Carbon flux monitoring method and system based on soil drought remote sensing data

Abstract

The invention relates to the technical field of carbon flux monitoring, and particularly discloses a carbon flux monitoring method and system based on soil drought remote sensing data, and the method comprises the steps: respectively collecting soil humidity remote sensing data and a soil vegetation coverage remote sensing image through a microwave sensor and an optical remote sensing sensor; performing resampling on the soil humidity remote sensing data and the soil vegetation coverage remote sensing image to realize spatial resolution alignment, and performing spatial distribution feature extraction on the soil humidity remote sensing data and the soil vegetation coverage remote sensing image after spatial alignment by adopting a deep learning algorithm; and pixel-level significance enhancement and fusion processing are carried out on the captured soil humidity distribution characteristics and vegetation distribution characteristics, so that nonlinear modeling of a dynamic influence mechanism of the soil drought degree on the vegetation growth state is realized, and intelligent estimation of the carbon flux is realized on the basis. In this way, the nonlinear regulation and control effect of soil drought stress on vegetation respiration and photosynthesis can be effectively described, and the efficiency and coverage range of carbon flux monitoring are improved.

Description

Technical Field

[0001] The present application relates to the technical field of carbon flux monitoring, and more specifically, to a carbon flux monitoring method and system based on soil drought remote sensing data. Background Art

[0002] Against the backdrop of global climate change, carbon cycle research has become a focus of scientific attention. As a key indicator of carbon exchange between ecosystems and the atmosphere, accurate monitoring of carbon flux is crucial for understanding the global carbon cycle, assessing ecosystem responses to climate change, and formulating effective climate policies. Traditional carbon flux measurement methods, such as the alkaline absorption method and the cover-and-cover method, have significant limitations. The alkaline absorption method requires a container attached to the soil surface, containing an open container containing an alkaline absorption solution. Changes in the pH of the solution over time are used to calculate soil CO2 release. The cover-and-cover method uses a plastic cover or container of appropriate size attached to the soil surface. The soil CO2 release is calculated based on the difference in CO2 concentration within the cover over a specified period. These methods are not only labor-intensive and time-consuming, but also applicable only to small-scale carbon flux measurements. Traditional methods are inefficient when applied to large monitoring areas and cannot meet the needs of real-time, comprehensive carbon flux monitoring.

[0003] With the continuous advancement of science and technology, remote sensing technology has provided a new approach for carbon flux monitoring due to its ability to achieve large-scale observations. Existing remote sensing carbon flux monitoring methods are mainly based on the inversion of vegetation photosynthesis parameters (such as leaf area index and photosynthetically active radiation absorption ratio) from optical remote sensing data (such as MODIS and Landsat), combined with meteorological data to drive light energy utilization models. However, such methods do not adequately describe the response mechanism of soil drought stress, especially ignoring the nonlinear regulatory effect of soil moisture on vegetation respiration and photosynthesis. Studies have shown that drought stress can cause a 30%-50% decrease in vegetation stomatal conductance, significantly affecting the net carbon exchange (NEE) of the ecosystem. Therefore, accurately sensing the degree of soil drought is a key link in improving the accuracy of carbon flux estimation.

[0004] However, existing studies have mostly isolated statistical correlations between soil moisture and vegetation growth, failing to establish a dynamic model of their interaction in carbon flux formation. This lacks the ability to adaptively model the dynamic response of carbon exchange processes under drought stress. For example, when soil moisture falls below the wilting point, stomata close, leading to a sharp decline in photosynthesis, which in turn affects carbon absorption and release. This modulation relationship has not been fully captured in existing carbon flux monitoring models.

[0005] Therefore, an optimized carbon flux monitoring method and system based on soil drought remote sensing data is expected. Summary of the Invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a carbon flux monitoring method and system based on soil drought remote sensing data, which utilizes microwave sensors and optical remote sensing sensors to respectively collect soil moisture remote sensing data and soil vegetation cover remote sensing images, and resamples the two to achieve spatial resolution alignment, and then uses deep learning-based data processing technology to extract spatial distribution features of the spatially aligned soil moisture remote sensing data and soil vegetation cover remote sensing images, and performs pixel-level significance enhancement and fusion processing on the captured soil moisture distribution features and vegetation distribution features to achieve nonlinear modeling of the dynamic impact mechanism of soil drought degree on vegetation growth state, thereby achieving intelligent estimation of carbon flux on this basis. In this way, the nonlinear regulatory effect of soil drought stress on vegetation respiration and photosynthesis can be effectively characterized, and the efficiency and coverage of carbon flux monitoring can be improved.

[0007] According to one aspect of the present application, a carbon flux monitoring method based on soil drought remote sensing data is provided, which comprises: Microwave sensors are used to collect soil moisture remote sensing data, while optical remote sensing sensors are used to collect soil vegetation cover remote sensing images; Extracting remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain a soil moisture distribution feature map and a soil vegetation distribution feature map, wherein extracting remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing image comprises: performing resolution alignment and remote sensing feature extraction on the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain the soil moisture distribution feature map and the soil vegetation distribution feature map; Performing feature enhancement based on dynamic receptive field anchoring on the soil moisture distribution feature map and the soil vegetation distribution feature map respectively to obtain a significant enhanced soil moisture distribution feature map and a significant enhanced soil vegetation distribution feature map; Carbon flux prediction is performed based on the fusion features of the significant enhanced soil moisture distribution characteristic map and the significant enhanced soil vegetation distribution characteristic map to obtain a carbon flux decoding value.

[0008] According to another aspect of the present application, a carbon flux monitoring system based on soil drought remote sensing data is provided, comprising: Remote sensing data acquisition module, used to collect soil moisture remote sensing data using microwave sensors, and simultaneously collect soil vegetation cover remote sensing images using optical remote sensing sensors; A remote sensing feature extraction module is configured to extract remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain a soil moisture distribution feature map and a soil vegetation distribution feature map, wherein extracting remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing image comprises: performing resolution alignment and remote sensing feature extraction on the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain the soil moisture distribution feature map and the soil vegetation distribution feature map; a dynamic feature enhancement module, configured to perform feature enhancement based on dynamic receptive field anchoring on the soil moisture distribution feature map and the soil vegetation distribution feature map to obtain a significant enhanced soil moisture distribution feature map and a significant enhanced soil vegetation distribution feature map; The carbon flux prediction module is used to perform carbon flux prediction based on the fusion characteristics of the significant enhanced soil moisture distribution characteristic map and the significant enhanced soil vegetation distribution characteristic map to obtain a carbon flux decoding value.

[0009] Compared with the existing technology, the carbon flux monitoring method and system based on soil drought remote sensing data provided by this application uses microwave sensors and optical remote sensing sensors to collect soil moisture remote sensing data and soil vegetation cover remote sensing images respectively, and resamples the two to achieve spatial resolution alignment, and then uses deep learning-based data processing technology to extract spatial distribution features of the spatially aligned soil moisture remote sensing data and soil vegetation cover remote sensing images, and performs pixel-level significance enhancement and fusion processing on the captured soil moisture distribution features and vegetation distribution features to achieve nonlinear modeling of the dynamic impact mechanism of soil drought degree on vegetation growth state, thereby achieving intelligent estimation of carbon flux on this basis. In this way, the nonlinear regulatory effect of soil drought stress on vegetation respiration and photosynthesis can be effectively portrayed, and the efficiency and coverage of carbon flux monitoring can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 Flowchart of a carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application.

[0012] Figure 2 Schematic diagram of data flow of a carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application.

[0013] Figure 3 4 is a flowchart of sub-step S2 of the carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application.

[0014] Figure 4 This is a flowchart of sub-step S3 of the carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application.

[0015] Figure 5 This is a flowchart of sub-step S33 of the carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application.

[0016] Figure 6 This is a flowchart of sub-step S4 of the carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application.

[0017] Figure 7 4 is a block diagram of a carbon flux monitoring system based on soil drought remote sensing data according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0019] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0020] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0021] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0022] It is worth noting that in this application, all actions to obtain data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0023] In response to the technical problems described in the above background technology, this application proposes a carbon flux monitoring method based on soil drought remote sensing data, which uses microwave sensors and optical remote sensing sensors to collect soil moisture remote sensing data and soil vegetation cover remote sensing images respectively, and resamples the two to achieve spatial resolution alignment, and then uses deep learning-based data processing technology to extract spatial distribution features of the spatially aligned soil moisture remote sensing data and soil vegetation cover remote sensing images, and performs pixel-level significance enhancement and fusion processing on the captured soil moisture distribution features and vegetation distribution features to achieve nonlinear modeling of the dynamic impact mechanism of soil drought degree on vegetation growth state, thereby achieving intelligent estimation of carbon flux on this basis. In this way, the nonlinear regulatory effect of soil drought stress on vegetation respiration and photosynthesis can be effectively portrayed, and the efficiency and coverage of carbon flux monitoring can be improved.

[0024] Figure 1 Flowchart of a carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the carbon flux monitoring method based on soil drought remote sensing data includes the following steps: S1, using a microwave sensor to collect soil moisture remote sensing data, and simultaneously using an optical remote sensing sensor to collect soil vegetation cover remote sensing images; S2, extracting remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing images to obtain a soil moisture distribution feature map and a soil vegetation distribution feature map, wherein extracting remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing images includes: performing resolution alignment and remote sensing feature extraction on the soil moisture remote sensing data and the soil vegetation cover remote sensing images to obtain the soil moisture distribution feature map and the soil vegetation distribution feature map; S3, performing feature enhancement based on dynamic receptive field anchoring on the soil moisture distribution feature map and the soil vegetation distribution feature map to obtain a significant enhanced soil moisture distribution feature map and a significant enhanced soil vegetation distribution feature map; S4, performing carbon flux prediction based on the fusion features of the significant enhanced soil moisture distribution feature map and the significant enhanced soil vegetation distribution feature map to obtain a carbon flux decoding value.

[0025] In the aforementioned carbon flux monitoring method based on soil drought remote sensing data, step S1 involves collecting soil moisture remote sensing data using microwave sensors, while simultaneously collecting soil vegetation cover remote sensing images using optical remote sensing sensors. It should be understood that existing carbon flux monitoring methods often rely solely on optical remote sensing data, making it difficult to directly perceive the stressful effects of soil moisture dynamics on vegetation physiological processes. Microwave remote sensing sensors detect soil moisture based on the property that the soil dielectric constant changes with soil moisture when microwaves interact with the soil medium. Higher soil moisture increases the dielectric constant, which in turn alters microwave reflection and scattering properties. This inversion yields soil moisture information, generating spatially gridded soil moisture remote sensing data. Optical remote sensing sensors capture color and texture changes in the vegetation canopy using red, green, and blue wavelengths, effectively revealing the physiological state of vegetation and accurately depicting the spatial heterogeneity of vegetation cover. By combining soil moisture remote sensing data with soil vegetation cover remote sensing images, this application enables bidirectional observation of both the water supply side (soil) and the physiological response side (vegetation), providing a data foundation for constructing drought-photosynthesis coupling mechanisms.

[0026] Specifically, when the soil contains a higher proportion of water, its dielectric constant increases, causing microwave signals to produce different reflection and scattering effects when penetrating the soil. By analyzing the changing patterns of these signals, the spatial distribution of soil moisture can be inverted. Modern microwave remote sensing technology usually uses L-band or C-band for observation. Among them, the L-band is more suitable for areas with thick vegetation cover due to its strong penetration ability. To ensure that the data obtained has high accuracy and resolution, microwave sensors on satellite platforms are usually equipped with advanced calibration systems that can automatically adjust parameters to meet the needs of different observation conditions. In addition, during the data processing stage, complex inversion algorithms need to be applied to convert the original microwave signals into specific soil moisture values, and then map them into the geospatial coordinate system to form a continuous soil moisture distribution map.

[0027] Meanwhile, optical remote sensing sensors are primarily used to capture information on surface vegetation cover and its health. By measuring the reflectance of vegetation canopies in multiple wavelengths—red, green, and blue—key physiological states such as vegetation growth and chlorophyll content can be revealed, which is crucial for understanding how vegetation responds to drought stress. In the visible light band, healthy vegetation typically exhibits a high green reflectance, while drought-stressed vegetation may show reduced reflectance, particularly in the red band, due to enhanced red light absorption by chlorophyll.

[0028] During implementation, the temporal and spatial consistency between the two types of data must be fully considered. On the one hand, due to the limitations of satellite orbital periods, it may not be possible to collect all necessary data at the same time. On the other hand, there may be differences in spatial resolution between different sensors, which also need to be addressed through subsequent data processing steps. To overcome these problems, a multi-source data fusion strategy is often adopted. This involves registering and resampling data from different sensors during the data preprocessing phase to ensure that they can be compared and analyzed within the same geographic reference frame. Especially when processing large-scale areas, multi-source data fusion strategies can help eliminate errors caused by factors such as terrain undulations and cloud cover.

[0029] Furthermore, special attention must be paid to the impact of atmospheric conditions on observational results. Both microwave and optical remote sensing data are subject to interference from atmospheric components such as water vapor and aerosols. To address this challenge, scientists have developed a variety of atmospheric correction algorithms designed to minimize the negative impact of these factors on the final results. For example, synchronized atmospheric profile data allows for precise atmospheric correction of microwave signals. In optical remote sensing, atmospheric radiation transfer models can be established to simulate and remove atmospheric scattering and absorption effects.

[0030] In the above-mentioned carbon flux monitoring method based on soil drought remote sensing data, the step S2 extracts remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain a soil moisture distribution feature map and a soil vegetation distribution feature map, wherein extracting remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing image includes: performing resolution alignment and remote sensing feature extraction on the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain the soil moisture distribution feature map and the soil vegetation distribution feature map. Figure 3 FIG. 1 is a flow chart of sub-step S2 of the carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application. Figure 3 As shown, the step S2 includes the steps of: S21, resampling the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain spatially aligned soil moisture remote sensing data and spatially aligned soil vegetation cover remote sensing image; S22, performing remote sensing feature extraction based on the SwinTransformer model on the spatially aligned soil moisture remote sensing data and the spatially aligned soil vegetation cover remote sensing image to obtain the soil moisture distribution feature map and the soil vegetation distribution feature map.

[0031] Specifically, the step S21 resamples the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain spatially aligned soil moisture remote sensing data and spatially aligned soil vegetation cover remote sensing images. Specifically, due to the large difference in resolution between microwave data and visible light data, directly performing joint data analysis will lead to information dislocation, affecting the subsequent feature extraction and carbon flux estimation accuracy. Therefore, the present application further adopts an appropriate resampling method, such as bilinear interpolation or nearest neighbor interpolation, to perform spatial resolution alignment on the soil moisture remote sensing data and the soil vegetation cover remote sensing image to ensure the consistency of the two in spatial position, thereby providing a spatially corresponding data basis for the subsequent joint analysis of soil drought degree and vegetation growth conditions.

[0032] In the specific implementation process, the resampling process is essentially to redistribute the pixel values ​​in the original dataset in order to generate a new dataset with uniform spatial resolution. In this process, choosing an appropriate resampling algorithm is crucial. Common resampling methods include bilinear interpolation and nearest neighbor interpolation. Bilinear interpolation calculates a weighted average as the value of the new pixel by considering the values ​​and relative positions of the four nearest neighboring pixels around the target pixel. This method can effectively reduce image distortion and maintain the smoothness of the image. It is particularly suitable for situations where it is necessary to accurately reflect the continuous changes in the surface characteristics, such as soil moisture distribution. In contrast, nearest neighbor interpolation sets the value of each new pixel to the value of the nearest original pixel. Although this simple and direct method may cause jagged effects on the edges of the image, it is still of great value when processing discrete data or when you want to maintain the accuracy of the original data.

[0033] In order to meet the spatial resolution alignment requirements for soil moisture remote sensing data and soil vegetation cover remote sensing images, it is first necessary to clarify the initial resolution of both and the desired target resolution. For soil moisture remote sensing data, given its low initial resolution, it may need to be upscaled to match the higher resolution standards of optical remote sensing images. In this process, bilinear interpolation can be used to effectively fill the gaps caused by upscaling, so that the newly generated soil moisture data retains the core features of the original information while matching the vegetation cover image in detail. In specific operations, the geometric relationship between the target pixel and its four adjacent original pixels can be calculated to determine their respective weight coefficients, and then the weighted average value can be obtained as the new value of the target pixel. This process not only helps to improve the spatial resolution of soil moisture data, but also ensures a smooth transition between data and avoids mutations.

[0034] At the same time, for remote sensing images of soil and vegetation cover, if their resolution is higher than the required standard, it may be necessary to reduce the size of the image. In this case, nearest neighbor interpolation becomes an ideal choice. By selecting the nearest original pixel value corresponding to each new pixel, the resolution adjustment can be completed quickly without changing the original information. It is worth noting that although this method may cause the edges of the image to be not smooth enough, it is sufficient to meet the requirements for tasks such as vegetation classification that do not require extremely high precision. In addition, in some specific applications, bilinear interpolation and nearest neighbor interpolation can also be used in combination, that is, while keeping the main structure unchanged, some smoothing elements are appropriately introduced to optimize the final result.

[0035] In addition to the basic resampling techniques described above, higher-level algorithms can be incorporated depending on the specific situation, such as resampling methods based on wavelet transforms. Wavelet transforms provide information in both the time and frequency domains, making them ideal for processing remote sensing data with distinct multi-scale features. By decomposing the raw data into wavelet coefficients at different scales and then recombining these coefficients according to the target resolution, high-quality spatial alignment can be achieved. This method not only effectively improves the spatial resolution of the data but also preserves more detailed information, making it particularly suitable for analyzing soil moisture and vegetation cover in complex terrain conditions.

[0036] Another factor that cannot be ignored in the entire resampling process is the consistency of the geographic reference frame. In order to ensure that all data can be compared and analyzed in the same geographic coordinate system, the original data must be strictly aligned before resampling begins. This involves identifying and correcting displacement errors caused by factors such as satellite orbit deviation and terrain undulation. Modern GIS software provides a wealth of tools to support this process, such as automated control point matching functions, which can quickly find point pairs in different datasets corresponding to the same geographic location and make precise position adjustments accordingly. In addition, it is also necessary to pay attention to the possible time delay between different sensors, and try to choose datasets that are acquired synchronously or nearly synchronously for processing to reduce observation differences caused by time differences.

[0037] In practice, resampling tasks often rely on specialized remote sensing image processing platforms. These platforms offer a variety of built-in resampling algorithms and allow users to flexibly configure parameters based on specific needs. For example, when performing bilinear interpolation, the interpolation window size can be adjusted based on data characteristics and desired results, balancing computational efficiency with output quality. Similarly, when applying nearest neighbor interpolation, consideration should be given to whether special processing is required for boundary regions to avoid unnecessary errors.

[0038] Specifically, in step S22, the spatially aligned soil moisture remote sensing data and the spatially aligned soil vegetation cover remote sensing images are respectively subjected to remote sensing feature extraction based on the Swin Transformer model to obtain the soil moisture distribution feature map and the soil vegetation distribution feature map. It should be understood that since traditional carbon flux monitoring methods mostly analyze the statistical correlation between soil moisture and vegetation growth in isolation, it is difficult to capture the nonlinear effects of soil drought stress on vegetation physiological processes. To this end, the present application adopts the Swin Transformer model to perform feature extraction on the spatially aligned soil moisture remote sensing data and the spatially aligned soil vegetation cover remote sensing images, respectively, to explore the spatial distribution pattern of soil moisture and the spatial structural characteristics of vegetation cover. Specifically, the Swin Transformer model, as a deep learning model based on the self-attention mechanism, can effectively capture long-distance dependencies in images and has a strong feature representation capability. In this application, the Swin Transformer model uses the self-attention mechanism to capture the long-range dependencies within spatially aligned soil moisture remote sensing data and spatially aligned soil vegetation cover remote sensing images. At the same time, the sliding window strategy is used to reduce computational complexity and improve the model's learning efficiency for soil moisture and vegetation cover characteristics. It can effectively capture the spatial distribution patterns of soil moisture and vegetation growth, generate soil moisture distribution feature maps and soil vegetation distribution feature maps, thereby revealing the spatial heterogeneity of soil moisture and vegetation growth, and providing effective feature input for subsequent nonlinear correlation response modeling between soil drought degree and vegetation growth status.

[0039] In the above-mentioned carbon flux monitoring method based on soil drought remote sensing data, the step S3 performs feature enhancement based on dynamic receptive field anchoring on the soil moisture distribution characteristic map and the soil vegetation distribution characteristic map to obtain a significant enhanced soil moisture distribution characteristic map and a significant enhanced soil vegetation distribution characteristic map. It should be understood that the present application takes into account the spatial non-uniformity of vegetation response to soil drought, that is, the degree of soil drought in different local areas is inconsistent, which leads to differences in the physiological response of vegetation. Therefore, in order to further dynamically focus on key areas, the present application proposes a feature enhancement method based on dynamic receptive field anchoring, which dynamically adjusts the receptive field size and performs pixel-level feature enhancement processing on the soil moisture distribution characteristic map and the soil vegetation distribution characteristic map to enhance the ability to capture the soil moisture distribution spatial gradient mutation area and the vegetation physiological response sensitive area, thereby generating a significant enhanced soil moisture distribution characteristic map and a significant enhanced soil vegetation distribution characteristic map. Below, the feature enhancement process of the soil moisture distribution characteristic map is described in detail as an example.

[0040] Figure 4FIG. 1 is a flow chart of sub-step S3 of the carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application. Figure 4 As shown, the step S3 includes the steps of: S31, extracting the channel feature vector of the (i, j)th pixel position in the soil moisture distribution feature map as the soil moisture distribution pixel-level feature vector to be enhanced; S32, based on the characteristic distribution of the soil moisture distribution pixel-level feature vector to be enhanced, performing local receptive field dynamic anchoring on the soil moisture distribution feature map to screen out a set of soil moisture distribution pixel-level feature vectors within the local receptive field; S33, based on the set of soil moisture distribution pixel-level feature vectors within the local receptive field, performing significance enhancement on the soil moisture distribution pixel-level feature vector to be enhanced to obtain an enhanced soil moisture distribution pixel-level feature vector, wherein the enhanced soil moisture distribution pixel-level feature vector is the channel feature vector of the (i, j)th pixel position of the significance enhanced soil moisture distribution feature map.

[0041] Specifically, in a specific example of the present application, step S31 includes: first, performing feature decoupling on the soil moisture distribution feature map along the channel dimension to obtain a set of soil moisture distribution pixel-level feature vectors, which is expressed as: in, represents the set of real numbers, 、 and Respectively represent the height, width and number of channels of the soil moisture distribution characteristic map, Represents the soil moisture distribution characteristic map, 、 、 and They represent the (1,1), (1, ), No. ( ,1) and ( , ) pixel-level feature vector of soil moisture distribution at the pixel location, represents the feature decoupling function.

[0042] That is, by decoupling features along the channel dimension, the multi-channel feature vectors at each pixel position can be independently analyzed, breaking the implicit associations between channels, thereby eliminating the interference of channel coupling on subsequent pixel-level feature analysis, and providing independent, fine-grained pixel-level feature inputs for dynamic receptive field anchoring. This allows the feature vector of each pixel to be adaptively screened for receptive fields based on its own distribution characteristics, rather than being restricted by the global statistical characteristics of inter-channel mixing. Based on this, the generated set of pixel-level feature vectors of soil moisture distribution not only retains the spatial heterogeneity information of soil moisture distribution in the original remote sensing data, but also enhances the interpretability of features through decoupling of the channel dimension, allowing subsequent feature processing to more accurately capture the dynamic interactive relationship between soil moisture and vegetation cover at the local spatial scale, thereby laying a data foundation for modeling the nonlinear regulatory effect of soil-vegetation in carbon flux prediction.

[0043] Then, the soil moisture distribution pixel-level feature vector at the (i, j)th pixel position is extracted from the set of the soil moisture distribution pixel-level feature vectors as the soil moisture distribution pixel-level feature vector to be enhanced, which is expressed as follows: in, Indicates the soil moisture distribution characteristic diagram ( , ) pixel-level feature vector of soil moisture distribution at the pixel location, Represents the pixel-level feature vector to be enhanced for soil moisture distribution.

[0044] Specifically, by extracting the pixel-level feature vector of soil moisture distribution at a specific coordinate (i, j) as the anchor point to be enhanced, global remote sensing feature processing can be transformed into fine-grained pixel-level analysis, thereby accurately locating the local core of drought stress. Specifically, by using the extracted pixel-level feature vector of soil moisture distribution to be enhanced as the feature benchmark for local feature aggregation, the spatial weights of the surrounding feature vectors can be adaptively adjusted, thereby capturing the nonlinear coupling relationship of soil moisture distribution at the microscopic spatial scale. This allows the model to independently analyze the local effect pattern of soil drought at each location, thereby accurately expressing the dynamic response characteristics caused by spatial heterogeneity in the soil-vegetation interaction process during the carbon flux decoding stage.

[0045] Specifically, in a specific example of the present application, step S32 includes: first, compressing the soil moisture distribution pixel-level feature vector to be enhanced to obtain a soil moisture distribution pixel-level feature distillation encoding vector to be enhanced, which is expressed as: in, represents the calculation norm, Represents the distilled encoding vector of the pixel-level features to be enhanced in soil moisture distribution.

[0046] Specifically, through information compression, the high-dimensional pixel-level soil moisture distribution feature vectors to be enhanced are distilled into low-dimensional but information-dense distilled encoding vectors of the soil moisture distribution pixel-level features to be enhanced. This removes noise or redundant components irrelevant to the dynamic response of soil drought while retaining core features closely related to carbon flux formation, thus enabling dynamic receptive field size adaptability. By focusing on key features, the compressed pixel-level distilled encoding vectors of the soil moisture distribution pixel-level features to be enhanced more clearly reflect the correlation between the target pixel and its surrounding area in terms of drought response, thus providing interpretable feature information for subsequent dynamic adjustment of the receptive field size. This knowledge distillation-based information purification method not only enhances the efficiency of feature representation in representing soil drought dynamics but also improves the structural separability of the feature space by removing redundant information. This enables subsequent receptive field anchoring to accurately capture the local regulatory relationship of soil moisture based on the compact distilled features, thereby providing high signal-to-noise ratio feature input for modeling nonlinear soil-vegetation interactions in carbon flux prediction.

[0047] Then, based on the feature distribution spatial structure characteristics of the distilled coding vector of the soil moisture distribution pixel-level feature to be enhanced, the size of the feature receptive field of the distilled coding vector of the soil moisture distribution pixel-level feature to be enhanced is determined, which can be expressed as follows: in, represents the logarithmic function with base 2, Represents the half-side length of the feature receptive field of the distilled encoding vector of the pixel-level features to be enhanced in soil moisture distribution.

[0048] That is, by utilizing the spatial structural characteristics of the feature distribution implicit in the pixel-level to-be-enhanced feature distillation coding vector of soil moisture distribution and dynamically mapping the receptive field size, the physical diffusion mechanism of drought stress can be converted into geometric constraints in the feature space, and the key action range of the local soil moisture area represented by the pixel-level to-be-enhanced feature distillation coding vector of soil moisture distribution can be accurately identified, so that the coverage of the receptive field matches the actual scope of the dynamic response of soil drought, thereby significantly improving the spatial generalization ability of the model for the nonlinear relationship between drought and vegetation.

[0049] Finally, based on the size of the feature receptive field, the set of pixel-level feature vectors of soil moisture distribution within the local receptive field is filtered out from the set of pixel-level feature vectors of soil moisture distribution, which can be expressed as follows: in, 、 、 and They represent the first ( , ), No. ( , ), No. ( , ) and ( , ) location, the pixel-level feature vector of soil moisture distribution, Represents the set of pixel-level feature vectors of soil moisture distribution within the local receptive field.

[0050] That is, by screening the pixel-level feature vectors of soil moisture distribution within the local receptive field according to the dynamically determined receptive field size, and constructing a local context feature library centered on the target pixel, the model can flexibly integrate multi-scale context information for different drought scenarios, thereby accurately characterizing the modulation effect of soil moisture spatial heterogeneity on vegetation photosynthesis-respiration balance in the carbon flux decoding stage, and significantly improving the modeling accuracy of the spatial propagation process of drought stress.

[0051] Figure 5 FIG. 1 is a flow chart of sub-step S33 of the carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application. Figure 5 As shown, the step S33 includes the steps of: S331, based on the feature correlation of each local receptive field pixel-level feature vector of the soil moisture distribution in the local receptive field relative to the soil moisture distribution pixel-level feature vector to be enhanced, performing significance aggregation on the set of the local receptive field pixel-level feature vectors of the soil moisture distribution to obtain a local receptive field pixel-level feature significance aggregation coding vector; S332, fusing the local receptive field pixel-level feature significance aggregation coding vector and the soil moisture distribution pixel-level feature vector to be enhanced to obtain the enhanced soil moisture distribution pixel-level feature vector.

[0052] More specifically, step S331 is expressed as follows: in, The first pixel in the set of feature vectors representing the soil moisture distribution in the local receptive field is ( , ) position in the local receptive field of the soil moisture distribution pixel-level feature vector, represents the transpose of a vector, represents vector multiplication, represents the softmax normalization function, express The significance weight of Represents the pixel-level feature saliency aggregation encoding vector of soil moisture distribution within the local receptive field.

[0053] Specifically, by calculating the correlation between the pixel-level feature vectors of soil moisture distribution within each local receptive field and the pixel-level feature vector to be enhanced for the central pixel, the contribution of neighborhood features to the drought response of the central pixel can be quantified, thereby screening key contextual features with high correlation. The resulting saliency aggregation coding vector of pixel-level features of soil moisture distribution within the local receptive field can amplify spatial patterns strongly correlated with drought stress, weaken irrelevant noise, and aggregate highly correlated gradient features, enabling the model to more accurately characterize the spatial distribution of soil drought stress and its modulation of vegetation carbon exchange, thereby improving the robustness and accuracy of modeling spatial heterogeneity of drought stress in carbon flux prediction.

[0054] More specifically, in a specific example of the present application, the step S332 includes: weighted fusion of the soil moisture distribution pixel-level feature significance aggregation coding vector within the local receptive field and the soil moisture distribution pixel-level feature vector to be enhanced to obtain the enhanced soil moisture distribution pixel-level feature vector.

[0055] In particular, in a preferred example of the present application, step S332 includes: first, based on the conformal feature fusion commutation constraint of the set of pixel-level feature vectors of the soil moisture distribution in the local receptive field, modulating the weighting coefficients of the pixel-level feature saliency aggregation coding vector of the soil moisture distribution in the local receptive field and the pixel-level feature vector to be enhanced of the soil moisture distribution, which can be expressed as follows: in, represents the sine function, Represents the spatial representation vector of the significant fusion volume of pixel-level features of soil moisture distribution in the local receptive field, Represents the significant fusion boundary representation vector of the pixel-level features of soil moisture distribution in the local receptive field, and They respectively represent the modulated weighted coefficients of the pixel-level feature saliency aggregation coding vector of the soil moisture distribution in the local receptive field and the pixel-level feature vector to be enhanced of the soil moisture distribution, represents the proportional scaling factor, Indicates the calculation of the bi-norm.

[0056] Specifically, by introducing the commutative difference norm constraint between the volumetric representation vector of the significant fusion of pixel-level soil moisture distribution features within the local receptive field and the boundary representation vector of the significant fusion of pixel-level soil moisture distribution features within the local receptive field, the regularity condition is enforced during weighted fusion between the two, eliminating fusion noise caused by non-conformal volume-edge features. In this way, by dynamically adjusting the weighting coefficients, the modulated weighting coefficients generated can enhance the model's ability to depict nonlinear coupling relationships by constraining the conformal commutativity of volume-edge features, significantly improving the accuracy and robustness of modeling the dynamics of drought-vegetation interactions in carbon flux prediction.

[0057] Then, based on the weighting coefficient, the soil moisture distribution pixel-level feature saliency aggregation coding vector within the local receptive field and the soil moisture distribution pixel-level feature vector to be enhanced are weightedly fused to obtain the enhanced soil moisture distribution pixel-level feature vector, which is expressed as follows: in, Represents the pixel-level feature vector of enhanced soil moisture distribution.

[0058] In other words, weighted fusion not only preserves the drought stress intensity of the central pixel but also introduces significant information about the coordinated changes in the neighborhood, thus overcoming the spatial limitations of isolated feature modeling. Specifically, the pixel-level feature vectors of enhanced soil moisture distribution generated by feature fusion can uniformly represent the local intensity and spatial dynamics of drought stress, significantly improving the model's accuracy in modeling the spatiotemporal heterogeneity of carbon flux and enhancing its ability to model the dynamic responses of vegetation photosynthesis and respiration.

[0059] In the above-mentioned carbon flux monitoring method based on soil drought remote sensing data, the step S4 performs carbon flux prediction based on the fusion features of the significant enhanced soil moisture distribution characteristic map and the significant enhanced soil vegetation distribution characteristic map to obtain a carbon flux decoding value. Figure 6 FIG4 is a flow chart of sub-step S4 of the carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application. Figure 6 As shown, the step S4 includes the following steps: S41, performing cascade fusion processing on the significance enhanced soil moisture distribution feature map and the significance enhanced soil vegetation distribution feature map along the channel dimension to obtain a soil moisture-vegetation distribution fusion coding feature map; S42, inputting the soil moisture-vegetation distribution fusion coding feature map into a codec-based carbon flux prediction module to obtain the carbon flux decoding value.

[0060] Specifically, in step S41, the significant enhanced soil moisture distribution feature map and the significant enhanced soil vegetation distribution feature map are subjected to cascade fusion processing along the channel dimension to obtain a soil moisture-vegetation distribution fusion coding feature map. It should be understood that the impact of soil drought stress on carbon flux is manifested as a nonlinear coupling between soil moisture and vegetation state. Therefore, the present application further performs cascade processing along the channel dimension on the significant enhanced soil moisture distribution feature map and the significant enhanced soil vegetation distribution feature map to model the coupling relationship between the two through cross-modal feature interaction, capture the nonlinear mapping relationship between the spatial difference of vegetation growth state and the spatial distribution pattern of soil moisture, and thus generate a soil moisture-vegetation distribution fusion coding feature map. The soil moisture-vegetation distribution fusion coding feature map not only integrates the spatial distribution information of soil moisture, but also embeds the spatial structural characteristics of vegetation physiological response, providing a comprehensive feature characterization for subsequent carbon flux estimation based on drought stress.

[0061] Specifically, in step S42, the soil moisture-vegetation distribution fusion coding feature map is input into the carbon flux prediction module based on the codec to obtain the carbon flux decoding value. Specifically, the carbon flux prediction module based on the codec is composed of an encoder part and a decoder part. Among them, the encoder part is used to perform deep feature extraction on the input soil moisture-vegetation distribution fusion coding feature map, and to explore the deep coupling relationship between soil moisture and vegetation growth status. The decoder part predicts the specific value of the carbon flux based on the deep features extracted by the encoder. During the training process, by introducing real carbon flux observation data as labels, the carbon flux prediction module based on the codec is trained using a supervised learning method, so that it learns the mapping relationship from soil moisture and vegetation growth status characteristics to carbon flux, and through the joint optimization of the codec structure, it is possible to achieve effective modeling of the complex relationship between soil drought degree and vegetation growth status. After the training is completed, the module can be used to perform carbon flux prediction on the input soil moisture-vegetation distribution fusion coding feature map to obtain a carbon flux decoding value. In this way, it is possible to estimate vegetation carbon flux under drought stress, providing strong support for agricultural water resources management and ecosystem carbon cycle research.

[0062] In summary, a carbon flux monitoring method based on soil drought remote sensing data according to an embodiment of the present application is explained, which utilizes microwave sensors and optical remote sensing sensors to respectively collect soil moisture remote sensing data and soil vegetation cover remote sensing images, and resamples the two to achieve spatial resolution alignment, and then uses deep learning-based data processing technology to extract spatial distribution features of the spatially aligned soil moisture remote sensing data and soil vegetation cover remote sensing images, and performs pixel-level significance enhancement and fusion processing on the captured soil moisture distribution features and vegetation distribution features to achieve nonlinear modeling of the dynamic impact mechanism of soil drought degree on vegetation growth state, thereby achieving intelligent estimation of carbon flux on this basis. In this way, the nonlinear regulatory effect of soil drought stress on vegetation respiration and photosynthesis can be effectively portrayed, thereby improving the efficiency and coverage of carbon flux monitoring.

[0063] Furthermore, a carbon flux monitoring system based on soil drought remote sensing data is also provided.

[0064] Figure 7 FIG is a block diagram of a carbon flux monitoring system based on soil drought remote sensing data according to an embodiment of the present application. Figure 7 As shown, according to an embodiment of the present application, a carbon flux monitoring system 100 based on soil drought remote sensing data includes: a remote sensing data acquisition module 110 for acquiring soil moisture remote sensing data using a microwave sensor, and simultaneously acquiring soil vegetation cover remote sensing images using an optical remote sensing sensor; a remote sensing feature extraction module 120 for extracting remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing images to obtain a soil moisture distribution feature map and a soil vegetation distribution feature map, wherein extracting remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing images includes: extracting remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing images; ... The soil and vegetation cover remote sensing image is subjected to resolution alignment and remote sensing feature extraction to obtain the soil moisture distribution feature map and the soil vegetation distribution feature map; a dynamic feature enhancement module 130 is used to perform feature enhancement based on dynamic receptive field anchoring on the soil moisture distribution feature map and the soil vegetation distribution feature map to obtain a significant enhanced soil moisture distribution feature map and a significant enhanced soil vegetation distribution feature map; a carbon flux prediction module 140 is used to perform carbon flux prediction based on the fusion features of the significant enhanced soil moisture distribution feature map and the significant enhanced soil vegetation distribution feature map to obtain a carbon flux decoding value.

[0065] Here, those skilled in the art will appreciate that the specific operations of each module in the carbon flux monitoring system based on soil drought remote sensing data have been described above with reference to Figures 1 to 6 The description of the carbon flux monitoring method based on soil drought remote sensing data has been introduced in detail, and therefore, its repeated description will be omitted.

[0066] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in the present invention are merely illustrative and non-limiting, and should not be construed as necessarily possessed by each embodiment of the present invention. Furthermore, the specific details of the above embodiments are provided for illustrative purposes and to facilitate understanding, and are not intended to be limiting. These details do not necessarily limit the present invention to being implemented using these specific details.

[0067] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0068] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be encompassed therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.

[0069] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0070] Finally, it should be noted that the above description has been provided for purposes of illustration and description. Furthermore, the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to be limiting. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art will appreciate that the technical solutions of the present invention may be modified or replaced with equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A carbon flux monitoring method based on soil drought remote sensing data, characterized in that: include: Microwave sensors are used to collect soil moisture remote sensing data, while optical remote sensing sensors are used to collect soil vegetation cover remote sensing images; Extracting remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain a soil moisture distribution feature map and a soil vegetation distribution feature map, wherein extracting remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing image comprises: performing resolution alignment and remote sensing feature extraction on the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain the soil moisture distribution feature map and the soil vegetation distribution feature map; Performing feature enhancement based on dynamic receptive field anchoring on the soil moisture distribution feature map and the soil vegetation distribution feature map respectively to obtain a significant enhanced soil moisture distribution feature map and a significant enhanced soil vegetation distribution feature map; Carbon flux prediction is performed based on the fusion features of the significant enhanced soil moisture distribution characteristic map and the significant enhanced soil vegetation distribution characteristic map to obtain a carbon flux decoding value.

2. The carbon flux monitoring method based on soil drought remote sensing data according to claim 1, characterized in that: Performing resolution alignment and remote sensing feature extraction on the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain the soil moisture distribution feature map and the soil vegetation distribution feature map, including: Resampling the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain spatially aligned soil moisture remote sensing data and spatially aligned soil vegetation cover remote sensing image; The spatially aligned soil moisture remote sensing data and the spatially aligned soil vegetation cover remote sensing image are respectively subjected to remote sensing feature extraction based on a Swin Transformer model to obtain the soil moisture distribution feature map and the soil vegetation distribution feature map.

3. The carbon flux monitoring method based on soil drought remote sensing data according to claim 2, characterized in that: The soil moisture distribution feature map and the soil vegetation distribution feature map are respectively subjected to feature enhancement based on dynamic receptive field anchoring to obtain a significant enhanced soil moisture distribution feature map and a significant enhanced soil vegetation distribution feature map, including: Extracting the channel feature vector of the (i, j)th pixel position in the soil moisture distribution feature map as the soil moisture distribution pixel-level feature vector to be enhanced; Based on the characteristic distribution of the soil moisture distribution pixel-level feature vector to be enhanced, the soil moisture distribution feature map is dynamically anchored in a local receptive field to screen out a set of soil moisture distribution pixel-level feature vectors within the local receptive field; Based on the set of pixel-level feature vectors of the soil moisture distribution within the local receptive field, the soil moisture distribution pixel-level feature vector to be enhanced is significantly enhanced to obtain an enhanced soil moisture distribution pixel-level feature vector, wherein the enhanced soil moisture distribution pixel-level feature vector is the channel feature vector of the (i, j)th pixel position of the significant enhanced soil moisture distribution feature map.

4. The carbon flux monitoring method based on soil drought remote sensing data according to claim 3 is characterized in that: Extracting the channel feature vector of the (i, j)th pixel position in the soil moisture distribution feature map as the soil moisture distribution pixel-level feature vector to be enhanced includes: Performing feature decoupling on the soil moisture distribution feature map along the channel dimension to obtain a set of soil moisture distribution pixel-level feature vectors; The soil moisture distribution pixel-level feature vector at the (i, j)th pixel position is extracted from the set of the soil moisture distribution pixel-level feature vectors as the soil moisture distribution pixel-level feature vector to be enhanced.

5. The carbon flux monitoring method based on soil drought remote sensing data according to claim 4 is characterized in that: Based on the characteristic distribution of the soil moisture distribution pixel-level feature vector to be enhanced, the soil moisture distribution feature map is dynamically anchored in a local receptive field to screen out a set of soil moisture distribution pixel-level feature vectors within the local receptive field, including: Performing information compression on the soil moisture distribution pixel-level feature vector to be enhanced to obtain a soil moisture distribution pixel-level feature distillation encoding vector to be enhanced; Determining a size of a feature receptive field of the soil moisture distribution pixel-level feature distillation coding vector to be enhanced based on a feature distribution spatial structure characteristic of the soil moisture distribution pixel-level feature distillation coding vector to be enhanced; Based on the size of the feature receptive field, a set of pixel-level feature vectors of soil moisture distribution within the local receptive field is filtered out from the set of pixel-level feature vectors of soil moisture distribution.

6. The carbon flux monitoring method based on soil drought remote sensing data according to claim 5, characterized in that: Based on the set of pixel-level feature vectors of soil moisture distribution in the local receptive field, the soil moisture distribution pixel-level feature vector to be enhanced is significantly enhanced to obtain an enhanced soil moisture distribution pixel-level feature vector, including: Based on the feature correlation of each local receptive field pixel-level feature vector of the soil moisture distribution in the set of the local receptive field pixel-level feature vectors with respect to the soil moisture distribution pixel-level feature vector to be enhanced, the set of the local receptive field pixel-level feature vectors of the soil moisture distribution is saliency aggregated to obtain a local receptive field pixel-level feature saliency aggregated coding vector; The enhanced soil moisture distribution pixel-level feature vector is obtained by fusing the soil moisture distribution pixel-level feature saliency aggregation coding vector within the local receptive field and the soil moisture distribution pixel-level feature vector to be enhanced.

7. The carbon flux monitoring method based on soil drought remote sensing data according to claim 6, characterized in that: The enhanced soil moisture distribution pixel-level feature vector is obtained by fusing the soil moisture distribution pixel-level feature saliency aggregate coding vector within the local receptive field and the soil moisture distribution pixel-level feature vector to be enhanced, comprising: The soil moisture distribution pixel-level feature saliency aggregation coding vector within the local receptive field and the soil moisture distribution pixel-level feature vector to be enhanced are weightedly fused to obtain the enhanced soil moisture distribution pixel-level feature vector.

8. The carbon flux monitoring method based on soil drought remote sensing data according to claim 6, characterized in that: The enhanced soil moisture distribution pixel-level feature vector is obtained by fusing the soil moisture distribution pixel-level feature saliency aggregate coding vector within the local receptive field and the soil moisture distribution pixel-level feature vector to be enhanced, comprising: Based on the conformal feature fusion commutation constraint of the set of pixel-level feature vectors of the soil moisture distribution in the local receptive field, modulating the weighting coefficients of the pixel-level feature saliency aggregation coding vector of the soil moisture distribution in the local receptive field and the pixel-level feature vector to be enhanced of the soil moisture distribution; Based on the weighting coefficient, the soil moisture distribution pixel-level feature saliency aggregation coding vector within the local receptive field and the soil moisture distribution pixel-level feature vector to be enhanced are weightedly fused to obtain the enhanced soil moisture distribution pixel-level feature vector.

9. The carbon flux monitoring method based on soil drought remote sensing data according to claim 8, characterized in that: The method comprises: performing carbon flux prediction based on the fusion characteristics of the significant enhanced soil moisture distribution characteristic map and the significant enhanced soil vegetation distribution characteristic map to obtain a carbon flux decoding value, including: Performing cascade fusion processing along the channel dimension on the significant enhanced soil moisture distribution feature map and the significant enhanced soil vegetation distribution feature map to obtain a soil moisture-vegetation distribution fusion coding feature map; The soil moisture-vegetation distribution fusion coding feature map is input into a codec-based carbon flux prediction module to obtain the carbon flux decoding value.

10. A carbon flux monitoring system based on soil drought remote sensing data, characterized in that: include: Remote sensing data acquisition module, used to collect soil moisture remote sensing data using microwave sensors, and simultaneously collect soil vegetation cover remote sensing images using optical remote sensing sensors; A remote sensing feature extraction module is configured to extract remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain a soil moisture distribution feature map and a soil vegetation distribution feature map, wherein extracting remote sensing features from the soil moisture remote sensing data and the soil vegetation cover remote sensing image comprises: performing resolution alignment and remote sensing feature extraction on the soil moisture remote sensing data and the soil vegetation cover remote sensing image to obtain the soil moisture distribution feature map and the soil vegetation distribution feature map; a dynamic feature enhancement module, configured to perform feature enhancement based on dynamic receptive field anchoring on the soil moisture distribution feature map and the soil vegetation distribution feature map to obtain a significant enhanced soil moisture distribution feature map and a significant enhanced soil vegetation distribution feature map; The carbon flux prediction module is used to perform carbon flux prediction based on the fusion characteristics of the significant enhanced soil moisture distribution characteristic map and the significant enhanced soil vegetation distribution characteristic map to obtain a carbon flux decoding value.

Images