006] S2, registering and correcting the training images in the training data set, and after the registration and correction, preprocessing the training images to obtain preprocessed images;
[0007] S3, extracting the spectral vector corresponding to the preprocessed image as the first spectral vector; calculating the angle between the first spectral vector and the reference spectral vector, and taking the area corresponding to the band where the angle is within the set angle range as the area of interest;
[0008] S4, inputting the region of interest at two time points into a pre-built low-altitude hyperspectral remote sensing basic large model to obtain a change detection result; calculating a loss value based on the change detection result and its corresponding label, and then updating the parameters of the low-altitude hyperspectral remote sensing basic large model;
[0009] S5, looping through S3-S4 until a trained low-altitude hyperspectral remote sensing basic large model is obtained;
[0010] S6, obtaining two low-altitude hyperspectral remote sensing images to be detected for changes, and obtaining a region of interest through the methods of S2 and S3, inputting the region of interest into the trained low-altitude hyperspectral remote sensing basic large model to obtain a change detection result.
[0011] In some preferred embodiments, the training images in the training data set are registered and corrected by:
[0012] Performing image registration on the training images in the training data set; after the image registration, performing geometric correction;
[0013] After geometric correction, the training image is calibrated.
[0014] In some preferred embodiments, the geometric correction method includes correction through RPC model, correction through DEM model assistance, and correction through polynomial fitting.
[0015] In some preferred embodiments, the correction includes atmospheric correction, radiation calibration, sun angle correction, and shadow processing.
[0016] In some preferred embodiments, the sun angle correction is performed by:
[0017] L corr =L×cos(θ s )×TIF
[0018] Among them, L corr represents the corrected reflectivity, i.e. the result of sun angle correction, L represents the original reflectivity, θ s represents the solar zenith angle, and TIF represents the correction factor calculated based on local terrain characteristics.
[0019] In some preferred embodiments, the correction factor is calculated as follows:
[0020] Calculate the slope and aspect of the ground surface, and obtain the solar zenith angle and solar azimuth angle;
[0021] Calculate the sine value of the solar zenith angle, the sine value of the slope of the ground surface, and the cosine value of the difference between the slope of the ground surface and the solar azimuth angle, and use them as a first vector value, a second vector value, and a third vector value;
[0022] Calculating the cosine value of the solar zenith angle and the cosine value of the slope of the ground surface as the fourth vector value and the fifth vector value, respectively;
[0023] respectively calculating the product of the first vector value, the second vector value, the third vector value, the fourth vector value, and the fifth vector value, and adding the two product results as the solar incidence angle;
[0024] The ratio of the solar incident angle to the fourth vector value is used as a correction factor.
[0025] In some preferred embodiments, the preprocessing includes noise removal, outlier processing, standardization processing, normalization processing, and data enhancement processing.
[0026] In some preferred embodiments, the angle between the first spectral vector and the reference spectral vector is calculated by: calculating the angle between the first spectral vector and the reference spectral vector by a spectral angle mapping method.
[0027] In some preferred embodiments, the low-altitude hyperspectral remote sensing basic large model includes a dual-phase input layer, a time difference encoding and fusion layer, a multi-scale convolution layer, a spatiotemporal feature fusion layer, and an output layer;
[0028] The dual-temporal input layer is used to input the regions of interest of the low-altitude hyperspectral remote sensing images at two time points in the same geographical area, respectively as the first region of interest and the second region of interest; convolution and pooling are performed on the first region of interest to obtain a first feature map; convolution and pooling are performed on the second region of interest to obtain a second feature map;
[0029] The temporal difference coding and fusion layer is used to calculate the difference between the first feature map and the second feature map to obtain a differential feature map;
[0030] The multi-scale convolution layer is used to perform multi-scale convolution on the differential feature map and fuse feature maps of different scales to obtain a fused feature map;
[0031] The spatiotemporal feature fusion layer is used to splice the differential feature map with the fused feature map to obtain a spliced feature map;
[0032] The output layer is used to process the spliced feature map through a global average pooling layer and a fully connected layer in sequence to obtain a change detection result.
[0033] In a second aspect of the present invention, a system for constructing a low-altitude hyperspectral remote sensing basic large model is proposed. The constructed low-altitude hyperspectral remote sensing basic large model is used for change detection. The system comprises:
[0034] A data acquisition module is configured to acquire low-altitude hyperspectral remote sensing images at different time points in the same geographical area as training images and construct a training data set;
[0035] A registration and correction module, configured to perform registration and correction on the training images in the training data set, and after the registration and correction, preprocess the training images to obtain preprocessed images;
[0036] A region extraction module is configured to extract a spectral vector corresponding to the preprocessed image as a first spectral vector; calculate an angle between the first spectral vector and a reference spectral vector, and take a region corresponding to a band in which the angle is within a set angle range as a region of interest;
[0037] The parameter updating module is configured to input the region of interest at two time points into a pre-built low-altitude hyperspectral remote sensing basic large model to obtain a change detection result; based on the change detection result and its corresponding label, calculate the loss value, and then update the parameters of the low-altitude hyperspectral remote sensing basic large model;
[0038] A loop module is configured to loop through the region extraction module and the parameter updating module until a trained low-altitude hyperspectral remote sensing basic large model is obtained;
[0039] The change detection module is configured to obtain two low-altitude hyperspectral remote sensing images to be detected for changes, and obtain the region of interest through the methods of the registration and correction module and the region extraction module, and input the region of interest into the trained low-altitude hyperspectral remote sensing basic large model to obtain the change detection result.
[0040] Beneficial effects of the present invention:
[0041] 1) The present invention reduces the error caused by external factors and improves the accuracy of change detection by performing detailed registration and correction on the image;
[0042] 2) The present invention introduces a variety of correction measures, such as atmospheric correction, solar angle correction, etc., so that the model can adapt to different environmental conditions;
[0043] 3) The present invention uses a basic large model constructed by a deep learning framework, which can process a large amount of hyperspectral data in a short time and quickly respond to user needs. It can not only be used for transformation detection, but also can be extended to multi-category classification or regression analysis, and is suitable for more application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings.
[0045] Figure 1 The present invention is a flowchart of a method for constructing a low-altitude hyperspectral remote sensing basic large model according to an embodiment of the present invention. DETAILED DESCRIPTION
[0046] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiment is a part of the embodiment 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 creative work are within the scope of protection of the present invention.
[0047] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.
[0048] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application may be combined with each other.
[0049] A method for constructing a low-altitude hyperspectral remote sensing basic large model according to a first embodiment of the present invention is used to extract features of ground objects and then perform category recognition, such as Figure 1 As shown, the following steps are included:
[0050] S1, obtain low-altitude hyperspectral remote sensing images of the same geographical area at different time points as training images, and construct a training data set;
[0051] S2, registering and correcting the training images in the training data set, and after the registration and correction, preprocessing the training images to obtain preprocessed images;
[0052] S3, extracting the spectral vector corresponding to the preprocessed image as the first spectral vector; calculating the angle between the first spectral vector and the reference spectral vector, and taking the area corresponding to the band where the angle is within the set angle range as the area of interest;
[0053] S4, inputting the region of interest at two time points into a pre-built low-altitude hyperspectral remote sensing basic large model to obtain a change detection result; calculating a loss value based on the change detection result and its corresponding label, and then updating the parameters of the low-altitude hyperspectral remote sensing basic large model;
[0054] S5, looping through S3-S4 until a trained low-altitude hyperspectral remote sensing basic large model is obtained;
[0055] S6, obtaining two low-altitude hyperspectral remote sensing images to be detected for changes, and obtaining a region of interest through the methods of S2 and S3, inputting the region of interest into the trained low-altitude hyperspectral remote sensing basic large model to obtain a change detection result.
[0056] In order to more clearly illustrate the method for constructing a low-altitude hyperspectral remote sensing basic large model of the present invention, the steps in an embodiment of the method of the present invention are described in detail below with reference to the accompanying drawings.
[0057] S1, obtain low-altitude hyperspectral remote sensing images of the same geographical area at different time points as training images, and construct a training data set;
[0058] In this embodiment, a drone equipped with a hyperspectral camera is used to perform multiple flight photography of the same geographical area within a selected time interval. The collected low-altitude hyperspectral remote sensing images are classified and sorted according to time and geographical location to ensure that each sample contains two images of different phases. According to the actual situation (such as ground measured data, historical archives, etc.), the changes (such as new buildings, changes in vegetation coverage, etc.) are annotated for each pair of images.
[0059] S2, registering and correcting the training images in the training data set, and after the registration and correction, preprocessing the training images to obtain preprocessed images;
[0060] In this embodiment, the training images in the training data set are registered and corrected to eliminate geometric deformation caused by sensor posture, terrain undulations, etc., and then preprocessed to improve data quality. The details are as follows:
[0061] It is preferred to use feature matching and other technologies (common algorithms include SIFT (Scale Invariant Feature Transform), SURF (Speeded Robust Features) and ORB (Oriented FAST and Rotated BRIEF)) to perform image registration on the training images in the training data set, so that the images of the two periods have a pixel-level correspondence in the same coordinate system, thereby ensuring that the images from different phases are accurately aligned in space;
[0062] After the image registration, geometric correction is performed; the methods of geometric correction include correction through RPC model, correction assisted by DEM model (digital elevation model), and correction through polynomial fitting;
[0063] After geometric correction, the training image is corrected; the correction includes atmospheric correction, radiation calibration, sun angle correction, and shadow processing.
[0064] Atmospheric correction: correct the effects caused by atmospheric scattering and absorption, and restore the true reflectivity of the ground objects; commonly used methods include the MODTRAN model;
[0065] Radiometric calibration: converting the original DN value into a physical quantity (such as reflectivity) to ensure that data at different time points are comparable;
[0066] Sun Angle Correction: Adjusts the brightness changes caused by differences in the sun's altitude angle to ensure consistency between images; specifically:
[0067] L corr =L×cos(θ s )×TIF
[0068] Among them, L corr represents the corrected reflectivity, i.e. the result of sun angle correction, L represents the original reflectivity, θ s represents the solar zenith angle, TIF represents the correction factor calculated based on local terrain characteristics;
[0069] The correction factor is calculated as follows:
[0070] Calculate the slope and aspect of the ground surface, and obtain the solar zenith angle and solar azimuth angle;
[0071] The slope S represents the inclination of the ground surface and is calculated using the following formula:
[0072]
[0073] in, They represent the rate of change of height in the x-axis and y-axis directions respectively, and can be calculated from the DEM by the difference method.
[0074] Slope aspect A represents the direction of the ground surface, that is, the angle of the steepest descent direction, and is calculated using the following formula:
[0075]
[0076] Since the result range of the inverse tangent function is -π to π, it needs to be adjusted to the range of 0 to 2π according to the specific situation, and the case where the denominator is zero (that is, the direction perpendicular to the y-axis) must be handled.
[0077] Calculate the sine value of the solar zenith angle, the sine value of the slope of the ground surface, and the cosine value of the difference between the slope of the ground surface and the solar azimuth angle, and use them as the first vector value, the second vector value, and the third vector value;
[0078] Calculate the cosine value of the solar zenith angle and the cosine value of the slope of the ground surface as the fourth vector value and the fifth vector value respectively;
[0079] Calculate the product of the first vector value, the second vector value, the third vector value, the fourth vector value, and the fifth vector value respectively, and add the two product results as the solar incidence angle;
[0080] The ratio of the solar incident angle to the fourth vector value is used as a correction factor.
[0081] The correction factor reflects the change in light intensity relative to flat ground after taking into account the effects of terrain. For slopes facing the sun, TIF>1; for slopes facing away from the sun, TIF<1.
[0082] Shadow processing: Identify and correct shadow areas, which can seriously affect the accuracy of spectral information;
[0083] Preprocessing includes noise removal, outlier processing, standardization processing, normalization processing, and data enhancement processing, all of which are existing technologies and will not be described here one by one.
[0084] S3, extracting the spectral vector corresponding to the preprocessed image as the first spectral vector; calculating the angle between the first spectral vector and the reference spectral vector, and taking the area corresponding to the band where the angle is within the set angle range as the area of interest;
[0085] In this embodiment, the low-altitude hyperspectral remote sensing image is usually stored in the form of a multidimensional array, which contains three dimensions: row (height), column (width) and number of bands. Each pair of row and column coordinates corresponds to a pixel point, and the band represents a different electromagnetic band. The reflectance value of each pixel point in different bands constitutes the spectral vector of the pixel.
[0086] Preferably, the angle between the spectral vector corresponding to the preprocessed image and the reference spectral vector (i.e., selecting an ideal spectral curve representing the target object) is calculated by the spectral angle mapping method, and the band or band combination with the smallest angle is selected to represent the spectral characteristics of the target object, thereby obtaining the area of interest in the low-altitude hyperspectral remote sensing image at two time points in the same geographical area.
[0087] S4, inputting the region of interest at two time points into a pre-built low-altitude hyperspectral remote sensing basic large model to obtain a change detection result; calculating a loss value based on the change detection result and its corresponding label, and then updating the parameters of the low-altitude hyperspectral remote sensing basic large model;
[0088] In this embodiment, the low-altitude hyperspectral remote sensing basic large model includes a dual-phase input layer, a time difference encoding and fusion layer, a multi-scale convolution layer, a spatiotemporal feature fusion layer, and an output layer;
[0089] The dual-temporal input layer is used to input the regions of interest of the low-altitude hyperspectral remote sensing images at two time points in the same geographical area, respectively as the first region of interest and the second region of interest; the first region of interest is subjected to a set number of convolution and pooling processes to obtain a first feature map; the second region of interest is subjected to a set number of convolution and pooling processes to obtain a second feature map;
[0090] The temporal differential encoding and fusion layer is used to calculate the difference between the first feature map and the second feature map to obtain a differential feature map, which represents the difference between the two temporal feature maps and captures the change information; the difference can be obtained by element-by-element subtraction.
[0091] The multi-scale convolution layer is used to perform multi-scale convolution on the differential feature map (such as convolution kernels of different sizes or dilated convolution) and fuse the feature maps of different scales to obtain a fused feature map;
[0092] The spatiotemporal feature fusion layer is used to splice the differential feature map and the fused feature map to obtain a spliced feature map;
[0093] The output layer is used to process the concatenated feature map in turn through a global average pooling layer and a fully connected layer to obtain a change detection result. In other embodiments, an appropriate activation function (such as sigmoid for a binary classification problem) may also be selected according to the task type.
[0094] According to the output result of the output layer, combined with the pre-calibrated labels, the loss value is calculated, and the weight of the model is updated through the back propagation algorithm.
[0095] S5, looping through S3-S4 until a trained low-altitude hyperspectral remote sensing basic large model is obtained;
[0096] In this embodiment, the model is trained in a loop until the model converges.
[0097] S6, obtaining two low-altitude hyperspectral remote sensing images to be detected for changes, and obtaining a region of interest through the methods of S2 and S3, inputting the region of interest into the trained low-altitude hyperspectral remote sensing basic large model to obtain a change detection result.
[0098] In this embodiment, after the training is completed, two low-altitude hyperspectral remote sensing images to be detected are obtained (i.e., low-altitude hyperspectral remote sensing images at two time points in the same geographical area), and the images are sequentially registered and corrected, preprocessed, and the region of interest is extracted, and then the region of interest is input into the trained low-altitude hyperspectral remote sensing basic large model to obtain the change detection result. For example, the urban planning department wants to understand the changes in the green space around a newly built community. The low-altitude hyperspectral remote sensing images collected by the drone (in the spring of 2023, the most prosperous period of greening, and in the autumn of 2024, the vegetation coverage is reduced or changed significantly) are sequentially registered and corrected, preprocessed, and the region of interest is extracted, and then the region of interest (i.e., the greening area) is input into the trained low-altitude hyperspectral remote sensing basic large model to obtain the final change detection result, including the specific location and degree of greening changes, such as: 1) Newly planted trees are found in some areas; 2) Some lawns show obvious signs of degradation; 3) Some original vegetation is replaced by other types of plants.
[0099] In summary, on the one hand, the present invention reduces the errors caused by external factors and improves the accuracy of change detection by performing detailed registration and correction of images. On the other hand, the basic large model constructed by the deep learning framework can process a large amount of hyperspectral data in a short time and quickly respond to user needs.
[0100] A system for constructing a low-altitude hyperspectral remote sensing basic large model according to a second embodiment of the present invention, wherein the constructed low-altitude hyperspectral remote sensing basic large model is used for change detection, and the system comprises:
[0101] A data acquisition module is configured to acquire low-altitude hyperspectral remote sensing images at different time points in the same geographical area as training images and construct a training data set;
[0102] A registration and correction module, configured to perform registration and correction on the training images in the training data set, and after the registration and correction, preprocess the training images to obtain preprocessed images;
[0103] A region extraction module is configured to extract a spectral vector corresponding to the preprocessed image as a first spectral vector; calculate an angle between the first spectral vector and a reference spectral vector, and take a region corresponding to a band in which the angle is within a set angle range as a region of interest;
[0104] The parameter updating module is configured to input the region of interest at two time points into a pre-built low-altitude hyperspectral remote sensing basic large model to obtain a change detection result; based on the change detection result and its corresponding label, calculate the loss value, and then update the parameters of the low-altitude hyperspectral remote sensing basic large model;
[0105] A loop module is configured to loop through the region extraction module and the parameter updating module until a trained low-altitude hyperspectral remote sensing basic large model is obtained;
[0106] The change detection module is configured to obtain two low-altitude hyperspectral remote sensing images to be detected for changes, and obtain the region of interest through the methods of the registration and correction module and the region extraction module, and input the region of interest into the trained low-altitude hyperspectral remote sensing basic large model to obtain the change detection result.
[0107] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.
[0108] It should be noted that the system for constructing a low-altitude hyperspectral remote sensing basic large model provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined. For example, the modules in the above embodiment can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not regarded as improper limitations of the present invention.
[0109] An electronic device according to the third embodiment of the present invention comprises: at least one processor; and a memory connected in communication with at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned method for constructing a low-altitude hyperspectral remote sensing basic large model.
[0110] A computer-readable storage medium according to a fourth embodiment of the present invention stores computer instructions, and the computer instructions are used to be executed by the computer to implement the above-mentioned method for constructing a low-altitude hyperspectral remote sensing basic large model.
[0111] Technicians in the technical field can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the electronic device and computer-readable storage medium described above can refer to the corresponding process in the aforementioned method example and will not be repeated here.
[0112] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.
[0113] The terms "first", "second", etc. are used to distinguish similar objects rather than to describe or indicate a particular order or sequence.
[0114] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that includes a list of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article, or apparatus / device.
[0115] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.
Claims
1. A method for constructing a low-altitude hyperspectral remote sensing basic large model, wherein the constructed low-altitude hyperspectral remote sensing basic large model is used for change detection, characterized in that: The method comprises the following steps: S1, obtain low-altitude hyperspectral remote sensing images of the same geographical area at different time points as training images, and construct a training data set; S2, registering and correcting the training images in the training data set, and after the registration and correction, preprocessing the training images to obtain preprocessed images; S3, extracting the spectral vector corresponding to the preprocessed image as the first spectral vector; calculating the angle between the first spectral vector and the reference spectral vector, and taking the area corresponding to the band where the angle is within the set angle range as the area of interest; S4, inputting the region of interest at two time points into a pre-built low-altitude hyperspectral remote sensing basic large model to obtain a change detection result; calculating a loss value based on the change detection result and its corresponding label, and then updating the parameters of the low-altitude hyperspectral remote sensing basic large model; S5, looping through S3-S4 until a trained low-altitude hyperspectral remote sensing basic large model is obtained; S6, obtaining two low-altitude hyperspectral remote sensing images to be detected for changes, and obtaining a region of interest through the methods of S2 and S3, inputting the region of interest into the trained low-altitude hyperspectral remote sensing basic large model to obtain a change detection result.
2. The method for constructing a low-altitude hyperspectral remote sensing basic large model according to claim 1, characterized in that: The training images in the training data set are registered and corrected by: Performing image registration on the training images in the training data set; after the image registration, performing geometric correction; After geometric correction, the training image is calibrated.
3. The method for constructing a low-altitude hyperspectral remote sensing basic large model according to claim 2 is characterized in that: The geometric correction method includes correction through RPC model, correction through DEM model assistance, and correction through polynomial fitting.
4. The method for constructing a low-altitude hyperspectral remote sensing basic large model according to claim 2, characterized in that: The correction includes atmospheric correction, radiation calibration, sun angle correction, and shadow processing.
5. The method for constructing a low-altitude hyperspectral remote sensing basic large model according to claim 4 is characterized in that: The sun angle correction method is as follows: L corr =L×cos(θ s )×TIF Among them, L corr represents the corrected reflectivity, i.e. the result of sun angle correction, L represents the original reflectivity, θ s represents the solar zenith angle, and TIF represents the correction factor calculated based on local terrain characteristics.
6. The method for constructing a low-altitude hyperspectral remote sensing basic large model according to claim 5, characterized in that: The correction factor is calculated as follows: Calculate the slope and aspect of the ground surface, and obtain the solar zenith angle and solar azimuth angle; Calculate the sine value of the solar zenith angle, the sine value of the slope of the ground surface, and the cosine value of the difference between the slope of the ground surface and the solar azimuth angle, and use them as a first vector value, a second vector value, and a third vector value; Calculating the cosine value of the solar zenith angle and the cosine value of the slope of the ground surface as the fourth vector value and the fifth vector value, respectively; respectively calculating the product of the first vector value, the second vector value, the third vector value, the fourth vector value, and the fifth vector value, and adding the two product results as the solar incidence angle; The ratio of the solar incident angle to the fourth vector value is used as a correction factor.
7. The method for constructing a low-altitude hyperspectral remote sensing basic large model according to claim 1, characterized in that: The preprocessing includes noise removal, outlier processing, standardization processing, normalization processing, and data enhancement processing.
8. The method for constructing a low-altitude hyperspectral remote sensing basic large model according to claim 1, characterized in that: The angle between the first spectrum vector and the reference spectrum vector is calculated by: calculating the angle between the first spectrum vector and the reference spectrum vector by a spectrum angle mapping method.
9. The method for constructing a low-altitude hyperspectral remote sensing basic large model according to claim 1, characterized in that: The low-altitude hyperspectral remote sensing basic model includes a dual-phase input layer, a time difference encoding and fusion layer, a multi-scale convolution layer, a spatiotemporal feature fusion layer, and an output layer; The dual-temporal input layer is used to input the regions of interest of the low-altitude hyperspectral remote sensing images at two time points in the same geographical area, respectively as the first region of interest and the second region of interest; Performing convolution and pooling processing on the first region of interest to obtain a first feature map; Performing convolution and pooling processing on the second region of interest to obtain a second feature map; The temporal difference coding and fusion layer is used to calculate the difference between the first feature map and the second feature map to obtain a differential feature map; The multi-scale convolution layer is used to perform multi-scale convolution on the differential feature map and fuse feature maps of different scales to obtain a fused feature map; The spatiotemporal feature fusion layer is used to splice the differential feature map with the fused feature map to obtain a spliced feature map; The output layer is used to process the spliced feature map through a global average pooling layer and a fully connected layer in sequence to obtain a change detection result.
10. A system for constructing a low-altitude hyperspectral remote sensing basic large model, wherein the constructed low-altitude hyperspectral remote sensing basic large model is used for change detection, characterized in that: The system includes: A data acquisition module is configured to acquire low-altitude hyperspectral remote sensing images at different time points in the same geographical area as training images and construct a training data set; A registration and correction module, configured to perform registration and correction on the training images in the training data set, and after the registration and correction, preprocess the training images to obtain preprocessed images; A region extraction module is configured to extract a spectral vector corresponding to the preprocessed image as a first spectral vector; calculate an angle between the first spectral vector and a reference spectral vector, and take a region corresponding to a band in which the angle is within a set angle range as a region of interest; The parameter updating module is configured to input the region of interest at two time points into a pre-built low-altitude hyperspectral remote sensing basic large model to obtain a change detection result; based on the change detection result and its corresponding label, calculate the loss value, and then update the parameters of the low-altitude hyperspectral remote sensing basic large model; A loop module is configured to loop through the region extraction module and the parameter updating module until a trained low-altitude hyperspectral remote sensing basic large model is obtained; The change detection module is configured to obtain two low-altitude hyperspectral remote sensing images to be detected for changes, and obtain the region of interest through the methods of the registration and correction module and the region extraction module, and input the region of interest into the trained low-altitude hyperspectral remote sensing basic large model to obtain the change detection result.