Agricultural Drought Monitoring Methods and Systems Based on Remote Sensing Technology

Abstract

This invention relates to the field of digital image processing technology, specifically to an agricultural drought monitoring method and system based on remote sensing technology. The technical solution of this invention acquires hyperspectral data and RGB images of farmland areas, utilizes the spatial feature differences of RGB images to complete image block segmentation, and each image block has corresponding hyperspectral data. Based on the variation characteristics of the block's spectral data in continuous bands, water content indicators for each block are extracted in different spectral bands, reflecting the crop's response capability in water-sensitive bands. Data dimensionality reduction is performed based on the parameter differences between blocks characterized by all water content indicators, and the crop water content index of each image block is processed to obtain the overall drought monitoring results for the farmland area. This technical solution achieves effective compression of hyperspectral redundant information and accurate preservation of water-sensitive features, thereby improving the accuracy of large-scale farmland drought monitoring.

Description

Technical Field

[0001] This invention relates to the field of digital image processing technology, and specifically to an agricultural drought monitoring method and system based on remote sensing technology. Background Technology

[0002] In modern agricultural production, drought monitoring is a crucial step in ensuring stable agricultural production and improving water resource utilization efficiency. Drought monitoring involves collecting and analyzing information such as meteorological data, soil moisture, and crop growth status to assess and predict the impact of drought on agricultural production.

[0003] Traditional drought monitoring in farmland typically begins with manual soil moisture measurement, followed by data from meteorological stations to obtain soil relative humidity and precipitation, and finally, assessment using agricultural drought standards. This method is costly, inefficient, and difficult to apply at large-scale field levels. With technological advancements, remote sensing-based drought monitoring methods are gradually being adopted in agricultural production. These methods primarily construct drought monitoring models by collecting hyperspectral data of the target crop area. However, due to the large data volume, the monitoring process can lead to the loss of dimensional data representing drought characteristics, resulting in inaccurate drought monitoring results. Summary of the Invention

[0004] To address the technical challenge of reducing the probability of losing water content information during dimensionality reduction of hyperspectral data in drought monitoring based on farmland areas, this invention aims to provide a remote sensing-based agricultural drought monitoring method and system to improve the accuracy of drought monitoring in large-scale farmland. The specific technical solution adopted is as follows:

[0005] In a first aspect, embodiments of the present invention provide a method for monitoring agricultural drought based on remote sensing technology, the method comprising:

[0006] Remote sensing data was collected from farmland areas under drought monitoring to obtain RGB images and hyperspectral data of the farmland areas.

[0007] Image blocks are segmented based on the differences in image features represented by the RGB images, and the block spectral data of the hyperspectral data corresponding to each segmented image block is determined.

[0008] Based on the variation characteristics of the block spectral data of each image block in continuous spectral bands, the water content indication parameters of the crops growing in each image block in different spectral bands are obtained.

[0009] Data dimensionality reduction is performed based on the differences in parameters representing the crop in each image block according to all moisture content hint parameters to obtain the crop moisture content index for each image block;

[0010] Based on the crop moisture content index of all image blocks, the drought monitoring results for farmland areas were determined.

[0011] In one optional embodiment, based on the variation characteristics of the block spectral data of each image block in continuous spectral bands, the water content indication parameters of the crops growing in each image block in different spectral bands are obtained, including:

[0012] The crop growth index for each image block is obtained based on the crop growth status represented by the block spectral data of each image block.

[0013] Based on the difference in crop growth index between each image block and its surrounding blocks, the water transfer characteristics of each image block are obtained.

[0014] Based on the differences in the spectral data of each image block in different spectral bands, and combined with the differences in crop growth index between the block and other blocks, the water content comparison parameter of each image block is obtained.

[0015] Based on the moisture transfer characteristics and water content comparison parameters of each image block, the water content indication parameters of the corresponding image block are obtained.

[0016] In one optional embodiment, the crop growth index for each image block is obtained based on the crop growth status characterized by the block spectral data of each image block, including:

[0017] The spectral data of each image block is input into the vegetation index processing model to obtain the normalized vegetation index of each image block.

[0018] The block spectral data of each image block is input into the variance mean calculation model to obtain the data variance mean of the block spectral data of the image block in different spectral bands.

[0019] The crop growth index for each image block is obtained by comparing the normalized vegetation index of each image block with the mean of the corresponding data variance.

[0020] In one optional embodiment, the water transfer characteristics of each image block are obtained based on the difference in crop growth index between each image block and its surrounding blocks, including:

[0021] Based on the difference in crop growth index between each image block and its surrounding blocks, the direction of the farmland center is determined in relation to the relative position of each image block and its surrounding blocks.

[0022] Connect the edge data points of the current image block with the centroid of the block in the direction of the farmland center of each image block to obtain the directional line segment representing the change in water content direction on the current image block;

[0023] The water transfer characteristics of each image block are obtained by comparing the spectral data of different spectral bands corresponding to the directional line segments of each image block.

[0024] In one optional embodiment, the farmland center direction relative to the surrounding blocks is determined based on the difference in crop growth index between each image block and its surrounding blocks, including:

[0025] Determine whether the crop growth index of the current image block is less than the crop growth index of the surrounding blocks;

[0026] If so, the surrounding blocks larger than the current image block are determined as regional direction blocks, and the farmland center direction of the current image block is obtained by summing the centroid vectors connecting the first centroid position of all regional direction blocks and the second centroid position of the current image block.

[0027] If not, the direction from which the centroid of the block adjacent to the current image block points to the second centroid of the current image block is determined as the direction of the farmland center of the current image block.

[0028] In one optional embodiment, based on the differences in spectral data of each image block across different spectral bands, and combined with the differences in crop growth index between that block and other blocks, a water content comparison parameter for each image block is obtained, including:

[0029] Interpolation was used to complete the spectral data of all blocks, and the mean square error of the spectral data between the current image block and other image blocks in different spectral bands, as well as the exponential difference of the crop growth index, were calculated.

[0030] Based on the mean square error of all spectral data and all exponential differences of the current image block, the water content comparison parameters of the current image block are obtained.

[0031] In one optional embodiment, the water content reference parameters for the current image block are obtained based on the mean square error of all spectral data and all exponential differences of the current image block, including:

[0032] The mean square error and all exponential differences of all spectral data of the current image block are sorted in descending order to obtain the first and second sorting results of the current image block;

[0033] Based on the degree of difference between the first and second sorting results, the water content comparison parameters of the current image block are obtained.

[0034] In one optional embodiment, data dimensionality reduction is performed based on the differences in parameters characterizing the crops within each image block according to all moisture content cue parameters, to obtain the crop moisture content index for each image block, including:

[0035] A data matrix is ​​constructed based on the spectral bands and corresponding spectral data of the block spectral data of each image block to obtain an initial data matrix representing the block spectral data of each image block in each spectral band.

[0036] The initial data matrix of each image block in each spectral band is subjected to mean-removal processing to obtain the block data matrix of each image block in each spectral band.

[0037] The covariance matrix of each image block is obtained by weighting the block data matrix of all water content cue parameters and all spectral bands for each image block.

[0038] The covariance matrix of each image block is decomposed into eigenvalues ​​to obtain the crop moisture index of each image block.

[0039] In one optional embodiment, drought monitoring results for farmland areas are determined based on crop moisture indexes of all image blocks, including:

[0040] The crop water content index of all image blocks is input into the drought analysis model, which is a machine learning model trained based on hyperspectral data and corresponding drought indices.

[0041] Based on the output of the drought analysis model, the drought monitoring results for farmland areas are determined.

[0042] Secondly, embodiments of the present invention also provide an agricultural drought monitoring system based on remote sensing technology, applicable to any of the monitoring methods in the first aspect, the system comprising:

[0043] The data acquisition terminal is used to perform remote sensing data acquisition in farmland areas under drought monitoring and to output RGB images and hyperspectral data of the farmland areas.

[0044] The data processing terminal, connected to the data acquisition terminal, is used to segment image blocks based on the differences in image features represented by the RGB images, and to determine the block spectral data corresponding to each segmented image block in the hyperspectral data. Based on the variation characteristics of the block spectral data of each image block in continuous spectral bands, it obtains the water content indication parameters of the crops growing within each image block in different spectral bands. Based on the parameter differences representing the crop parameters of each image block using all water content indication parameters, it performs data dimensionality reduction processing to obtain the crop water content index of each image block. Based on the crop water content index of all image blocks, it determines the drought monitoring results for the farmland area.

[0045] The present invention has the following beneficial effects:

[0046] The technical solution of this invention performs joint acquisition of hyperspectral data and RGB images of farmland areas, utilizes the spatial feature differences of RGB images to complete image block segmentation, and refines the crop growth blocks within the farmland area, ensuring that each image block has corresponding spectral data. Based on the variation characteristics of the block's spectral data in continuous bands, water content indication parameters are extracted for each block under different spectral bands, reflecting the crop's response capability in water-sensitive bands. Data dimensionality reduction is performed based on the parameter differences between blocks represented by all water content indication parameters, allowing the dimensionality reduction process to prioritize the preservation of spectral features sensitive to water changes, thereby reducing the probability of losing water content information dimensions and improving the reliability of the crop water content index. The crop water content index of each image block is processed to obtain the overall drought monitoring results for the farmland area. This technical solution, through a logical chain of spatial segmentation, spectral feature extraction, constrained dimensionality reduction of water content indication parameters, crop water content index calculation, and regional drought output, achieves effective compression of hyperspectral redundant information and accurate preservation of water-sensitive features, thereby improving the accuracy of drought monitoring in large-scale farmland. Attached Figure Description

[0047] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 A flowchart illustrating an agricultural drought monitoring method based on remote sensing technology, as provided in one embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of the direction of the center of a farmland area provided in one embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram of an agricultural drought monitoring system based on remote sensing technology, provided as an embodiment of the present invention. Detailed Implementation

[0051] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an agricultural drought monitoring method and system based on remote sensing technology proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0053] For drought monitoring in agriculture, the surface cover of crops (or agricultural crops) before the growing season is mainly bare soil, while during the growing season it is mainly vegetation. At this time, the accuracy of soil moisture monitoring under vegetation cover is significantly affected by the vegetation layer, making monitoring more difficult. Therefore, the combination of multi-source remote sensing data is widely used for regional-scale soil moisture content and agricultural drought monitoring. Optical remote sensing is one of the important technologies in agricultural drought monitoring. It mainly uses remote sensing data in the visible, near-infrared, and thermal infrared bands to monitor the condition of soil and vegetation. This method assesses the degree of crop drought by analyzing differences in spectral characteristics and bands sensitive to moisture changes, such as by calculating the Vegetation Supply Water Index (VSWI).

[0054] A crucial preprocessing step in all hyperspectral imaging applications is data dimensionality reduction. Due to the continuity of band images, redundant information exists between bands. The high correlation between adjacent bands in hyperspectral images leads to data redundancy. Those skilled in the art will understand that for farmland areas of a certain size, the large volume of hyperspectral data makes direct analysis computationally expensive and time-consuming, thus necessitating dimensionality reduction. However, different areas within farmland exhibit varying spectral bands that reflect drought characteristics. Traditional dimensionality reduction methods, relying solely on the distribution relationships of data across dimensions as a reference standard, may result in the loss of dimensions representing drought characteristics, leading to insufficient accuracy in drought monitoring results. To address these technical problems, the following detailed description, in conjunction with the accompanying drawings, illustrates a specific scheme for an agricultural drought monitoring method and system based on remote sensing technology provided by this invention.

[0055] Please see Figure 1 , Figure 1 This is a flowchart illustrating an agricultural drought monitoring method based on remote sensing technology, provided as an embodiment of the present invention. This drought monitoring method can be applied to a data processing terminal for drought monitoring. The data processing terminal can be a computer or a server, as long as it can run the method; no specific limitations are imposed here. The drought monitoring method includes:

[0056] S11. Perform remote sensing data collection on farmland areas to be monitored for drought conditions in order to obtain RGB images and hyperspectral data of the farmland areas.

[0057] Specifically, drones equipped with hyperspectral imagers and cameras can be used to collect hyperspectral data and RGB images of farmland areas. RGB images are color images composed of three color channels: Red, Green, and Blue. Each pixel consists of three components: R, G, and B, with each component's value ranging from 0 to 255. During the acquisition process, it is necessary to avoid the impact of adverse weather conditions (such as rain, fog, or snow). In clear weather, RGB images can be acquired by taking pictures during the drone's flight. The hyperspectral imager used in this embodiment has a spectral range of 400-1000 nm, a spectral resolution of 2.1 nm, 150 spectral channels, a sampling interval of 1.07 nm, and a maximum frame rate of 249 frames per second. After acquiring hyperspectral data, the data is stored using a band-interleaved-by-pixel (BIP) encoding format.

[0058] It should be noted that after planning the flight mission based on the location of the target farmland, the drone's flight altitude, speed, and other information are set to ensure that the drone's flight range covers the entire monitored farmland area. For example, the drone's flight altitude is set to 50m and its flight speed to 3m / s. The spatial resolution of the acquired hyperspectral image is 0.03m × 0.03m. During the drone's flight mission, the hyperspectral data and RGB images of the farmland area collected by the drone are simultaneously transmitted to the data processing terminal. It can be understood that the data recorded in the RGB image is two-dimensional, with each pixel corresponding to an RGB value containing information from the red, green, and blue bands; the hyperspectral data is three-dimensional, with each pixel corresponding to a spectral curve, which is a series of continuous spectral reflectance values ​​containing information from tens to hundreds of consecutive narrow bands.

[0059] At this point, RGB images and hyperspectral data of the farmland area have been acquired, and we proceed to step S12.

[0060] S12. Perform image block segmentation based on the differences in image features represented by the RGB image, and determine the block spectral data of the hyperspectral data corresponding to each segmented image block.

[0061] Specifically, the collected farmland area consists of areas including crop fields, bare land, and general vegetation. These different areas have different image characteristics, and the characteristics of hyperspectral data also vary across different image areas. Using a uniform data dimensionality reduction method for different areas would lead to the loss of important information. Therefore, it is necessary to segment the RGB image according to its different internal texture features. This can be achieved using a superpixel segmentation algorithm, dividing the RGB image into multiple superpixel blocks, i.e., multiple image blocks. The differences in image features between blocks of the RGB image are represented in different image blocks. Since superpixel segmentation can divide different regions in an RGB image into multiple image blocks according to their color and texture, in the RGB image of the farmland area, different superpixel blocks can represent farmland areas with different characteristics. To perform accurate dimensionality reduction on the hyperspectral data of the farmland area, the segmented image blocks of the RGB image are synchronized to the hyperspectral data, which is then divided into multiple block spectral data. Each segmented image block is mapped to the block spectral data of its corresponding block.

[0062] It should be noted that different soil types have varying water retention capacities, which directly affect the spectral representation of soil moisture and crop water status. Differences in crop species imply varying photosynthetic and water use efficiencies, thus influencing the spectral characteristics of the crop canopy. Changes in climatic conditions such as temperature, humidity, and light intensity also affect the physiological state and spectral response of crops. Therefore, it is necessary to segment the hyperspectral data to provide an analytical foundation for the subsequent dimensionality reduction process.

[0063] S13. Based on the variation characteristics of the block spectral data of each image block in continuous spectral bands, obtain the water content indication parameters of the crops growing in each image block in different spectral bands.

[0064] Specifically, the spectral information representing different regions within different blocks varies. To calculate the drought information characteristics reflected in the block spectral data of a specific block, it is necessary to compare and analyze the block spectral data from different blocks. For the block spectral data of any block, each pixel is constructed as multi-dimensional data, with different dimensions representing the reflectance information of different spectral bands collected by the spectrometer at a given pixel location. The water content indicator parameter represents the reference value of the water content of the crops growing within the corresponding image block, which can be used to assess the drought characteristics of that block. Since the block spectral data includes reflectance information from multiple spectral bands, but only some spectral bands can reflect the crop's water content, the relationship between each band and changes in crop moisture can be analyzed to evaluate the target band that is more sensitive to water content and derive the water content indicator parameter.

[0065] For example, step S13 includes sub-steps S13-1 to S13-4, which are described in detail below:

[0066] S13-1. Based on the crop growth status represented by the block spectral data of each image block, obtain the crop growth index for each image block. Different regions have different regional characteristics, and the block spectral data they represent also differ. In the analysis of agricultural drought monitoring, the main focus is on analyzing the hyperspectral data within farmland areas for drought monitoring. Therefore, it is necessary to determine the characteristics of the crop growth status of farmland represented by different block spectral data, and to characterize whether there is water shortage in the corresponding image block through the crop growth index. Based on the correspondence between block spectral data and crop growth index, a data calculation model can be formulated. By calculating and processing the block spectral data corresponding to each image block through the data calculation model, the crop growth index of the corresponding image block can be obtained.

[0067] In one specific implementation, sub-step S13-1 includes:

[0068] The first step involves inputting the spectral data of each image block into the vegetation index processing model to obtain the Normalized Difference Vegetation Index (NDVI) for each image block. The NDVI is a crucial parameter reflecting the vegetation growth status and coverage within a block; its value ranges from [0, 1] and increases with increasing vegetation coverage. The reflectance values ​​in the near-infrared and red bands can be extracted from the spectral data of each image block, and the NDVI value can be calculated using the NDVI calculation formula.

[0069] In the collected hyperspectral data, the vegetation-covered areas may include not only the areas covered by crops but also surrounding vegetation areas. Compared to crop areas, these areas have a richer variety of plant species, and due to differences in composition, different plant species exhibit certain differences in their representation across different bands in the hyperspectral data. To distinguish between crop areas and general vegetation-covered areas, further computational analysis of the hyperspectral data is required.

[0070] The second step involves inputting the block spectral data of each image block into the variance-mean calculation model to obtain the mean variance of the block spectral data across different spectral bands. For each image block, the block spectral data can be calculated using the variance-mean calculation formula. First, the variance of the data in different spectral bands within the current image block is calculated, and then the mean variance of each spectral band is calculated to obtain the mean variance. .

[0071] The third step is to obtain the crop growth index for each image block by calculating the ratio of the normalized vegetation index to the mean of the corresponding data variance. Calculate the crop growth index for the current image block. The specific calculation process is as follows:

[0072] ,in, The crop growth index, The mean of the data variance The normalized vegetation index is used.

[0073] Based on the above analysis, the normalized vegetation index (NVI) of the image block is relatively large, and for general vegetation areas, the internal plant distribution is relatively complex, meaning the mean variance of the data is relatively large. The crop growth index of the image block is calculated by the ratio of the NVI to the mean variance. The larger the NVI and the smaller the mean variance, the larger the crop growth index, indicating a better crop growth level within the current image block; conversely, the smaller the crop growth index, the worse the crop growth level. It should be noted that in this embodiment of the invention, the mean variance is not zero. That is, when the mean variance is 0, the overall scheme has no analytical significance. Therefore, it is limited. Of course, in other embodiments of the invention, when the mean variance is detected to be 0, it can be adaptively adjusted to a preset value such as 0.1 to avoid the formula being meaningless. The same principle applies to the calculation of other formulas in this application.

[0074] To analyze how different spectral bands represent crop water content information, it is necessary to evaluate and analyze the data in each spectral band across different regions. Different crops have different spectral bands that can reflect water content information. In the case of unknown crop species, it is necessary to further analyze the water content expression ability of each dimension of data by combining the spectral data characteristics of different regions.

[0075] S13-2. Based on the difference in crop growth index between each image block and its surrounding blocks, the water transfer characteristics of each image block are obtained. By combining the crop growth index of the image block, the data characteristics within the block can be analyzed, and the crop water content of different dimensions can be calculated. Single-dimensional characterization can be achieved through water transfer characteristics. Water transfer characteristics quantify the parameters of water loss in an image block based on the difference in crop growth index between the image block and its surrounding blocks. When performing spectral analysis, it is difficult to accurately determine water information by only looking at the data within a single block, as it may be isolated. The direction of the farmland center can be introduced, placing a single block within the overall gradient from the center to the edge for analysis. This allows for the calculation of the smoothness of the spectrum with spatial gradient, thereby deriving the water transfer characteristics of each image block.

[0076] While irrigation in farmland is typically uniform, the rate of water loss is not uniform. Generally, water loss is influenced by factors such as wind speed and surface temperature. The edges of farmland are more susceptible to environmental influences, resulting in higher water loss rates, while the central areas experience lower rates. Therefore, spectral data representing crop moisture content usually exhibits a diffusion trend from the center of the farmland outwards. However, image blocks obtained through superpixel segmentation often cannot represent the entire farmland area; thus, comparative analysis with adjacent blocks is necessary.

[0077] Based on this, in one specific implementation, sub-step S13-2 includes:

[0078] The first step is to determine the farmland center direction based on the difference in crop growth index between each image block and its surrounding blocks. The farmland center direction represents the relative spatial position of the current image block and the center of the actual farmland area. Under this direction, the farmland crops exhibit a gradual degree of water loss, and their spectral data also show a continuous and similar change relationship. Therefore, it is necessary to determine the farmland center direction based on the difference in crop growth index between each image block and its surrounding blocks.

[0079] Specifically, when determining the farmland center direction of each image block, it is necessary to determine whether the crop growth index of the current image block is less than that of the surrounding blocks. If the crop growth index of the current image block is less than that of the surrounding blocks, it indicates that the crop growth index of the current image block is not a maximum value for that region. In this case, the surrounding blocks with a higher crop growth index than the current image block are identified as regional directional blocks. The farmland center direction of the current image block is then obtained by summing the centroid vectors connecting the first centroid position of all regional directional blocks to the second centroid position of the current image block.

[0080] A vector can be constructed pointing from the centroid of the current image patch to the centroid of the farmland area block, with the difference in crop growth index between the farmland area block and the current image patch used as the magnitude of this vector. The sum of these vectors is calculated, and the farmland area block whose direction is closest to this vector is selected. The direction from the current image patch to the centroid of this closest farmland area block is then combined with the direction of the farmland center at that point. (See also...) Figure 2 , Figure 2 This is a schematic diagram of the farmland center direction of image block A1 in the farmland area. When determining the farmland center direction, the crop growth index of image block A1 is compared with that of its surrounding blocks A2-A7. If the crop growth index of image block A1 is not a maximum value, then vector J1 is determined as the farmland center direction.

[0081] If the crop growth index of the current image block is greater than that of the surrounding blocks, it indicates that the crop growth index of the current image block is the maximum value in that region. Therefore, the direction from the centroid of the adjacent image block to the second centroid of the current image block is defined as the farmland center direction of the current image block. If no regional directional blocks exist, the farmland center direction is defined as the direction from the centroid of the adjacent image block to the centroid of the current image block; that is, the current image block has multiple farmland center directions. After obtaining the farmland center directions of each point in the image block, a preliminary assessment can be made of the ability to represent water content information in different dimensions based on the changes in data in each dimension of the image block.

[0082] The second step involves connecting the edge data points of each image block to its centroid along the direction of the farmland center. If multiple edge data points exist, the one closest to the centroid is selected for connection, thus obtaining the directional line segment representing the change in water content direction within the current image block. Please continue reading... Figure 2 Connect the edge data points of the current image block with the centroid of the block in the direction of the farmland center to obtain multi-dimensional hyperspectral data of the data points through which the line segment of the water content change direction passes; Figure 2 The center point represents the centroid of each image block, the direction of arrow J1 is the direction of the center of the farmland in the middle block, and the direction line segment of image block A1 is J2.

[0083] Among them, edge data points are points located at the edges of the current image block along the direction of the farmland center after determining the centroid of the current image block in the direction of the farmland center. These points are then used as endpoints. Figure 2 The starting point on line segment J2 in the middle direction.

[0084] The third step involves obtaining the water transfer characteristics of each image block based on the differences in spectral data corresponding to different spectral bands along the directional line segments. This can be achieved by calculating the differences between the data points along the directional line segments and the data points along the next data point, subtracting the data points along the next data point from each dimension of the data points along the directional line segments. And construct the difference sequence according to the order of the connections. Then calculate the sequence The difference between the middle element and the next element. The crop water transfer characteristic T of a certain dimension of spectral data in the current block is calculated using the following formula:

[0085]

[0086] In the above formula, m represents the number of farmland center directions in the current image block; This represents the number of pixels traversed in the direction of the center of the j-th farmland; This represents the sequence of pixel differences along the line connecting the edge point and the center point of the j-th farmland in the direction of the center. The difference between the data of the i-th pixel and the data of the (i+1)-th pixel; This represents the mean of the differences along the direction of the center of the j-th farmland; exp is the natural exponential function, and m, n, and j are all natural numbers. (This is because...) The value is obtained after two difference operations; therefore, here... In real-world scenarios, it is greater than or equal to 3.

[0087] In the above analysis, farmland water loss exhibits certain characteristics. By analyzing the directions of adjacent image blocks and the current image block to determine the direction of the farmland center, the variation factors of the data differences in each dimension of the farmland center direction within the image block are extracted. The greater the distance between farmland centers, the greater the degree of water loss. Spectral data representing water changes will gradually and slowly change in the same direction as the distance increases. Therefore, the difference in spectral data that best represents water changes should be smaller, meaning the difference in the value of that dimension of data should be smaller. The value is relatively small. After adjusting the logical relationship by exp(-), T represents the water transfer characteristics of crops under different spectral bands.

[0088] For data that can indicate the moisture content of crops, the differences should be greater in different regions: the leaf structure of crop areas differs from that of general vegetation-covered areas, and the spectral data bands that can reflect their moisture content also differ; when compared with areas without vegetation cover, the differences in these dimensions of spectral data will be even greater.

[0089] S13-3. Based on the differences in spectral data across different spectral bands for each image block, and combined with the differences in crop growth indices between this block and other blocks, obtain the water content comparison parameter for each image block. The water content comparison parameter measures the degree of matching between the spectral differences in a certain band and the differences in crop growth in different regions. If the spectral differences in a band can accurately reflect the differences in crop moisture and growth levels in different regions, it indicates that this band is sensitive to water. If there is no correlation between the spectral differences and the differences in crop growth levels, then this band is not suitable as a key indicator for water content monitoring.

[0090] For example, sub-step S13-3 includes:

[0091] The first step involves using interpolation to complete the spectral data of all blocks, and then calculating the mean square error (MSE) of the spectral data between the current image block and other image blocks in different spectral bands, as well as the exponential difference of the crop growth index. Since the number of pixels in different image blocks may vary, interpolation can be used to complete the spectral data of all blocks to the same length, ensuring comparability. The spectral differences between the current block and other blocks are calculated band by band and characterized by the MSE. Simultaneously, the difference in the crop growth index between two blocks is also calculated.

[0092] The second step involves obtaining the water content comparison parameter for the current image block based on the mean square error (MSE) and all exponential differences of all spectral data for that block. For example, the absolute values ​​of the MSE and exponential differences of all spectral data for the current image block are sorted in ascending order to obtain a first and second sorting result. The water content comparison parameter is then obtained based on the degree of difference between the first and second sorting results. Specifically, the mean square error (MSE) of the spectral data for the current image block and the exponential differences of the crop growth index with other image blocks can be sorted in ascending order. The water content comparison parameter for a specific dimension of the current block can then be calculated. :

[0093]

[0094] In the above formula, u represents the number of image blocks. This indicates the order of the mean square error (MSE) sequence of the spectral data from the current block and the v-th block. This represents the sequence order of the exponential differences between the current image block and the v-th image block, tanh() is the hyperbolic tangent function, and c represents the adjustment coefficient. In this embodiment, c=1, a safety value set to prevent the denominator from being 0. In the above formula... The smaller the value, the closer the difference between the current image block and the crop growth index in this dimension of data is. This parameter can be used as an auxiliary feature in analyzing water content differences. The water content comparison parameter is obtained by calculating the consistency between the spectral difference sorting sequence and the growth difference sorting sequence.

[0095] S13-4. Based on the water transfer characteristics and water content comparison parameters of each image block, obtain the corresponding water content indication parameters for that image block. Obtain the crop water transfer characteristics T and water content comparison parameters for different image blocks in each spectral band. Then, the water content indicator parameters for each spectral band can be calculated. :

[0096]

[0097] In the above formula, u represents the number of image blocks. Let v be the crop growth index of the v-th image block. Let v be the water content comparison parameter for the v-th image block. Let v be the moisture transfer feature of the v-th image block, and norm() be the normalization function. The above method combines the crop moisture transfer features and water content comparison parameters of different image blocks, and adjusts the calculation weights according to the crop growth index of the blocks to amplify the proportion of crop blocks.

[0098] At this point, the water content parameters of the crops growing in each image block in different spectral bands have been obtained, and we proceed to step S14.

[0099] S14. Perform data dimensionality reduction based on the differences in parameters representing the crop in each image block according to all water content hint parameters, in order to obtain the crop water content index for each image block.

[0100] Specifically, the water content indicators for each image block in different spectral bands can be used as priority parameters for data dimensionality reduction. Prioritizing data with more pronounced water content indicators improves the retention of data in that dimension during dimensionality reduction. This ensures that data with drought analysis characteristics is accurately preserved after dimensionality reduction, and a crop water content index can be derived based on this data. It should be noted that the crop water content index can be the spectral data after dimensionality reduction for each image block; or it can be an indicator parameter calculated based on this spectral data. This index parameter characterizes the drought status of crops in the corresponding image block. For example, a higher crop water content index indicates a more severe drought in the image block; conversely, a lower index indicates less impact from drought on crops.

[0101] For example, step S14 includes sub-steps S14-1 to S14-4, which are described in detail below:

[0102] S14-1. Construct a data matrix based on the spectral bands and corresponding spectral data of each image block's spectral data to obtain an initial data matrix representing the block's spectral data in each spectral band. Arrange the block spectral data into an initial data matrix A with n rows and m columns, where n is the number of data points and m is the number of spectral bands.

[0103] S14-2. The initial data matrix of each image block in each spectral band is mean-reduced to obtain the block data matrix of each image block in each spectral band. The data in each spectral band of the initial data matrix A is zero-mean-reduced by subtracting the mean of each data point in a row from the mean of the data in that row, resulting in the processed block data matrix B.

[0104] S14-3. Based on all water content indicator parameters and the block data matrix of all spectral bands for each image block, a weighted calculation is performed to obtain the covariance matrix of each image block. To preserve, during dimensionality reduction, the dimensionality reduction matrix needs to be analyzed by considering the optimization priorities of different dimensions of the hyperspectral data, focusing on the dimensions that best represent the current crop's water content. The water content indicator parameters for each image block are used as priority reference standards, and the dimensionality reduction priority parameters for each dimension are then used to determine the optimal dimensionality reduction matrix. Multiply the corresponding dimension (row) data of block data matrix B, and further adjust the differences between data using data priority to obtain the adjustment matrix. ; Calculate the adjustment matrix The covariance matrix C is obtained, which is the covariance matrix of each image block.

[0105] S14-4. Eigenvalue decomposition is performed on the covariance matrix of each image block to obtain the crop moisture content index of each image block. The eigenvalues ​​and eigenvectors of the covariance matrix C can be calculated. The eigenvectors are arranged row-wise from top to bottom according to the corresponding eigenvalues, and the first k rows are taken to form matrix P. The first k rows are determined by the cumulative contribution rate of the eigenvalues. In this embodiment, the cumulative contribution rate threshold is set to 80% based on empirical values. Finally, matrix P is multiplied by the initial data matrix A to obtain the data after dimensionality reduction to k dimensions. According to the above dimensionality reduction method, matrix P is multiplied by the initial data matrix A, and the product is represented by matrix D. The data in matrix D is the dimensionality-reduced data, and the data after dimensionality reduction is determined as the crop moisture content index.

[0106] S15. Based on the crop moisture content index of all image blocks, determine the drought monitoring results for the farmland area.

[0107] Specifically, the crop moisture content index is data after dimensionality reduction processing. This reduces the deviation in drought monitoring results caused by data redundancy and effectively reduces the amount of data computation and processing. When carrying out large-scale farmland drought monitoring, the drought monitoring results of farmland areas can be determined based on the data after dimensionality reduction processing.

[0108] For example, step S15 includes sub-steps S15-1 to S15-2, which are described in detail below:

[0109] S15-1. Input the crop water content index of all image blocks into the drought analysis model, which is a machine learning model trained based on hyperspectral data and corresponding drought indices. Taking the dimensionality-reduced spectral data as an example, the crop water content index is input into the machine learning model for drought analysis. The model training process is as follows: Hyperspectral data is acquired through the internet and remote sensing platforms, and objective annotations are constructed using existing drought indices (such as SPI, SPEI, etc.). The hyperspectral data of each set is processed to obtain the crop water content index. The data is divided into training and validation sets in a 7:3 ratio, and K-fold cross-validation is used in the training set to improve robustness. During the training phase, random forest regression is used as the core, and subsets are constructed using Bootstrap sampling. Each tree is split on the feature subset according to the principle of minimizing the mean square error, and the final result is the mean of all tree predictions. The number, depth, and number of features of trees are adjusted through grid search, and the optimal parameters are selected based on the cross-validation error to avoid overfitting. After the model training is completed, it is evaluated on an independent test set, and the predicted drought index is output.

[0110] S15-2. Based on the output of the drought analysis model, determine the drought monitoring results for the farmland area. The output of the drought analysis model can be a predicted drought index value, or a classification such as mild drought, moderate drought, or severe drought. The predicted results of all image blocks are stitched together to obtain a drought distribution map of the entire farmland area, ultimately outputting a macro-level drought monitoring result.

[0111] Based on the same technical concept as the monitoring method, this invention also provides an agricultural drought monitoring system based on remote sensing technology, applicable to any of the above-mentioned monitoring methods. Please refer to [link to relevant documentation]. Figure 3 , Figure 3 This is a schematic diagram of the monitoring system. The monitoring system includes a data acquisition terminal 1 and a data processing terminal 2, with the data processing terminal 2 connected to the data acquisition terminal 1.

[0112] The data acquisition terminal can be a drone or other types of aircraft. It is used to collect remote sensing data in farmland areas under drought monitoring and output RGB images and hyperspectral data of the farmland areas.

[0113] The data processing terminal can be configured as a computer device or server, as long as it can perform data processing. The data processing terminal is used to segment image blocks based on the differences in image features represented by the RGB images, and determine the block spectral data of the hyperspectral data corresponding to each segmented image block; based on the variation characteristics of the block spectral data of each image block in continuous spectral bands, obtain the water content indication parameters of the crops growing in each image block in different spectral bands; perform data dimensionality reduction processing based on the parameter differences of the crops in each image block represented by all water content indication parameters to obtain the crop water content index of each image block; and determine the drought monitoring results of the farmland area based on the crop water content index of all image blocks.

[0114] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0115] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for monitoring agricultural drought based on remote sensing technology, characterized in that, The method comprises: Perform remote sensing data collection on the farmland area to be monitored for drought to obtain RGB images and hyperspectral data of the farmland area; According to the image block feature difference represented by the RGB image, image block segmentation is performed, and block spectral data corresponding to the hyperspectral data of each segmented image block is determined; According to the change characteristics of the block spectral data of each image block in the continuous spectral band, the water content prompt parameter of the crop grown in each image block in different spectral bands is obtained; According to the parameter difference of each image block representing the crop in the block, data dimension reduction processing is performed to obtain the crop water index of each image block; According to the crop water index of all image blocks, the drought monitoring result of the farmland area is determined; The method for obtaining the water content prompt parameter comprises: According to the crop growth state represented by the block spectral data of each image block, the crop growth index of each image block is obtained; According to the crop growth index difference between each image block and its surrounding blocks, the water transfer feature of each image block is obtained; According to the data difference of the block spectral data of each image block in different spectral bands, and combining the crop growth index difference between the block and other blocks, the water content reference parameter of each image block is obtained; According to the water transfer feature and the water content reference parameter of each image block, the water content prompt parameter of the corresponding image block is obtained; Calculating moisture content prompt parameters under each spectral band : ; in the above formula, u represents the number of image blocks, is the crop growth index of the vth image block, is the water content reference parameter of the vth image block, is the water transfer characteristic of the vth image block, and norm() is a normalization function. The method for obtaining the water transfer feature of each image block comprises: According to the crop growth index difference between each image block and its surrounding blocks, the farmland center direction of the relative position relationship between each image block and its surrounding blocks is determined; The edge data points and block center of the current image block are connected by the farmland center direction of each image block to obtain the direction line segment representing the direction change of the water content on the current image block; According to the difference between the spectral data corresponding to the direction line segment of each image block in different spectral bands, the water transfer feature of each image block is obtained; The method for obtaining the water content reference parameter comprises: Interpolation is used to fill in the block spectral data of all blocks, and the spectral data mean square error between the current image block and other image blocks in different spectral bands and the index difference value of the crop growth index are calculated respectively; All spectral data mean square errors and all index difference values of the current image block are arranged in descending order respectively to obtain the first sorting result and the second sorting result of the current image block; According to the sorting difference degree between the first sorting result and the second sorting result, the water content reference parameter of the current image block is obtained; sequentially arrange the mean square error (MSE) and the index difference of the crop growth index of the current image block and the spectral data of each of the other image blocks in order from small to large : ; in the above formula, u represents the number of image blocks, represents the sequence order of the spectral data mean square error MSE of the current block and the vth block, represents the sequence order of the exponential difference value of the current image block and the vth image block, tanh() is the hyperbolic tangent function, and c represents an adjustment coefficient, c=1. The method for obtaining the crop water index of each image block by data dimension reduction processing comprises: According to the spectral band and corresponding spectral data of the block spectral data of each image block, a data matrix is constructed to obtain the initial data matrix representing the block spectral data of each image block in each spectral band; de-meaning the initial data matrix of each image block in each spectral band to obtain a block data matrix of each image block in each spectral band; performing weighted calculation according to all water content prompt parameters of each image block and the block data matrix of all spectral bands to obtain a covariance matrix of each image block; performing eigenvalue decomposition on the covariance matrix of each image block to obtain a crop water index of each image block.

2. The remote sensing technology-based agricultural drought monitoring method according to claim 1, characterized in that, The crop growth index of each image block is obtained according to the crop growth state represented by the block spectral data of each image block, and the method comprises: inputting the block spectral data of each image block into a vegetation index processing model to obtain a normalized vegetation index of each image block; inputting the block spectral data of each image block into a variance mean calculation model to obtain a data variance mean of the block spectral data of the image block in different spectral bands; obtaining the crop growth index of each image block by dividing the normalized vegetation index of each image block by the corresponding data variance mean. 3.The remote-sensing-based agricultural drought monitoring method according to claim 1, characterized in that, The farmland center direction of the relative position relationship between each image block and its surrounding blocks is determined according to the difference between the crop growth index of each image block and that of its surrounding blocks, and the method comprises: determining whether the crop growth index of the current image block is less than that of the surrounding blocks; if yes, determining the surrounding blocks greater than the current image block as regional directional blocks, and obtaining the farmland center direction of the current image block according to the summation calculation result of the first centroid position of all regional directional blocks and the second centroid position of the current image block; if no, determining the direction from the block centroid point adjacent to the current image block to the second centroid position of the current image block as the farmland center direction of the current image block.

4. The remote sensing technology-based agricultural drought monitoring method according to claim 1, characterized in that, The drought monitoring result of the farmland region is determined according to the crop water index of all image blocks, and the method comprises: inputting the crop water index of all image blocks into a drought analysis model, wherein the drought analysis model is a machine learning model trained based on hyperspectral data and a corresponding drought index; determining the drought monitoring result of the farmland region according to the output result of the drought analysis model.

5. A remote sensing technology-based agricultural drought monitoring system, characterized by, The system is applied to the monitoring method of any one of claims 1-4, and the system comprises: a data acquisition terminal configured to perform remote sensing data acquisition on a farmland region to be monitored for drought, and output an RGB image and hyperspectral data of the farmland region; The data processing terminal is connected to the data acquisition terminal, and is configured to perform image block segmentation according to a difference in image block features represented by the RGB image, and determine block spectral data of the hyperspectral data corresponding to each segmented image block; obtain a water content prompt parameter of the growing crop in each image block at different spectral bands according to a variation feature of the block spectral data of each image block at continuous spectral bands; perform data dimension reduction processing according to a parameter difference of the water content prompt parameters of each image block, so as to obtain a crop water index of each image block; and determine a drought monitoring result of the farmland region according to the crop water indexes of all image blocks. The method for obtaining the water content prompt parameter comprises: obtaining a crop growth index of each image block according to a crop growth state represented by the block spectral data of each image block; obtaining a water transfer feature of each image block according to a difference in the crop growth indexes of each image block and its surrounding blocks; obtaining a water content reference parameter of each image block according to a data difference of the block spectral data of each image block at different spectral bands, and combining the difference in the crop growth indexes of the block and other blocks; obtaining the water content prompt parameter of the corresponding image block according to the water transfer feature and the water content reference parameter of each image block; Calculating moisture content prompt parameters under each spectral band : ; u denotes the number of image blocks, crop growth index for the vth image block, water content reference parameter for the vth image block, water transfer characteristic for the vth image block, norm() is a normalization function; The method for obtaining the water transfer feature of each image block comprises: determining a farmland center direction of a relative position relationship between each image block and its surrounding blocks according to a difference in the crop growth indexes of each image block and its surrounding blocks; connecting an edge data point of a current image block and a block center of each image block in the farmland center direction to obtain a direction line segment representing a direction change of the water content on the current image block; obtaining the water transfer feature of each image block according to a difference between spectral data corresponding to the direction line segment of each image block at different spectral bands; The method for obtaining the water content reference parameter comprises: performing interpolation on the block spectral data of all blocks to fill in the block spectral data, and respectively calculating a spectral data mean square error between a current image block and other image blocks at different spectral bands, and an index difference value of the crop growth indexes; performing descending arrangement on all spectral data mean square errors and all index difference values of the current image block respectively to obtain a first sorting result and a second sorting result of the current image block; obtaining the water content reference parameter of the current image block according to a sorting difference degree between the first sorting result and the second sorting result; sequentially arrange the mean square error (MSE) of the spectral data and the exponential difference value of the crop growth index of the current image block and other image blocks in order from small to large; calculate the water content reference parameter of the current block in a certain dimension : ; in the above formula, u represents the number of image blocks, represents the sequence order of the spectral data mean square error MSE of the current block and the vth block, represents the sequence order of the exponential difference value of the current image block and the vth image block, tanh() is the hyperbolic tangent function, and c represents an adjustment coefficient, c=1. The method for performing data dimension reduction processing to obtain the crop water index of each image block comprises: performing data matrix construction on spectral bands and corresponding spectral data of the block spectral data of each image block to obtain an initial data matrix representing the block spectral data of each image block at each spectral band; performing mean value elimination processing on the initial data matrix of each image block at each spectral band to obtain a block data matrix of each image block at each spectral band; and The covariance matrix of each image block is obtained by weighted calculation according to all water content prompt parameters of the image block and block data matrices of all spectral bands; The crop water index of each image block is obtained by eigenvalue decomposition of the covariance matrix of each image block.

Images