Regional soil component detection method based on remote sensing image

Abstract

The invention provides a regional soil component detection method based on a remote sensing image, and belongs to the technical field of soil detection. Aiming at the problems that a traditional detection method depends on a single optical remote sensing image, is interfered by weather and vegetation, and is low in detection precision, low in efficiency, unguaranteed in data credibility and the like, radar and thermal infrared remote sensing data are introduced to be fused with an optical image, features are extracted through principal component analysis, a generative adversarial network is adopted to carry out hyperspectral image super-resolution reconstruction, and a high-resolution hyperspectral image is obtained. A genetic algorithm is utilized to realize adaptive wave band selection, a real-time atmospheric correction model is constructed, a recurrent neural network and a spatial convolutional neural network are utilized to mine spatio-temporal context information, a quantum computing acceleration algorithm is introduced, block chain management data is utilized, and unmanned aerial vehicle cooperative operation is combined. According to the method, the detection precision, efficiency and data credibility are remarkably improved, and the regional soil component detection process is effectively optimized.

Description

Technical Field

[0001] The present invention belongs to the technical field of soil detection, and particularly relates to a method for detecting regional soil components based on remote sensing images. Background Technique

[0002] At present, in many fields such as agriculture and geological exploration, methods for detecting regional soil components based on remote sensing images are widely used. The traditional method mainly relies on single optical remote sensing images and analyzes the combination of specific bands to infer soil components. However, this method has many drawbacks. On the one hand, optical remote sensing is easily affected by weather conditions such as cloud cover, precipitation interference, and vegetation coverage, resulting in a decline in the quality of the acquired images and seriously affecting the accuracy of soil component inversion. On the other hand, soil components are extremely complex, and it is difficult to comprehensively and accurately reflect the differences in soil characteristics in different regions only by relying on fixed band combinations, resulting in a significant reduction in the accuracy and reliability of the detection results. At the same time, when the existing technology processes hyperspectral images, due to its low spatial resolution, it is difficult to capture the changes in fine soil components; moreover, the computational efficiency is low, and when facing large-area detection tasks, it takes a long time and cannot meet the real-time requirements in practical applications. In addition, in the data management link, there is a lack of effective means to ensure the credibility and traceability of the detection data.

[0003] In order to solve the problems of low detection accuracy, many interference factors, low data processing efficiency, and lack of guarantee of data credibility existing in the above-mentioned existing technology, the present invention proposes a brand-new method for detecting regional soil components based on remote sensing images. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for detecting regional soil components based on remote sensing images, aiming to solve the problems of low detection accuracy, many interference factors, low data processing efficiency, and lack of guarantee of data credibility existing in the existing technology.

[0005] To achieve the above purpose, the present invention provides the following technical solutions:

[0006] A method for detecting regional soil components based on remote sensing images includes the following steps:

[0007] S1. Obtain multi-modal remote sensing image data of the target area, where the multi-modal remote sensing image data includes optical remote sensing image data, radar remote sensing data, and thermal infrared remote sensing data; at the same time, obtain the atmospheric parameters monitored in real time by a ground meteorological station and the time information when the satellite passes; collect a small amount of soil samples in the target area to obtain the measured data of soil components;

[0008] S2. Perform multi-modal data fusion on the multi-modal remote sensing image data, extract features from each modal data through the principal component analysis algorithm, splice the feature vectors of different modal data, and construct a comprehensive feature vector set. The principal component analysis algorithm includes the following steps:

[0009] S21. Let the multi-modal data matrix be, with a dimension of,

[0010] S22. First, calculate the mean vector of the data, then calculate the covariance matrix, and then solve the eigenvalues and eigenvectors () of the covariance matrix. The eigenvalues of the covariance matrix are obtained by solving the characteristic equation, and the corresponding eigenvectors are obtained by solving the linear equations.

[0011] S23. Select the top eigenvectors with larger eigenvalues, splice the corresponding eigenvectors of different modal data, and construct a comprehensive feature vector set.

[0012] In the formula: is the number of samples, and is the number of features.

[0013] S3. For the obtained low-resolution hyperspectral image data, use a generative adversarial network for hyperspectral image super-resolution reconstruction. The generator network learns the mapping relationship between the low-resolution hyperspectral image and the high-resolution image to generate a high-resolution image, and the discriminator network compares and judges the generated image with the real high-resolution image to optimize the generator.

[0014] S4. Based on the prior knowledge of the soil in the target area or a small amount of soil sample data collected, use the genetic algorithm for adaptive band selection, construct a fitness function, randomly generate an initial subset of wavebands as a population, and search for the subset of wavebands that makes the fitness function value optimal through genetic operations such as selection, crossover, and mutation. Let the sensitivity score function of the band to the soil composition be, and the correlation score function between bands be, then the fitness function is;

[0015] In the formula: m is the number of bands, b represents the subset of wavebands, and is the correlation score function between bands, which is used to evaluate the correlation between the i-th band and the j-th band.

[0016] S5. Establish a real-time atmospheric correction model. According to the obtained atmospheric parameters and satellite transit time information, use the atmospheric radiation transfer model to calculate the absorption and scattering coefficients of the atmosphere on the remote sensing signal, and perform real-time correction on the remote sensing image.

[0017] S6. Use a recurrent neural network combined with a spatial convolutional neural network to mine spatio-temporal context information. Use the recurrent neural network to process remote sensing images of time series to capture the changing trend of soil components over time, and use the spatial convolutional neural network to perform convolutional operations on remote sensing images in the spatial neighborhood to mine the correlation of soil information in the spatial neighborhood. After fusing the outputs of the recurrent neural network (RNN) and the spatial convolutional neural network, input them into the classifier;

[0018] S7. For complex algorithms in soil component detection, introduce quantum computing technology and use the quantum annealing algorithm to optimize the parameters of the soil component inversion model;

[0019] S8. Use blockchain technology to manage the detection data trustworthily, and record information such as the original remote sensing image data, algorithm parameters of each processing module, intermediate processing results, and the final soil component detection conclusion on the blockchain;

[0020] S9. Deploy intelligent drones equipped with high-resolution cameras and small soil detection devices. According to the area of interest preliminarily defined by satellite remote sensing, the drones use path planning algorithms to autonomously plan flight paths to collect low-altitude high-resolution remote sensing images, and land in key areas to collect soil samples for on-site detection, and fuse and analyze the data obtained by the drones with satellite remote sensing data.

[0021] As a preferred solution of the present invention, the preprocessing of the multi-modal remote sensing image data includes radiometric calibration and geometric correction of optical remote sensing images, radar remote sensing data, and thermal infrared remote sensing data respectively.

[0022] As a preferred solution of the present invention, when constructing the fitness function, comprehensively consider the sensitivity of the band to soil components and the correlation between bands. The fitness function is the sensitivity score of the band to soil components minus the correlation score between bands.

[0023] As a preferred solution of the present invention, the selection operation adopts the roulette wheel selection method, and the probability P i of an individual being selected is i proportional to its fitness value F The calculation formula is:

[0024] where N is the number of population individuals.

[0025] As a preferred solution of the present invention, the recurrent neural network adopts a long short-term memory network structure, and the output feature vectors of the recurrent neural network and the spatial convolutional neural network are fused by using the splicing or weighted summation method.

[0026] As a preferred embodiment of the present invention, assuming that the output feature vector of the recurrent neural network is r and the output feature vector of the spatial convolutional neural network is s, the weighted sum fusion formula is:

[0027] f = w1r + w2s;

[0028] In the formula: w1 and w2 are weights, and w1 + w2 = 1.

[0029] As a preferred embodiment of the present invention, the blockchain platform selects a proof-of-work or proof-of-stake consensus algorithm, encrypts the detection data, and packages it into blocks according to the requirements of the blockchain format and adds them to the blockchain network.

[0030] As a preferred embodiment of the present invention, the UAV path planning algorithm takes into account factors such as terrain, obstacles, and image overlap.

[0031] Compared with the prior art, the beneficial effects of the present invention are:

[0032] 1. In this solution, through multi-modal data fusion, the ability to identify soil components is improved, the interference of factors such as weather and vegetation cover is reduced, and the detection results are more accurate and reliable. The super-resolution reconstruction of hyperspectral images effectively improves the image spatial resolution, can capture the subtle changes in soil components, and enhances the detection accuracy.

[0033] 2. In this solution, by adaptively selecting bands to dynamically adjust the detection bands according to the soil characteristics of different regions, the detection efficiency and pertinence are improved. Real-time atmospheric correction accurately eliminates atmospheric interference, provides high-quality image data for soil component inversion, and the mining of spatio-temporal context information makes full use of the information in the time and space dimensions to more accurately judge the dynamic changes of soil components.

[0034] 3. In this solution, the calculation time is significantly shortened through quantum computing acceleration, meeting the rapid detection requirements for large areas. Blockchain data management ensures the credibility and traceability of detection data.

[0035] 4. In this solution, the cooperation between UAVs and satellite remote sensing optimizes the detection process, improves the comprehensiveness and accuracy of detection. Compared with the prior art, the present invention has distinct distinguishing technical features and creativity, and can effectively improve the regional soil component detection level based on remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0037] Figure 1Flow chart of a method for detecting regional soil components based on remote sensing images according to the present invention. Detailed implementation manners

[0038] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0039] Embodiment 1

[0040] Please refer to Figure 1 , the present invention provides the following technical solutions:

[0041] A method for detecting regional soil components based on remote sensing images, characterized by comprising the following steps:

[0042] S1. Obtain multi-modal remote sensing image data of the target area, where the multi-modal remote sensing image data includes optical remote sensing image data, radar remote sensing data, and thermal infrared remote sensing data; at the same time, obtain the atmospheric parameters monitored by the ground weather station in real time and the time information when the satellite passes; collect a small amount of soil samples in the target area to obtain the measured data of soil components;

[0043] S2. Perform multi-modal data fusion on the multi-modal remote sensing image data, extract features of each modal data through the principal component analysis (PCA) algorithm, splice the feature vectors of different modal data, and construct a comprehensive feature vector set; where the principal component analysis (PCA) algorithm includes the following steps:

[0044] S21. Let the multi-modal data matrix be X, and its dimension be n×p,

[0045] S22. First, calculate the mean vector of the data Then calculate the covariance matrix Next, solve the eigenvalues λ i and eigenvectors v i (i = 1, 2,..., p) of the covariance matrix, and obtain the eigenvalues λ of the covariance matrix C by solving the characteristic equation |C - λI| = 0 i , and obtain the corresponding eigenvectors v i I)v i = 0 by solving the linear equation system (C - λ i ,

[0046] S23. Select the first k eigenvectors with larger eigenvalues, splice the corresponding eigenvectors of different modal data, and construct a comprehensive feature vector set;

[0047] Where: n is the number of samples, and p is the number of features.

[0048] S3. For the acquired low-resolution hyperspectral image data, use a generative adversarial network (GAN) for hyperspectral image super-resolution reconstruction. The generator network learns the mapping relationship between the low-resolution hyperspectral image and the high-resolution image to generate a high-resolution image, and the discriminator network compares and judges the generated image with the real high-resolution image to optimize the generator.

[0049] S4. Based on the prior knowledge of the soil in the target area or the small amount of soil sample data collected, use a genetic algorithm for adaptive band selection. Construct a fitness function, randomly generate an initial subset of wavebands as a population, and search for the subset of wavebands that optimizes the fitness function value through genetic operations such as selection, crossover, and mutation. Among them, let the sensitivity score function of the band to the soil component be S(b), and the correlation score function between bands be R(b1, b2), then the fitness function is

[0050]

[0051] Where: m is the number of wavebands, b represents the subset of wavebands, and R(b1, b2) is the correlation score function between bands, which is used to evaluate the correlation between the i-th band and the j-th band.

[0052] S5. Establish a real-time atmospheric correction model. According to the acquired atmospheric parameters and satellite transit time information, use an atmospheric radiative transfer model to calculate the absorption and scattering coefficients of the atmosphere to the remote sensing signal, and perform real-time correction on the remote sensing image.

[0053] S6. Use a recurrent neural network (RNN) combined with a spatial convolutional neural network (SCNN) to mine spatio-temporal context information. Use the recurrent neural network RNN to process the time-series remote sensing images to capture the changing trend of soil components over time, and the spatial convolutional neural network SCNN performs convolutional operations on the remote sensing images in the spatial neighborhood to mine the correlation of soil information in the spatial neighborhood. The outputs of the recurrent neural network RNN and the spatial convolutional neural network SCNN are fused and then input into the classifier.

[0054] S7. For the complex algorithms in soil component detection, introduce quantum computing technology, and use a quantum annealing algorithm to optimize the parameters of the soil component inversion model.

[0055] S8. Use blockchain technology to manage the detection data in a trustworthy manner, and record information such as the original remote sensing image data, the algorithm parameters of each processing module, the intermediate processing results, and the final soil component detection conclusion on the blockchain.

[0056] S9. Deploy intelligent drones equipped with high-resolution cameras and small soil detection devices. The drones autonomously plan flight paths according to the regions of interest preliminarily delineated by satellite remote sensing for low-altitude high-resolution remote sensing image acquisition, and land in key areas to collect soil samples for on-site detection. Then, the data obtained by the drones is fused and analyzed with the satellite remote sensing data.

[0057] In a specific embodiment of the present invention, the method introduces radar remote sensing data and thermal infrared remote sensing data for fusion with traditional optical remote sensing images. The principal component analysis (PCA) algorithm is used to extract features from multi-modal data, and the feature vectors of different modal data are spliced to construct a comprehensive feature vector set. For example, the reflectivity features obtained from optical images, the soil roughness and dielectric constant features obtained from radar remote sensing, and the temperature features obtained from thermal infrared remote sensing are fused after PCA processing, enabling the detection model to obtain more comprehensive soil information and enhancing the ability to identify soil components.

[0058] Use a generative adversarial network (GAN) for hyperspectral image super-resolution reconstruction. The generator network learns the mapping relationship between low-resolution hyperspectral images and high-resolution images to generate high-resolution images. The discriminator network compares and judges the generated images with real high-resolution images to continuously optimize the generator. During training, a large number of paired low-resolution and high-resolution hyperspectral images are used as samples, and after multiple rounds of iterative training, the generator can effectively improve the spatial resolution of hyperspectral images, thereby accurately identifying subtle changes in soil components.

[0059] Use a genetic algorithm to achieve adaptive band selection. First, based on the prior knowledge of the soil in the target area or a small amount of soil sample data collected on-site, a fitness function is constructed. The fitness function comprehensively considers factors such as the sensitivity of bands to soil components and the correlation between bands. Then, an initial subset of wavebands is randomly generated as a population, and through genetic operations such as selection, crossover, and mutation, the subset of wavebands that optimizes the fitness function value is continuously searched. This optimal subset of wavebands is the sensitive band combination for soil component detection in this area, reducing data redundancy while improving detection accuracy and efficiency.

[0060] Establish a real-time atmospheric correction model. This model combines the atmospheric parameters monitored in real-time by a ground meteorological station, including temperature, humidity, aerosol concentration, etc., and the time information during satellite transit. By constructing an atmospheric radiation transfer model, the absorption and scattering coefficients of the atmosphere for remote sensing signals are calculated based on the real-time atmospheric parameters to perform real-time correction on remote sensing images. Compared with traditional correction methods based on empirical formulas, it can more accurately eliminate atmospheric interference, improve image quality, and provide a reliable data basis for accurately retrieving soil components.

[0061] Use a recurrent neural network (RNN) combined with a spatial convolutional neural network (SCNN) to mine spatio-temporal context information; the RNN processes remote sensing images of time series to capture the changing trend of soil components over time, such as the content changes of certain components in the soil in different seasons; the SCNN performs convolutional operations on remote sensing images in the spatial neighborhood to mine the correlation of soil information in the spatial neighborhood, such as the gradual change relationship of soil types in adjacent areas; fuse the outputs of the RNN and SCNN and input them into a classifier, which can more accurately judge the dynamic changes of soil components and overcome the limitations of traditional methods that only detect based on single-image or single-time-point data;

[0062] For complex algorithms in soil component detection, such as feature extraction of high-dimensional data and optimization of model parameters, introduce quantum computing technology; use the quantum annealing algorithm to optimize the parameters of the soil component inversion model; the parallel computing characteristics of qubits enable the algorithm to search for the global optimal solution in a shorter time, greatly shortening the calculation time and achieving fast and accurate soil component detection to meet the needs of rapid detection in large areas;

[0063] Use blockchain technology to manage detection data in a trustworthy manner; during the detection process, record information such as the original remote sensing image data, algorithm parameters of each processing module, intermediate processing results, and the final soil component detection conclusion on the blockchain; the distributed ledger of the blockchain ensures the immutability of data, and encryption technology ensures the security and traceability of data; any access and modification to data will be recorded on the blockchain, providing a highly trustworthy guarantee for the soil component detection results;

[0064] Deploy intelligent drones equipped with high-resolution cameras and small soil detection devices; the drones autonomously plan flight paths according to the regions of interest preliminarily delineated by satellite remote sensing to collect low-altitude high-resolution remote sensing images; in key areas, the drones land and use small soil detection devices to collect soil samples for on-site detection; by establishing a collaborative mechanism between drones and satellite remote sensing, combine the high-precision local information obtained by drones with the macroscopic information of satellite remote sensing to optimize the soil component detection process and improve the comprehensiveness and accuracy of detection.

[0065] For details, please refer to Figure 1 where the preprocessing of multi-modal remote sensing image data includes radiometric calibration and geometric correction of optical remote sensing images, radar remote sensing data, and thermal infrared remote sensing data respectively.

[0066] For details, please refer to Figure 1 where when constructing the fitness function, comprehensively consider the sensitivity of bands to soil components and the correlation between bands, and the fitness function is the sensitivity score of bands to soil components minus the correlation score between bands.

[0067] For details, please refer to Figure 1 , where the selection operation adopts the roulette wheel selection method, and the probability P that an individual is selected i is proportional to its fitness value F i , and the calculation formula is: In the formula, N is the number of individuals in the population.

[0068] For details, please refer to Figure 1 , where the crossover operation adopts single-point crossover or multi-point crossover, and randomly selects a crossover point to exchange the gene segments of two individuals after the crossover point; the mutation operation randomly changes the gene values in an individual with a certain probability.

[0069] For details, please refer to Figure 1 , where the recurrent neural network adopts the long short-term memory network (LSTM) structure, and the output feature vectors of the recurrent neural network RNN and the spatial convolutional neural network SCNN are fused by using the concatenation or weighted summation method.

[0070] For details, please refer to Figure 1 , assuming that the output feature vector of the recurrent neural network is r and the output feature vector of the spatial convolutional neural network is s, then the weighted summation fusion formula is:

[0071] f = w1r + w2s;

[0072] In the formula: w1 and w2 are weights, and w1 + w2 = 1.

[0073] For details, please refer to Figure 1 , where the blockchain platform selects a consensus algorithm such as proof of work (PoW) or proof of stake (PoS), encrypts the detection data, and packs it into a block according to the requirements of the blockchain format and adds it to the blockchain network.

[0074] For details, please refer to Figure 1 , where the UAV path planning algorithm takes into account factors such as terrain, obstacles, and image overlap.

[0075] Embodiment 2

[0076] Embodiment 2 of the present invention also provides an implementation manner of a method for detecting regional soil components based on remote sensing images, which is used to further elaborate and explain the method for detecting regional soil components based on remote sensing images provided in Embodiment 1 above, specifically as follows:

[0077] Data acquisition stage:

[0078] Optical, radar, and thermal infrared remote sensing devices carried by satellites are used to obtain multi-modal remote sensing image data of the target area. At the same time, a ground weather station monitors atmospheric parameters in real time and transmits the data to the data processing center;

[0079] For hyperspectral images, obtain the original data with low resolution for subsequent super-resolution reconstruction;

[0080] Collect a small amount of soil samples in the target area to obtain the measured data of soil components for the initial training and model verification of adaptive band selection.

[0081] Implementation steps of multi-modal data fusion:

[0082] Import optical remote sensing images, radar remote sensing data, and thermal infrared remote sensing data into the data processing system;

[0083] Preprocess each modal data, including radiometric calibration, geometric correction, etc.;

[0084] Use the principal component analysis (PCA) algorithm to extract features from the preprocessed multi-modal data. First, calculate the covariance matrix of each modal data, and then solve the eigenvalues and eigenvectors of the covariance matrix; select the first several eigenvectors with larger eigenvalues, and splice the corresponding eigenvectors of different modal data to obtain the fused feature vector set.

[0085] Implementation steps of hyperspectral image super-resolution reconstruction:

[0086] Construct a generative adversarial network (GAN), including a generator network and a discriminator network; the generator network adopts a convolutional neural network structure and learns the mapping relationship from low-resolution hyperspectral images to high-resolution images through multiple convolutional and deconvolutional operations; the discriminator network also adopts a convolutional neural network to judge the similarity between the generated high-resolution images and the real high-resolution images;

[0087] Prepare a large number of paired low-resolution and high-resolution hyperspectral images as training samples; input the low-resolution images into the generator network to generate high-resolution images, and then input the generated images and the real high-resolution images into the discriminator network together; the discriminator network outputs the judgment results, and adjust the network parameters of the generator and the discriminator according to the judgment results. After multiple rounds of iterative training, the generator can generate high-quality high-resolution hyperspectral images.

[0088] Implementation steps of adaptive band selection:

[0089] According to the prior knowledge of the soil in the target area and the data of a small amount of soil samples collected, construct a fitness function; the fitness function can be defined as the sensitivity score of the band to soil components minus the correlation score between bands to ensure that the selected bands are both sensitive to soil components and independent of each other;

[0090] Randomly generate an initial subset of wavebands as the population of the genetic algorithm. Each waveband subset can be represented by a binary code, where "1" indicates that the waveband is selected and "0" indicates that it is not selected;

[0091] Calculate the fitness value for each individual (waveband subset) in the population; perform a selection operation based on the fitness value, and select individuals with higher fitness values to enter the next generation. The roulette wheel selection method is used, that is, the probability of an individual being selected is proportional to its fitness value;

[0092] Perform crossover and mutation operations on the selected individuals; the crossover operation can use single-point crossover or multi-point crossover, randomly select the crossover point, and exchange the gene segments of two individuals after the crossover point; the mutation operation randomly changes the gene values in the individual with a certain probability, for example, changing "0" to "1" or vice versa;

[0093] After multiple generations of genetic operations, search for the waveband subset that optimizes the fitness function value, and this waveband subset is the optimal waveband combination for soil component detection in this area.

[0094] Implementation steps of real-time atmospheric correction:

[0095] Obtain real-time atmospheric parameters from the ground meteorological station, including temperature, humidity, aerosol concentration, etc., and at the same time obtain the time information when the satellite passes;

[0096] According to the real-time atmospheric parameters, use the atmospheric radiative transfer model to calculate the absorption and scattering coefficients of the atmosphere for the remote sensing signal; commonly used atmospheric radiative transfer models include the 6S model, etc.;

[0097] According to the calculated absorption and scattering coefficients, perform pixel-by-pixel correction on the remote sensing image; for example, for each pixel of the optical remote sensing image, adjust its radiance value according to the corresponding atmospheric parameters and model calculation results to eliminate atmospheric interference and obtain the corrected image.

[0098] Implementation steps of spatio-temporal context information mining:

[0099] Collect time-series remote sensing image data and arrange it in chronological order;

[0100] Construct a recurrent neural network (RNN) and a spatial convolutional neural network (SCNN); the RNN adopts a long short-term memory network (LSTM) structure, which can effectively handle the long-term dependencies in time-series data. The SCNN performs convolutional operations on the remote sensing images in the spatial neighborhood to extract spatial features;

[0101] Input the remote sensing images of the time series into the RNN in sequence. The RNN outputs the feature vectors corresponding to each time point, and these feature vectors reflect the changing trend of soil components over time. At the same time, input a single remote sensing image into the SCNN, and the SCNN outputs the spatial feature vector of this image.

[0102] Fuse the output feature vectors of the RNN and the SCNN, which can be done by methods such as concatenation or weighted summation. Input the fused feature vectors into a classifier, such as a support vector machine (SVM) or a neural network classifier, to classify and predict the soil components.

[0103] Steps for accelerating the implementation of quantum computing:

[0104] For complex algorithms in soil component detection, such as feature extraction algorithms for high-dimensional data and parameter optimization algorithms for soil component inversion models, convert them into forms suitable for quantum computing. For example, for parameter optimization problems, convert the objective function and constraint conditions into the energy function of quantum bits.

[0105] Use the quantum annealing algorithm for calculation. The quantum annealing algorithm searches for the global optimal solution in the state space of quantum bits by simulating the annealing process of a quantum system. Initialize the state of the quantum bits, and then according to the principles of quantum mechanics, let the quantum bits perform operations such as quantum tunneling at a certain temperature, gradually reduce the temperature, and make the quantum bits converge to the state with the lowest energy, which is the optimal solution to the problem.

[0106] Convert the results obtained by quantum computing back to the format in the traditional computing field and apply them to the soil component detection process to improve the computing efficiency and detection speed.

[0107] Steps for blockchain data management:

[0108] Build a blockchain platform and select a suitable blockchain consensus algorithm, such as proof of work (PoW) or proof of stake (PoS), etc.

[0109] During the detection process, when obtaining information such as the original remote sensing image data, the algorithm parameters of each processing module, the intermediate processing results, and the final soil component detection conclusion, encrypt this information and package it into a block according to the format requirements of the blockchain.

[0110] Add the newly generated block to the blockchain network, and use the consensus algorithm to make each node reach an agreement to ensure the consistency and immutability of the data. Any request for accessing and modifying the data needs to be verified and authorized by the blockchain network, and the relevant operation records will be saved on the blockchain to achieve the traceability of the data.

[0111] Steps for drone collaboration:

[0112] Transmit information such as the regional boundary coordinates to the unmanned aerial vehicle (UAV) according to the region of interest preliminarily defined by satellite remote sensing.

[0113] The UAV is equipped with a high-resolution camera and a small soil detection device. It autonomously plans the flight path according to the path planning algorithm and conducts low-altitude high-resolution remote sensing image acquisition in the region of interest. The path planning algorithm can consider factors such as terrain, obstacles, and image overlap to ensure comprehensive and high-quality image acquisition.

[0114] In key areas, the UAV lands and uses the small soil detection device to collect soil samples for on-site detection; transmit the on-site detection results and the collected image data back to the data processing center in real time.

[0115] The data processing center fuses and analyzes the high-precision local information obtained by the UAV with the macroscopic information of satellite remote sensing, comprehensively judges the soil composition, and optimizes the soil composition detection results.

[0116] Through the above specific implementation manners, the method for detecting regional soil composition based on remote sensing images of the present invention can be effectively implemented, achieving the improvement of detection accuracy, efficiency, and data credibility; this method introduces multi-modal data fusion, combines radar remote sensing data, thermal infrared remote sensing data with traditional optical remote sensing images, and enhances the ability to identify soil composition; a series of innovative technical means such as hyperspectral image super-resolution reconstruction, adaptive band selection, real-time atmospheric correction, spatio-temporal context information mining, quantum computing acceleration, blockchain data management, and UAV collaboration are adopted to comprehensively optimize the soil composition detection process and effectively improve the detection level.

[0117] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for detecting regional soil composition based on remote sensing images, characterized in that: The following steps are involved: S1. Acquire multimodal remote sensing image data of a target area, wherein the multimodal remote sensing image data includes optical remote sensing image data, radar remote sensing data, and thermal infrared remote sensing data; and simultaneously acquire atmospheric parameters monitored in real time by a ground meteorological station and time information when a satellite passes over the target area; Collect a small amount of soil samples from the target area to obtain measured data on soil composition; S2. Perform multimodal data fusion on the multimodal remote sensing image data, extract features from each modal data using a principal component analysis algorithm, concatenate feature vectors of different modal data, and construct a comprehensive feature vector set; wherein the principal component analysis algorithm includes the following steps: S21. Let the multimodal data matrix be X, whose dimension is n×p, S22. First calculate the mean vector of the data Then calculate the covariance matrix Then solve the eigenvalue λ of the covariance matrix i and the eigenvector v i (i=1, 2, ...p), and the eigenvalue λ of the covariance matrix C is obtained by solving the characteristic equation |C-λI|=0 i , by solving the linear equations (C-λ i I)v i =0 to get the corresponding eigenvector v i , S23, selecting the first k eigenvectors with the largest eigenvalues, concatenating the corresponding eigenvectors of different modal data, and constructing a comprehensive eigenvector set; Where n is the number of samples and p is the number of features. S3. For the acquired low-resolution hyperspectral image data, a generative adversarial network is used to perform super-resolution reconstruction of the hyperspectral image. The generator network learns the mapping relationship between the low-resolution hyperspectral image and the high-resolution image to generate a high-resolution image. The discriminator network compares the generated image with the real high-resolution image to optimize the generator. S4. Based on the prior knowledge of the soil in the target area or a small amount of soil sample data collected, a genetic algorithm is used to perform adaptive band selection, construct a fitness function, randomly generate an initial band subset as a population, and search for the band subset that optimizes the fitness function value through genetic operations such as selection, crossover, and mutation. Suppose the band sensitivity score function to soil composition is S(b), and the inter-band correlation score function is R(b1, b2), then the fitness function is Where m is the number of bands, b represents the band subset, and R(b1, b2) is the inter-band correlation score function, which is used to evaluate the correlation between the i-th band and the j-th band. S5. Establish a real-time atmospheric correction model. Based on the acquired atmospheric parameters and satellite transit time information, use the atmospheric radiation transfer model to calculate the atmospheric absorption and scattering coefficients of remote sensing signals, and perform real-time correction on remote sensing images. S6. Use recurrent neural networks combined with spatial convolutional neural networks to mine spatiotemporal context information. Use recurrent neural networks to process time series remote sensing images to capture the changing trends of soil composition over time. Use spatial convolutional neural networks to perform convolution operations on remote sensing images of spatial neighborhoods to mine the correlation of spatial neighborhood soil information. The outputs of the recurrent neural network (RNN) and the spatial convolutional neural network are fused and input into the classifier. S7. In response to the complex algorithms in soil composition detection, quantum computing technology is introduced, and the quantum annealing algorithm is used to optimize the parameters of the soil composition inversion model; S8. Use blockchain technology to manage test data reliably, and record information such as original remote sensing image data, algorithm parameters of each processing module, intermediate processing results, and final soil composition test conclusions on the blockchain; S9. Deploy intelligent drones equipped with high-resolution cameras and small soil testing equipment. Based on the areas of interest initially delineated by satellite remote sensing, the drones use path planning algorithms to autonomously plan flight paths for low-altitude high-resolution remote sensing image acquisition, land in key areas to collect soil samples for on-site testing, and integrate the data obtained by the drones with satellite remote sensing data for analysis.

2. The method for detecting regional soil composition based on remote sensing images according to claim 1, characterized in that: The preprocessing of the multimodal remote sensing image data includes performing radiation calibration and geometric correction on the optical remote sensing image, the radar remote sensing data and the thermal infrared remote sensing data respectively.

3. The method for detecting regional soil composition based on remote sensing images according to claim 1, wherein: When constructing the fitness function, the sensitivity of the band to the soil components and the correlation between the bands are comprehensively considered. The fitness function is the sensitivity score of the band to the soil components minus the correlation score between the bands.

4. The method for detecting regional soil composition based on remote sensing images according to claim 1, characterized in that: The selection operation adopts the roulette selection method, and the probability of an individual being selected is P i Its fitness value F i The calculation formula is: Where N is the number of individuals in the population.

5. The method for detecting regional soil composition based on remote sensing images according to claim 1, characterized in that: The crossover operation adopts single-point crossover or multi-point crossover, randomly selecting a crossover point to exchange the gene fragments of two individuals after the crossover point; the mutation operation randomly changes the gene value in the individual with a certain probability.

6. The method for detecting regional soil composition based on remote sensing images according to claim 1, characterized in that: The recurrent neural network adopts a long short-term memory network structure, and the output feature vectors of the recurrent neural network and the spatial convolutional neural network are fused by splicing or weighted summation.

7. The method for detecting regional soil composition based on remote sensing images according to claim 6, characterized in that: Assume that the output feature vector of the recurrent neural network is r and the output feature vector of the spatial convolutional neural network is s, then the weighted summation fusion formula is: f=w1r+w2s; Where: w1 and w2 are weights, and w1+w2=1.

8. The method for detecting regional soil composition based on remote sensing images according to claim 1, characterized in that: The blockchain platform selects a proof-of-work or proof-of-stake consensus algorithm, encrypts the detection data, and packages it into blocks according to the blockchain format requirements and adds it to the blockchain network.

9. The method for detecting regional soil composition based on remote sensing images according to claim 1, characterized in that: The UAV path planning algorithm takes into account factors such as terrain, obstacles, and image overlap.

Images