Natural resource remote sensing comprehensive monitoring method and device, terminal and medium

Abstract

The application provides a natural resource remote sensing comprehensive monitoring method, device, terminal and medium, wherein the natural resource remote sensing comprehensive monitoring method comprises the following steps: obtaining time-series remote sensing data of a target area and performing preprocessing; obtaining quantity information of each natural resource in the target area based on the time-series remote sensing data as a natural resource quantity distribution result; obtaining quality information of each natural resource in the target area based on the time-series remote sensing data as a natural resource quality result; obtaining an ecological analysis result based on the natural resource quantity distribution result and the natural resource quality result; and outputting the natural resource quantity distribution result, the natural resource quality result and the ecological analysis result as a natural resource remote sensing monitoring result. The interaction relationship between each natural resource is obtained based on the natural resource quantity distribution result and the natural resource quality result, so as to reflect the correlation between each natural resource, facilitate overall analysis of the target area, and achieve a more comprehensive monitoring effect.

Description

Technical Field

[0001] This application belongs to the field of remote sensing monitoring, and relates to a natural resource remote sensing monitoring technology, and in particular to a natural resource remote sensing integrated monitoring method, device, terminal and medium. Background Technology

[0002] Natural resource monitoring refers to the process of observing, assessing, and analyzing the current status and dynamic changes of natural resources such as arable land, water bodies, minerals, forests, grasslands, wetlands, and oceans. Remote sensing monitoring technology can acquire information about the Earth's surface over a large area, over a long period, and periodically. Compared to traditional monitoring methods that rely on manual field surveys, combining remote sensing technology with natural resource monitoring can achieve efficient and accurate monitoring results. Remote sensing monitoring of natural resources includes monitoring aspects such as quantity, quality, and ecological condition.

[0003] Currently, monitoring of natural resources in terms of quantity, quality, and ecological condition typically involves analysis and identification using various models. For example, trained target detection models are used to identify the spatial distribution area and location of natural resources for quantity monitoring; inversion or regression models are used to obtain the quality of natural resources; and analytical models are used to input data on various animals, plants, and the environment to analyze the ecological condition of the current area. However, the above monitoring process separates the acquisition of quantity, quality, and ecological condition data, failing to capture the correlations between these aspects. This makes it difficult to comprehensively analyze the state of natural resources from these perspectives. For instance, changes in the quantity or quality of one type of natural resource can cause changes in the quantity and quality of other types of natural resources, as well as the overall ecological condition of the region. The aforementioned natural resource monitoring processes struggle to accurately capture the correlations between these changes, resulting in poor monitoring effectiveness.

[0004] Meanwhile, when using regression models to assess the quality of natural resources, these models typically require extensive training with a large number of samples to improve accuracy. However, in actual production processes, the number of samples may be limited, leading to lower accuracy in the regression models and consequently, lower precision in the assessment of natural resource quality. Summary of the Invention

[0005] The purpose of this application is to provide a method, device, terminal and medium for integrated remote sensing monitoring of natural resources, in order to solve the problem that the acquisition methods of quantity, quality and ecological conditions in the prior art are independent of each other, and the correlation between quantity, quality and ecological conditions cannot be obtained, resulting in poor natural resource monitoring effect.

[0006] In a first aspect, this application provides a method for comprehensive remote sensing monitoring of natural resources, comprising: acquiring time-series remote sensing data of a target area and preprocessing it; based on the time-series remote sensing data, acquiring quantity information of each natural resource in the target area through a natural resource identification model, as a result of natural resource quantity distribution; constructing a set of remote sensing feature indices; based on the time-series remote sensing data and combined with the set of remote sensing feature indices, calculating and acquiring land parcel remote sensing feature index data using a step-by-step calculation method; inputting the land parcel remote sensing feature index data into a trained regression-physical model to obtain quality information of each natural resource in the target area, as a result of natural resource quality; wherein, the set of remote sensing feature indices includes Several types of remote sensing feature indices are used; based on the results of the quantity distribution of natural resources and the results of the quality of natural resources, initial graph structure data is obtained; based on the initial graph structure data, the interaction relationships between the natural resources are extracted; based on the interaction relationships between the natural resources, the initial graph structure data is updated to obtain updated graph structure data; based on the updated graph structure data, feature extraction is performed to obtain several global and local features; based on the global and local features, ecological analysis results are obtained through a graph structure neural network; the results of the quantity distribution of natural resources, the results of the quality of natural resources, and the results of the ecological analysis are output as the results of remote sensing monitoring of natural resources.

[0007] In one embodiment of this application, the step of calculating and obtaining land parcel remote sensing feature index data based on the time-series remote sensing data and the remote sensing feature index set using a step-by-step calculation method includes: calculating the values ​​of various types of remote sensing feature indices corresponding to each pixel based on the time-series remote sensing data to obtain pixel remote sensing feature index data; performing pixel segmentation based on the pixel remote sensing feature index data to obtain remote sensing features of each object and obtain object remote sensing feature index data; and clustering the remote sensing features of each object based on the quantity distribution results to obtain remote sensing features of each land parcel and obtain land parcel remote sensing feature index data.

[0008] In one embodiment of this application, the construction of the remote sensing feature index set includes: obtaining each sampling grid based on the quantity information of each natural resource in the natural resource quantity distribution results; conducting field sampling based on each sampling grid to obtain the measured data corresponding to each sampling grid; selecting several types of initial remote sensing feature indices; calculating the values ​​of each type of initial remote sensing feature index corresponding to each sampling grid based on the time-series remote sensing data and in combination with each sampling grid; performing correlation analysis on each type of initial remote sensing feature index based on the measured data to screen the initial remote sensing feature indices that meet the correlation requirements as each of the remote sensing feature indices, and forming the remote sensing feature index set based on the screened remote sensing feature indices.

[0009] In one embodiment of this application, updating the initial graph structure data based on the interaction relationships between the natural resources to obtain updated graph structure data includes: obtaining ecological monitoring data; for each node in the initial graph structure data, extracting corresponding ecological information from the ecological monitoring data, and using the ecological information as the node feature corresponding to each node, and updating the node feature as the node feature of the corresponding node; obtaining the mutual relationships between each node in the initial graph structure data based on the interaction relationships between the natural resources, and updating the edge features between the corresponding nodes based on the mutual relationships between the nodes.

[0010] In one embodiment of this application, the step of extracting the interaction relationships between the natural resources based on the initial graph structure includes: performing temporal convolution based on the initial graph structure data to obtain several node change feature vectors; performing graph convolution based on each node change feature vector to obtain edge feature vectors between nodes; and using each edge feature vector as the interaction relationship.

[0011] In one embodiment of this application, the regression-physics model includes a regression sub-model and a physical neural network sub-model; the regression sub-model and the physical neural network sub-model are connected together by a loss function; the regression sub-model is a model based on the CNN framework; and the physical neural network sub-model is a model based on the PINN framework.

[0012] In one embodiment of this application, after obtaining the quality information of each of the natural resources in the target area, the method further includes: obtaining quality change driving information of the target area, and inputting the quality change driving information and the quality information of each of the natural resources into a geographic interpretation model; the geographic interpretation model performs causal analysis on the quality change of each of the natural resources based on the quality change driving information, so as to update and output the quality information of each of the natural resources; and the updated output quality information of each of the natural resources is used as the natural resource quality result.

[0013] Secondly, this application provides a comprehensive remote sensing monitoring device for natural resources, including a preprocessing module, a quantity analysis module, a quality analysis module, an ecological analysis module, and an output module. The preprocessing module acquires and preprocesses time-series remote sensing data of a target area. The quantity analysis module, based on the time-series remote sensing data and using a natural resource identification model, acquires the quantity information of each natural resource within the target area, serving as the natural resource quantity distribution result. The quality analysis module constructs a set of remote sensing feature indices; based on the time-series remote sensing data and the set of remote sensing feature indices, it calculates and acquires land parcel remote sensing feature index data using a step-by-step calculation method; and inputs the land parcel remote sensing feature index data into a trained regression-physical model to obtain the quality information of each natural resource within the target area, serving as the natural resource quantity distribution result. The system comprises the following components: resource quality results; the remote sensing feature index set includes several types of remote sensing feature indices; the ecological analysis module is used to obtain initial graph structure data based on the natural resource quantity distribution results and the natural resource quality results; extract the interaction relationships between the natural resources based on the initial graph structure data; update the initial graph structure data based on the interaction relationships between the natural resources to obtain updated graph structure data; perform feature extraction based on the updated graph structure data to obtain several global and local features; and obtain ecological analysis results through a graph structure neural network based on the global and local features; and the output module is used to output the natural resource quantity distribution results, the natural resource quality results, and the ecological analysis results as natural resource remote sensing monitoring results.

[0014] Thirdly, this application provides a terminal, including: a processor and a memory, wherein the memory and the processor are communicatively connected;

[0015] The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory, so that the terminal performs the natural resource remote sensing integrated monitoring method as described above.

[0016] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a processor, implements the aforementioned integrated remote sensing monitoring method for natural resources.

[0017] As described above, this application provides a method, device, terminal, and medium for comprehensive remote sensing monitoring of natural resources. It acquires quantitative information of each natural resource, including its spatial location and area, to obtain the distribution of natural resource quantities. It also acquires quality information of each natural resource to obtain quality results. Based on the quantitative and quality results, it obtains the interaction relationships between the natural resources and, further, the ecological analysis results. This allows for a holistic analysis of the natural resources and the entire ecosystem, ultimately outputting the quantitative distribution, quality, and ecological analysis results, achieving a more refined and comprehensive monitoring effect. Simultaneously, a trained regression-physics model is used to acquire the quality information of each natural resource. Because the regression-physics model is constrained by the physical laws governing electromagnetic radiation during satellite remote sensing, it maintains high accuracy even with a small sample size, thus ensuring high accuracy of the acquired natural resource quality results. Attached Figure Description

[0018] Figure 1 The diagram shown is a flowchart illustrating a method for integrated remote sensing monitoring of natural resources as described in an embodiment of this application.

[0019] Figure 2 The diagram shown is a schematic representation of time-series remote sensing data as described in an embodiment of this application.

[0020] Figure 3 The diagram shown is a schematic representation of the distribution of natural resources obtained in an embodiment of this application.

[0021] Figure 4 The diagram shown is a flowchart illustrating a method for obtaining natural resource quality results as described in an embodiment of this application.

[0022] Figure 5 The diagram shown is a flowchart illustrating a method for constructing a remote sensing feature index set as described in an embodiment of this application.

[0023] Figure 6 The diagram shows a flowchart illustrating a method for acquiring land parcel remote sensing feature index data as described in an embodiment of this application.

[0024] Figure 7 The diagram shown is a flowchart illustrating another method for obtaining natural resource quality results as described in an embodiment of this application.

[0025] Figure 8 The diagram shown is a flowchart illustrating a method for updating natural resource quality results as described in an embodiment of this application.

[0026] Figure 9The diagram shown is a schematic representation of a natural resource quality result obtained in an embodiment of this application.

[0027] Figure 10 The diagram shown is a flowchart illustrating a method for obtaining ecological analysis results as described in an embodiment of this application.

[0028] Figure 11 The diagram shown is a flowchart illustrating a method for obtaining interaction relationships as described in an embodiment of this application.

[0029] Figure 12 The diagram shown is a schematic representation of an interaction relationship obtained in an embodiment of this application.

[0030] Figure 13 The diagram shown is a flowchart illustrating a method for obtaining updated graph structure data as described in an embodiment of this application.

[0031] Figure 14 The diagram shown is a schematic representation of ecological monitoring data as described in an embodiment of this application.

[0032] Figure 15 The diagram shown is a schematic representation of an ecological analysis result obtained in an embodiment of this application.

[0033] Figure 16 The diagram shown is a structural schematic of a natural resource remote sensing integrated monitoring device according to an embodiment of this application.

[0034] Figure 17 The diagram shown is a structural schematic of a terminal as described in an embodiment of this application.

[0035] Explanation of reference numerals in the attached figures

[0036] 61: Preprocessing module; 62: Quantitative analysis module; 63: Quality analysis module; 64: Ecological analysis module; 65: Output module; 70: Terminal; 71: Processor; 72: Memory; 721: Operating system; 722: Application program; 73: User interface; 74: Network interface; 75: Bus system. Detailed Implementation

[0037] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0038] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0039] Natural resource monitoring includes monitoring in terms of quantity, quality, and ecological condition. However, existing natural resource monitoring methods often focus on monitoring only one aspect of these three aspects, or conduct monitoring of these three aspects relatively independently. This makes it impossible to obtain the correlation between the three aspects and their mutual influence, making it difficult to analyze the overall state of natural resources and resulting in poor monitoring effectiveness.

[0040] The quantity of natural resources refers to the distribution information of natural resources. For example, the quantity of natural resources includes the spatial location and area of ​​the natural resources. For arable land, the quantity distribution information includes its spatial location and the area of ​​the arable land region.

[0041] The quality of natural resources refers to their usability. For example, the quality of natural resources includes their health, i.e., their sustainable utilization, such as whether forest land is affected by pests and diseases, or whether mineral resources are depleted or renewable. The quality of natural resources also includes stress information, i.e., factors that pose a threat to natural resources. For example, for water sources, stress information includes the pollution status of the water body and nearby pollution sources. For productive natural resources, the quality of natural resources also includes their growth status. For example, for arable land, the growth status refers to the growth of crops on that land; for forest land, the growth status refers to the forest stock volume.

[0042] The ecological status of natural resources refers to the impact of changes in various natural resources on the ecosystem within the target area. For example, if the area of ​​forest land expands, the ecological status will be: the ecosystem will have significant positive effects on soil and water conservation, biodiversity, and carbon sources and sinks; if arable land is attacked by pests and diseases, the ecological status will be: the ecosystem's food security will be at great risk.

[0043] To address the technical problems existing in the prior art, the following embodiments of this application provide a method, device, terminal, and medium for integrated remote sensing monitoring of natural resources. By acquiring the results of the quantity distribution and quality of natural resources, and based on the results of the quantity distribution and quality of natural resources, the interaction relationships between various natural resources are obtained to reflect the correlation and mutual influence between various natural resources. This facilitates the overall analysis of the ecological situation of the entire target area, achieving a more refined and comprehensive monitoring effect.

[0044] The technical solutions in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0045] like Figure 1 As shown, this embodiment provides a comprehensive remote sensing monitoring method for natural resources, used to achieve three-dimensional monitoring of the quantity, quality, and ecological condition of natural resources within a target area. Specifically, the comprehensive remote sensing monitoring method for natural resources includes:

[0046] S100: Acquire time-series remote sensing data of the target area and perform preprocessing.

[0047] Among them, time-series remote sensing data includes several remote sensing images arranged in time sequence. Time sequence arrangement means that the images are arranged from early to late based on the acquisition time. The acquisition range of the remote sensing images is the target area, which is used for remote sensing monitoring of the target area.

[0048] It should be noted that time-series remote sensing data includes remote sensing images from different detection bands. For example, time-series remote sensing data includes several panchromatic remote sensing images, multispectral remote sensing images, hyperspectral remote sensing images, microwave remote sensing images, and thermal infrared remote sensing images, with the images of the same detection band arranged in time sequence. For example, such as... Figure 2 As shown, in one optional implementation, the acquired time-series remote sensing data includes multispectral remote sensing images, panchromatic remote sensing images, and microwave remote sensing images.

[0049] Furthermore, these remote sensing images undergo preprocessing to form preprocessed time-series remote sensing data, which is then arranged chronologically to facilitate subsequent natural resource monitoring. Specifically, for panchromatic and multispectral remote sensing images, the preprocessing steps include: orthorectifying each image and fusing panchromatic images acquired at the same time with multispectral images corresponding to each band to obtain high-resolution remote sensing images that reflect the acquisition status of multiple bands. For hyperspectral remote sensing images, the preprocessing steps include atmospheric correction, radiometric calibration, and orthorectification. For microwave remote sensing images, the preprocessing steps include differential interferometry processing, i.e., calibration through baseline estimation and acquisition of interferograms based on the calibrated images. Coherence calculation, adaptive filtering, phase unwrapping, orbit refinement, and leveling are then performed based on the interferograms to extract and quantify minute surface deformation information with high precision, while also eliminating the influence of topographic factors on the remote sensing images. For thermal infrared remote sensing images, preprocessing steps include atmospheric correction, radiometric calibration, orthorectification, and blackbody radiance calculation.

[0050] Furthermore, after processing the remote sensing images of each detection band accordingly, images acquired at the same time are fused to obtain image data that integrates information from each detection band. This facilitates the subsequent extraction of natural resource-related information based on the remote sensing images. Specifically, taking time-series remote sensing data including several panchromatic, multispectral, hyperspectral, microwave, and thermal infrared images as an example, after processing the remote sensing images of each detection band accordingly, images acquired at the same time are fused to obtain fused remote sensing images corresponding to each acquisition time. These fused remote sensing images are then arranged chronologically as preprocessed time-series remote sensing data.

[0051] S200, based on time-series remote sensing data, obtains the quantity information of various natural resources within the target area, serving as the result of natural resource quantity distribution.

[0052] The distribution of natural resource quantities is used to characterize the spatial distribution of various natural resources within the target area. For example, for each remote sensing image arranged chronologically in time-series remote sensing data, the type, location, and area of ​​each natural resource are labeled to provide quantitative information about each resource, i.e., its spatial location and area, thus forming the distribution of natural resource quantities. For instance, for each remote sensing image arranged chronologically, cultivated land is labeled with text or other identifiers, along with its geographical location (latitude and longitude) and area. The labeled remote sensing images arranged chronologically are then output as the distribution of natural resource quantities.

[0053] Specifically, time-series remote sensing data is input into the trained natural resource identification model to obtain the quantity information of each natural resource in the target area, and the quantity information of each natural resource is used as the result of the natural resource quantity distribution.

[0054] The natural resource identification model is used to identify and extract various natural resources. For example, the natural resource identification model can be any of the following: a machine learning model such as a vector machine, decision tree, or random forest, or a deep learning model such as a convolutional neural network. The natural resources are delineated and marked on remote sensing images by manual means or software, such as Easy Feature, and used as training samples to train the natural resource identification model, thereby improving the identification accuracy of the natural resource identification model.

[0055] Furthermore, each remote sensing image from the time-series remote sensing data is input into the natural resource identification model, enabling the model to identify, extract, and label the natural resources on each remote sensing image. It should be noted that since each pixel in a remote sensing image has corresponding geographic coordinate information, meaning the remote sensing image possesses spatial location information, each natural resource on each remote sensing image also contains spatial location information. Based on this spatial location information, the quantity information of each natural resource is obtained and output as the natural resource quantity distribution result.

[0056] For example, such as Figure 3 As shown, it is displayed as based on Figure 2 The image shows the distribution of natural resources obtained from time-series remote sensing data. Red areas represent arable land, and blue areas represent water resources. It should be noted that... Figure 2 Each pixel in each remote sensing image has geographic coordinate information. Based on this, the red and blue areas also have spatial location information, and the spatial extent of the corresponding natural resources can be represented by the number of pixels in the red and blue areas.

[0057] S300, based on time-series remote sensing data, acquires quality information of various natural resources within the target area, which serves as the natural resource quality outcome.

[0058] The quality information of each natural resource is used to characterize the quality status of each resource, including but not limited to its health, stress information, and growth status. For example, for arable land, its quality information includes crop growth status, pest and disease erosion, and soil condition; for water sources, its quality information includes water pollution, turbidity, and eutrophication. It should be noted that those skilled in the art should understand the specific aspects that need to be set for the quality information of each natural resource; this embodiment does not impose specific limitations on this.

[0059] Furthermore, in order to obtain quality information of various natural resources through time-series remote sensing data, specific remote sensing characteristic indices are calculated from each remote sensing image in the time-series remote sensing imagery, so as to reflect the quality of natural resources through specific remote sensing characteristic indices. Here, the remote sensing characteristic index refers to the parameter value calculated from the optical band information in the remote sensing imagery.

[0060] Specifically, taking arable land as an example, crop growth status can be reflected through remote sensing imagery using remote sensing characteristic indices such as NDVI (Normalized Difference Vegetation Index), GNDVI (Green Normalized Difference Vegetation Index), and FPAR (Fraction of Photosynthetically Active Radiation). Pest and disease erosion can be reflected through remote sensing imagery using remote sensing characteristic indices such as NDVI (Normalized Difference Vegetation Index) and GLCM (Gray Level Co-occurrence Matrix). For arable land soil conditions, remote sensing imagery can be used to obtain BI (Brightness Index), iron oxide index, EVI (Enhanced Vegetation Index), and NDVI (Normalized Difference Vegetation Index). Remote sensing indices such as the Normalized Difference Vegetation Index (NDWI) reflect soil heavy metal pollution; remote sensing indices such as the Iron Oxide Index, Carbonate Index, and Brightness Index (BI) reflect soil pH; and remote sensing indices such as the Normalized Difference Water Index (NDWI), Moisture Stress Index (MSI), and Crop Water Stress Index (CWSI) reflect soil moisture. Of course, the above is merely an illustrative explanation of how various remote sensing indices reflect the quality of arable land. In actual production applications, those skilled in the art can select the same or different remote sensing indices to reflect the quality information of various natural resources according to actual needs. This embodiment does not specifically limit this.

[0061] It should be noted that those skilled in the art should know the calculation methods of various remote sensing image feature indices, and this embodiment does not make specific limitations here.

[0062] Natural resource quality results are used to characterize the quality of various natural resources within a target area. For example, for each remote sensing image arranged chronologically in time-series remote sensing data, quality parameter values ​​for each natural resource are labeled to intuitively represent the quality of the natural resources, forming natural resource quality results. The quality parameter values ​​for each natural resource are obtained based on the conversion of various remote sensing feature indices, used to more intuitively characterize the quality of natural resources. Specifically, taking cultivated land as an example, for crop growth status, various remote sensing feature indices are converted into predicted crop yield quality; for pest and disease erosion, the presence of pest and disease erosion is analyzed based on various remote sensing feature indices, exemplarily represented by the number 1 if pests and diseases are present, and 0 if they are not; cultivated land soil conditions are converted based on various remote sensing feature indices into parameters such as heavy metal element concentration, soil pH value, and soil moisture. Based on this, the converted quality parameter values ​​intuitively reflect the quality of cultivated land. It should be noted that the above is only an illustrative example. In actual production applications, those skilled in the art can convert the calculated remote sensing feature indices into quality parameter values ​​that are the same as or different from those in this embodiment, based on the characteristics of each natural resource and actual needs. This embodiment does not make any specific limitations in this regard.

[0063] Based on this, using time-series remote sensing data, selected remote sensing feature indices are calculated to obtain remote sensing feature index data for the target area. This data is then converted into quality parameter values ​​for each natural resource, thus intuitively reflecting the quality information of each resource and serving as a natural resource quality outcome. The conversion of the remote sensing feature index data for the target area into quality parameter values ​​for each natural resource can be achieved through a trained regression-physical model. The regression-physical model includes a regression sub-model and a physical neural network sub-model. The regression sub-model is based on a CNN framework, and the physical neural network sub-model is based on a PINN framework. The regression sub-model and the physical neural network sub-model are connected by a loss function to achieve synchronous training of both sub-models. During a single feedback process, both sub-models are optimized simultaneously, ensuring high accuracy even with a small number of training samples.

[0064] The specific steps and principles for obtaining results on the quality of natural resources will be explained in detail below.

[0065] In some alternative implementations, such as Figure 4 As shown, methods for obtaining natural resource quality results based on time-series remote sensing data include:

[0066] S310, Construct a set of remote sensing feature indices.

[0067] The remote sensing feature index set includes several types of remote sensing feature indices. All types of remote sensing feature indices in the set are selected remote sensing feature indices that can well characterize the quality of natural resources, so as to reflect the quality information of each natural resource through the selected remote sensing feature indices.

[0068] For example, such as Figure 5 As shown, the method for constructing a set of remote sensing feature indices includes:

[0069] S311. Based on the spatial range of each natural resource in the natural resource quantity distribution results, obtain each sampling grid; conduct field sampling based on each sampling grid to obtain the measured data corresponding to each sampling grid.

[0070] Among them, the measured data is used to characterize the actual quality of the geographical location corresponding to each sampling grid in the target area. Specifically, the measured data is a set of several sampling parameter values ​​obtained from on-site sampling of each sampling grid.

[0071] It should be noted that the various sampling parameter values ​​obtained from each sampling grid are of different types. Please refer to the aforementioned content on quality parameter values. The sampling parameter values ​​corresponding to each sampling grid are used to intuitively characterize the quality information of the natural resource corresponding to each sampling grid. Based on this, the sampling type corresponding to each sampling grid is determined based on the quality information of the natural resource corresponding to each sampling grid. For example, taking cultivated land as an example, if the quality information of cultivated land includes crop growth status, pest and disease erosion status, and cultivated land soil condition, the corresponding sampling type may include, but is not limited to: actual crop yield quality, pest and disease erosion status, heavy metal element concentration, soil pH value, and soil moisture. Of course, the above is only an illustrative description of the sampling types that can be set when the natural resource is cultivated land. In actual production applications, the same or different sampling types as in this embodiment can be selected.

[0072] It is worth noting that, for ease of recording and storage, each sampling grid records the field sampling conditions numerically. For sampling types where specific values ​​can be obtained, such as actual crop yield quality, heavy metal concentration, soil pH, and soil moisture, the measured values ​​are recorded. For sampling types where values ​​cannot be directly measured, such as pest and disease infestation, the values ​​are converted into numerical representations; for example, the presence of pests and diseases is represented by the number 1, and their absence by the number 0. Similarly, for the aforementioned quality parameter values ​​where actual values ​​cannot be obtained, the actual situation will be analyzed based on various remote sensing feature indices, and then converted into numerical representations.

[0073] Based on this, the actual method for obtaining the measured data is to obtain the sampling parameter values ​​corresponding to each sampling grid, and then use the set of sampling parameter values ​​of all sampling grids as the measured data.

[0074] Specifically, the spatial range to be sampled is obtained based on the spatial range displayed on each remote sensing image in the results of natural resource quantity distribution. For example, the sampling range is the sum of the spatial ranges of each natural resource on its corresponding remote sensing image.

[0075] Specifically, for any remote sensing image, the sampling area displayed on it is divided into several sampling grids of preset size. For example, the preset size of each sampling grid is 1km*1km, which corresponds to an area in the actual geographic space. Of course, those skilled in the art can set the size of the sampling grid based on actual needs or experience, and this embodiment does not impose specific limitations here.

[0076] Based on each sampling grid, the sampling route and sampling method are obtained. Specifically, based on the geographic coordinate information of each pixel in each remote sensing image, the actual geographical location of each sampling grid divided on the remote sensing image is obtained, and based on the actual geographical location of these sampling grids, the sampling route and sampling method are obtained. For example, based on the actual geographical location corresponding to each sampling grid, the shortest route is planned as the sampling route; based on the sampling type that requires on-site sampling for each sampling grid, the corresponding sampling method is set. The sampling method includes, but is not limited to, manual field sampling, high-altitude camera acquisition, and drone acquisition, etc., and this embodiment does not specifically limit it.

[0077] Based on the planned sampling route, on-site sampling was conducted for each sampling grid using the corresponding sampling method. The corresponding sampling parameter values ​​were obtained and recorded as measured data. It should be noted that, to avoid discrepancies between the measured data from the on-site sampling and the actual quality reflected in the corresponding remote sensing images in the time-series remote sensing data after a long period of time, the on-site sampling time for each sampling grid in each remote sensing image was consistent with the acquisition time of that remote sensing image.

[0078] S312, Select several types of initial remote sensing feature indices; Based on time-series remote sensing data and combined with each sampling grid, calculate the values ​​of each type of initial remote sensing feature index corresponding to each sampling grid.

[0079] Each initial remote sensing feature index is used to characterize all indicators calculated from remote sensing data that can reflect the quality information of natural resources. Specifically, for each natural resource, those skilled in the art can set corresponding initial remote sensing feature indices based on experience, literature, or actual needs to reflect the quality information of each natural resource. This embodiment does not impose specific limitations on this.

[0080] Furthermore, for any sampling grid divided from any remote sensing image in the time-series remote sensing data, the values ​​of all corresponding initial remote sensing feature indices are calculated to obtain the values ​​of each initial remote sensing feature index corresponding to each sampling grid in all remote sensing images. It should be noted that those skilled in the art should know how to calculate the values ​​of each initial remote sensing feature index based on remote sensing images, and this embodiment does not limit this.

[0081] S313. Based on measured data, the correlation degree of various types of initial remote sensing feature indices is analyzed to select the various types of initial remote sensing feature indices that meet the correlation degree requirements as each remote sensing feature index, and a set of remote sensing feature indices is formed.

[0082] Among them, correlation analysis refers to analyzing the strength of the correlation between various types of initial remote sensing feature indices and measured data, and screening various types of initial remote sensing feature indices based on the strength of the correlation, so as to select remote sensing feature indices with stronger correlation and better reflecting the quality information of various natural resources.

[0083] In some optional implementations, a supervised model is used to analyze the correlation between each initial remote sensing feature index to select the appropriate index. For example, the supervised model framework may employ any of the following: Random Forest (RF), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), or Support Vector Machine (SVM). The values ​​of various initial remote sensing feature indices corresponding to each sampling grid are input as independent variables, and measured data are used as the dependent variable. One or more initial remote sensing feature indices are randomly selected, and the correspondence between the values ​​of these initial remote sensing feature indices and the sampling grids is shuffled. The supervised model is then used to analyze and calculate the shuffled prediction data, analyzing the difference between the prediction data and the measured data. For example, the difference between the sampling parameter values ​​corresponding to each sampling grid and the corresponding values ​​in the prediction data is calculated to analyze the prediction accuracy of the shuffled model. The process of randomly selecting initial remote sensing feature indices, shuffling them, and obtaining the prediction accuracy after shuffling is repeated. The importance of each initial remote sensing feature index is scored based on this prediction accuracy; the lower the prediction accuracy, the higher the importance score of the selected initial remote sensing feature index. Finally, each initial remote sensing feature index whose importance score exceeds an importance score threshold is output as a remote sensing feature index, and the set of all these remote sensing feature indices is called the remote sensing feature index set. Those skilled in the art should know the specific value of the importance score threshold based on actual needs; this embodiment does not impose specific limitations here.

[0084] In some alternative implementations, for any sampling type, sampling grids whose sampling parameter value differences do not exceed a sampling difference threshold are grouped together. Based on the values ​​of various initial remote sensing feature indices corresponding to each sampling grid within the same group, the standard deviation of each initial remote sensing feature index is calculated. All initial remote sensing feature indices whose standard deviations within all groups do not exceed a standard deviation threshold are used as the remote sensing feature indices corresponding to that sampling type. The set of all remote sensing feature indices corresponding to all sampling types is then used as the remote sensing feature index set. Those skilled in the art should know the specific values ​​for setting the sampling difference threshold and standard deviation threshold based on actual needs; this embodiment does not impose specific limitations here.

[0085] Based on this, the actual quality of each natural resource is reflected by measured data, and the correlation between each selected initial remote sensing feature index and the measured data is analyzed to select strongly correlated remote sensing feature indices. Based on these remote sensing feature indices, the quality of each natural resource is analyzed in subsequent monitoring processes.

[0086] Furthermore, the measured data obtained in step S311, the values ​​of various initial remote sensing feature indices corresponding to each sampling grid in step S312, and the remote sensing feature indices selected in step S313 are also used in the training process of the regression-physics model, so that the regression-physics model can convert the values ​​of the remote sensing feature indices into the quality parameter values ​​of each natural resource, so as to more intuitively reflect the quality information of each natural resource.

[0087] Specifically, based on the remote sensing feature indices selected in step S313, the values ​​of each remote sensing feature index corresponding to each sampling grid are extracted from the values ​​of various initial remote sensing feature indices corresponding to each sampling grid obtained in step S312, and used as independent variables. The measured data, i.e., the sampling parameter values ​​corresponding to each sampling grid, are used as dependent variables and input into the regression sub-model so that the regression sub-model can learn to obtain the fitting relationship between each remote sensing feature index and each sampling parameter value.

[0088] Simultaneously, based on the physical laws governing the electromagnetic radiation process of satellite remote sensing, physical partial differential equations are constructed in the physical neural network sub-model. The physical neural network sub-model is then trained based on the values ​​of each remote sensing feature index and each sampling parameter corresponding to each sampling grid. It should be noted that during the training process, the fitting process between each remote sensing feature index and each sampling parameter value must conform to the constructed partial differential equations, so that the physical neural network sub-model fits under the constraints of physical laws, thereby improving the accuracy of the obtained fitting relationship. It should be noted that the aforementioned physical partial differential equations are used to characterize the electromagnetic radiation transmission laws of remote sensing images. Specifically, based on information such as the solar reflectivity, satellite elevation angle, solar irradiance, and detection wavelength obtained from the satellite acquiring each remote sensing image, corresponding physical partial differential equations are constructed. Those skilled in the art should be familiar with the specific form of the constructed physical partial differential equations; this embodiment does not impose specific limitations here.

[0089] Based on this, the training process of the regression-physical model involves inputting the values ​​of the selected remote sensing feature indices into the trained model, which is then fitted by the regression sub-model and the physical neural network sub-model respectively, to obtain the actual quality of the geographical location corresponding to each sampling grid in the target area. The fitting results of the two sub-models are then compared with the measured data, and error feedback is performed based on the comparison results to optimize the two sub-models.

[0090] Furthermore, both the regression sub-model and the physical neural network sub-model employ mean squared error (MSE) for error feedback to optimize the model. Specifically, the weighted sum of the MSEs of the regression sub-model and the physical neural network sub-model is used as the error function of the regression-physical model. That is, during the training process of the regression sub-model and the physical neural network sub-model, this weighted sum of MSEs is used to simultaneously optimize the two models. This allows the regression-physical model to perform regression analysis under the constraints of actual physical laws. Compared to existing regression models that require a large number of samples for training, the regression-physical model in this embodiment can still achieve high accuracy with a smaller number of training samples, thereby improving the accuracy of the acquired natural resource quality information and contributing to better monitoring results.

[0091] S320, based on time-series remote sensing data and combined with a set of remote sensing feature indices, uses a step-by-step calculation method to calculate and obtain the remote sensing feature index data of land parcels.

[0092] Among them, the remote sensing feature index data of land parcels is used to characterize the distribution of remote sensing feature indices of various land parcels within the target area. A single land parcel is used to characterize a collection of objects with close spatial relationships within the target area, and a single object is used to characterize a single land feature within the target area. For example, a single plot of cultivated land is considered a single object, while a cultivated area formed by multiple plots of cultivated land is considered a single land parcel.

[0093] Specifically, such as Figure 6 As shown, the methods for obtaining land parcel remote sensing feature index data include:

[0094] S321, based on time-series remote sensing data, calculates the values ​​of various types of remote sensing feature indices corresponding to each pixel, and obtains pixel remote sensing feature index data.

[0095] Specifically, based on the remote sensing feature index set, each remote sensing feature index is obtained. For any remote sensing image in the time-series remote sensing data, the value of each remote sensing feature index corresponding to each pixel is calculated to represent the distribution of remote sensing feature indices for that remote sensing image. The set of remote sensing feature index distributions for all remote sensing images is then used as the pixel remote sensing feature index data. It should be noted that those skilled in the art should be familiar with the calculation methods of each remote sensing feature index, and this embodiment does not impose specific limitations on them.

[0096] S322: Pixel segmentation is performed based on pixel remote sensing feature index data to obtain the remote sensing features of each object and the object remote sensing feature index data.

[0097] Among them, object remote sensing features are remote sensing features that characterize a single object, and object remote sensing feature index data are used to characterize the distribution of remote sensing feature indices of various parcels within the target area.

[0098] Based on the remote sensing feature indices corresponding to each pixel, the similarity between pixels is analyzed to classify pixels with high similarity as pixels of the same ground feature, forming an object. Specifically, the difference or ratio of the remote sensing feature indices corresponding to any two pixels is calculated, and the average of the differences or ratios of the various remote sensing feature indices is taken as the difference value between the two pixels. Pixels whose difference value does not exceed the difference threshold and can be interconnected are divided into an object, forming the object's remote sensing feature. Here, interconnected pixels refer to pixels where there are no multiple pixels whose difference value exceeds the difference threshold that would divide these pixels into multiple blocks; that is, these pixels can be connected to form an object. It should be noted that those skilled in the art can set the specific value of the difference threshold according to actual needs, and this embodiment does not impose specific limitations here.

[0099] Furthermore, for different objects, due to the different natural resources they correspond to, the selected remote sensing feature indices are also different. Based on this, in order to improve the efficiency and accuracy of image segmentation, this embodiment also analyzes the index overlap between pixels to classify pixels with high remote sensing feature index overlap into one object. Specifically, the number of remote sensing feature indices of the same type between any two pixels is calculated, and the ratio of the number of remote sensing feature indices of the same type to the maximum number of remote sensing feature index types between the two pixels is used as the index overlap rate between the two pixels. An index overlap rate threshold is set, and at the same time, the difference value between any two pixels is calculated to classify pixels whose difference value does not exceed the difference threshold, whose index overlap rate is not less than the index overlap rate threshold, and who can be interconnected into one object, forming the object's remote sensing feature. Similarly, those skilled in the art can set the specific value of the index overlap rate threshold based on actual needs, and this embodiment does not impose specific limitations here.

[0100] Furthermore, for any object remote sensing feature, based on the values ​​of various types of remote sensing feature indices corresponding to all pixels within the object remote sensing feature, the values ​​of various types of remote sensing feature indices corresponding to the object remote sensing feature are obtained, so as to form object remote sensing feature index data based on each object remote sensing feature.

[0101] Specifically, when obtaining the values ​​of various types of remote sensing feature indices corresponding to the remote sensing features of a single object, for any type of remote sensing feature index, the values ​​of the remote sensing feature index of that type corresponding to all pixels within the remote sensing features of the object are obtained, and the average, median, or mode of the remote sensing feature index of that type is obtained as the value of the remote sensing feature index of that type for the object's joystick feature.

[0102] Based on this, the values ​​of various types of remote sensing feature indices of all objects within the target area are obtained to form object remote sensing feature index data.

[0103] S323, based on the quantitative distribution results, clusters the remote sensing features of each object to obtain the remote sensing features of each land parcel and obtain land parcel remote sensing feature index data.

[0104] Specifically, images from the quantity distribution results are overlaid onto images from the object remote sensing feature index data. This overlays the quantity information of each natural resource in the quantity distribution results—namely, the type, location, and area of ​​each natural resource—onto the object's remote sensing features. This facilitates clustering of the object's remote sensing features based on the quantity information, allowing for the formation of remote sensing features for each land parcel that are geographically close and spatially related. For example, spatial autocorrelation or raster-based clustering methods are used. It should be noted that those skilled in the art should be familiar with methods for clustering object remote sensing features based on quantity information; this embodiment does not limit such methods.

[0105] Furthermore, the values ​​of various types of remote sensing feature indices corresponding to the remote sensing feature indices of each land parcel are obtained to form land parcel remote sensing feature index data. Specifically, for any land parcel remote sensing feature, the values ​​of various types of remote sensing feature indices of all corresponding object remote sensing features are obtained. The Z-Score standardization method is used to normalize the values ​​of the same type of remote sensing feature index for different object remote sensing features, and then the average value is taken as the land parcel remote sensing feature value for that type of remote sensing feature index.

[0106] Based on this, land parcel remote sensing characteristic index data can be obtained from time-series remote sensing data.

[0107] S330 inputs the remote sensing feature index data of the land parcel into the trained regression-physical model to obtain the quality information of each natural resource in the target area, which serves as the natural resource quality result.

[0108] Specifically, as mentioned earlier, the trained regression-physics model transforms the values ​​of various remote sensing feature indices to obtain the values ​​of various quality parameters, so as to intuitively reflect the quality information of various natural resources.

[0109] For example, for each image in the remote sensing feature index data of land parcels, a regression-physical model is used to label the corresponding quality parameter values ​​on each land parcel, thereby reflecting the quality information of each natural resource. The set of all images labeled with each quality parameter value is taken as the natural resource quality result.

[0110] Based on this, this embodiment can obtain the quality information of various natural resources through time-series remote sensing data, so as to form natural resource quality results.

[0111] It should be noted that the quality of various natural resources is typically affected by numerous external factors, including but not limited to topographical changes, economic development, weather changes, policies, and population changes. Therefore, in order to reflect the quality of various natural resources and infer the possible causes of their quality changes, and to achieve a more refined monitoring effect, such as... Figure 7 As shown, after obtaining the quality information of each natural resource, the process also includes:

[0112] S340: Obtain quality change-driven information for the target area to update the quality information of each natural resource, as a natural resource quality outcome.

[0113] Among them, the quality change driving information refers to the external interference factors that alter the quality of various natural resources. It should be noted that those skilled in the art should be aware of the methods for obtaining the quality change driving information of the target area, such as collection based on official information disclosure channels; this embodiment does not impose any limitations on this method.

[0114] Based on the information driven by quality changes, the reasons for changes in various natural resources are obtained, and the quality information of each natural resource is updated accordingly to obtain new natural resource quality results. This makes the natural resource quality results include factors of quality change, so as to achieve a more refined monitoring effect.

[0115] Specifically, such as Figure 8 As shown, the methods for updating natural resource quality results include:

[0116] S341, Obtain the quality change driving information of the target area, and input the quality change driving information and the quality information of each natural resource into the geographic interpretation model.

[0117] S342, the geographic interpretation model, is based on quality change-driven information to analyze the causes of quality changes in various natural resources, so as to update and output the quality information of each natural resource.

[0118] S343 specifies that the updated output quality information of each natural resource shall be used as the natural resource quality outcome.

[0119] Among them, the geographic interpretation model is a model that interprets the changes in the quality information of various natural resources based on quality change-driven information.

[0120] Specifically, images corresponding to different acquisition times are acquired from the natural resource quality results to obtain the quality status of each natural resource at different times. Based on the time-series arrangement of the quality status of each natural resource, the quality changes of each natural resource are obtained. Furthermore, based on the quality changes of each natural resource, combined with the quality change-driving information of the target area, the quality changes of each natural resource are interpreted, and the acquired interpretations are integrated into the images of the areas where quality changes have occurred to form new natural resource quality results.

[0121] For example, the geographic interpretation model is the GeoXAI (Geospatial eXplainable Artificial Intelligence) model, which includes an XGboost (eXtreme Gradient Boosting) layer and a SHAP (SHapley Additive exPlanation) layer. The XGboost layer is used to analyze and predict the possible changes in the quality of natural resources caused by various quality change driving information, and to interpret the causes of quality changes based on the SHAP layer. The obtained interpretation is then integrated into the image of the quality change to form new natural resource quality results.

[0122] Based on this, the quality of natural resources within the target area can be more accurately reflected through the results of natural resource quality assessment.

[0123] For example, such as Figure 9 As shown, it is displayed as based on Figure 2 The time-series remote sensing data displayed in the image and Figure 3 The data shows the distribution of natural resource quantities and the obtained natural resource quality results. For arable land, the yellow areas within the red zones represent areas with lodging and weeds, while for water resources, the green areas within the blue zones represent areas with eutrophication. Based on this, it can be seen that the arable land in the region has relatively good lodging and weed growth, but the eutrophication of water resources is quite serious, thus reflecting the quality of arable land and water resources, and serving as the natural resource quality results.

[0124] S400 obtains ecological analysis results based on the results of natural resource quantity distribution and natural resource quality.

[0125] The ecological analysis results are used to characterize the ecosystem of the entire target area. Specifically, the ecological analysis results reflect changes in the area or location of each natural resource, as well as changes in the quality of each natural resource and their causes. Furthermore, since changes in each natural resource are interconnected, obtaining information on changes in the area or location of each natural resource, as well as changes in the quality of each natural resource and their causes, requires analysis based on the interactions between the natural resources.

[0126] In some alternative implementations, such as Figure 10 As shown, the methods for obtaining ecological analysis results include:

[0127] S410: Based on the results of the quantity distribution of natural resources and the quality of natural resources, obtain the initial graph structure data.

[0128] The initial graph structure data is used to characterize the quantity and quality of each natural resource within the target area, as well as the spatial relationships between them, so as to facilitate the analysis of the ecological situation within the target area based on the graph structure data.

[0129] Specifically, the initial map structure data includes several initial map structures arranged chronologically. Based on images labeled with quantity information in the natural resource quantity distribution results and images labeled with quality information in the natural resource quality results, images corresponding to the same acquisition time are extracted. Each natural resource labeled on the images in the natural resource quantity distribution results is used as a node. Based on the location of each natural resource, the spatial relationships between them are obtained, such as spatial adjacency, and these are used as edges between nodes. The quality information of remote sensing features of each land parcel labeled on the images in the natural resource quality results is used as the node features. The initial map structures at each acquisition time are obtained and arranged chronologically to form the initial map structure data.

[0130] It should be noted that the relationship between remote sensing features of each land parcel and each natural resource may be one-to-one or many-to-one. When a single natural resource corresponds to remote sensing features of multiple land parcels, the natural resource is split into multiple nodes to correspond one-to-one with the remote sensing features of each land parcel.

[0131] S420 extracts the interaction relationships between various natural resources based on the initial graph structure data.

[0132] Among them, the interaction relationship between natural resources is used to characterize the mutual influence of various natural resources.

[0133] Specifically, such as Figure 11 The methods for obtaining the interaction relationships between various natural resources, as shown, include:

[0134] S421 performs temporal convolution based on the initial graph structure data to obtain several node change feature vectors.

[0135] Specifically, the temporal convolution method is used to analyze the corresponding nodes of each initial graph structure in the initial graph structure data to obtain the changes in node features over time, which are then used as node change feature vectors.

[0136] S422 performs graph convolution based on the changing feature vectors of each node to obtain the edge feature vectors between each node.

[0137] In this context, the feature vectors of each edge are used as the interaction relationships between various natural resources.

[0138] Specifically, by using graph convolution methods and combining the changes in node features over time, the interaction relationships between nodes can be obtained.

[0139] For example, if the spatial relationship between water sources and arable land is one of adjacency, and over time, the pollution level of water sources worsens, while crop growth declines in the arable land, the interaction can be understood as follows: water pollution affects the output of arable land. For example, ... Figure 12As shown, the interaction between the obtained water resources and arable land is illustrated. The yellow area represents a decline in arable land quality, the blue area represents no change in arable land quality, and the green area represents an improvement in arable land quality. Obviously, eutrophication of water resources is the cause of the decline in arable land quality. That is, the interaction between water resources and arable land is: eutrophication of water resources leads to a decline in arable land quality.

[0140] S430, based on the interaction relationships between various natural resources, updates the initial graph structure data to obtain updated graph structure data.

[0141] Among them, the updated graph structure data is the graph structure data after updating the interaction relationships between various natural resources.

[0142] Specifically, the edges of each node in each initial graph structure are replaced with the interaction relationships between each natural resource to update the initial graph structure, obtain the updated graph structure, and arrange all the updated graph structures in chronological order as the updated graph structure data.

[0143] It should be noted that when conducting ecological analysis, it is also necessary to add other ecological monitoring data, such as information on temperature, precipitation, and slope. Based on this, such as... Figure 13 As shown, the methods for obtaining updated graph structure data can also include:

[0144] S431, Acquire ecological monitoring data.

[0145] Among them, ecological monitoring data refers to the monitoring information required for ecological analysis. It should be noted that those skilled in the art should know how to set the content of ecological monitoring data based on actual needs. For example, ecological monitoring data includes information such as temperature, precipitation, slope, natural resource distribution density, and soil fertility. Furthermore, those skilled in the art should know the specific methods for obtaining ecological monitoring data, such as obtaining it based on publicly available information released by meteorological stations. This embodiment does not impose specific limitations here. For example, as shown... Figure 14 As shown, the ecological monitoring data includes the distribution density of cultivated land and soil fertility.

[0146] S432, for each node in the initial graph structure data, extract the corresponding ecological information from the ecological monitoring data, and use each ecological information as the node feature corresponding to each node to update the node feature.

[0147] The ecological information corresponding to each node is the data value of that node in the ecological monitoring data, such as the temperature value, precipitation, and slope value corresponding to each node.

[0148] Specifically, this ecological information is added to the node characteristics of each node, in order to update the information of ecological monitoring data in the graph structure data.

[0149] S433: Based on the interaction relationships between various natural resources, obtain the relationships between nodes in the initial graph structure data, and update the edge features between corresponding nodes based on the relationships between nodes.

[0150] Specifically, the edges of each node in each initial graph structure are replaced with the interaction relationships between each natural resource, so as to add the interaction relationships between each natural resource in the updated graph structure.

[0151] S440 performs feature extraction based on updated graph structure data to obtain several global and local features.

[0152] For example, a trained graph neural network can be used for feature extraction. The graph neural network can be DCRNN or ST-GNN, etc. It should be noted that those skilled in the art should know the construction and training methods of graph neural networks, and this embodiment does not make specific limitations here.

[0153] Specifically, through graph neural network, several local and global features are obtained based on each updated graph structure. Local features are used to characterize the features extracted from at least some regions of each updated graph structure, while global features are used to characterize the features extracted from each updated graph structure as a whole.

[0154] It should be noted that graph structure neural networks extract features based on actual ecological analysis needs. For example, for arable land, if the ecological analysis need is to predict the yield of a certain crop, then the corresponding crop planting area, as well as features related to the growth status of the crop such as precipitation and sunlight, are extracted and updated from the graph structure data as global and local features. For water resources, if the ecological analysis need is to analyze the water conservation capacity, then the topography, precipitation, vegetation cover, and other features of the target area are extracted and updated from the graph structure data as global and local features.

[0155] S450 obtains ecological analysis results based on various global and local features through a graph-structured neural network.

[0156] Specifically, after the trained graph neural network extracts the global and local features, it fuses the global and local features corresponding to the same updated graph structure into an image. Based on the actual ecological analysis needs, the image is analyzed and arranged in time sequence as the ecological analysis result.

[0157] It should be noted that the graph structure neural network in steps S440 and S450 is a trained model. The graph structure neural network employs different training methods based on different practical ecological analysis needs. Specifically, for each type of natural resource, different features are extracted and updated from the graph structure based on different practical needs. The actual situation corresponding to the ecological analysis needs in the target area is obtained, and this actual situation is used as the dependent variable. Updated graph structure data is obtained based on the time-series remote sensing data of the corresponding year as the independent variable, input into the model to extract corresponding features. A fusion analysis is then performed based on these features to obtain the model's analysis results. The obtained actual situation is compared with the analysis results to optimize the graph structure neural network. For example, specific examples will be given below when the natural resource is arable land or water resources.

[0158] For arable land, if the ecological analysis requirement is to predict the yield of a certain crop, the training method includes: obtaining the actual yield of the crop in several years for several arable lands within the target area as the dependent variable; obtaining the corresponding updated map structure data based on the time-series remote sensing image data for those years; using this updated map structure data as the independent variable for model training; extracting the corresponding local and global features; fusing these local and global features; analyzing and obtaining the yield of the crop in the target area; and comparing it with the actual yield of the corresponding year to optimize the model. For the specific implementation method of obtaining the updated map structure data based on time-series remote sensing imagery as the independent variable, please refer to the foregoing content; this embodiment will not elaborate further.

[0159] Alternatively, for water resources, if the ecological analysis requirement is to analyze the water conservation capacity, the training method includes: obtaining the actual water conservation values ​​for several locations within the target area over any number of years as the dependent variable, and obtaining the corresponding updated graph structure data based on the time-series remote sensing image data for those years as the independent variable. These are then input into the graph neural network to be trained to extract the corresponding local and global features and perform fusion analysis to obtain the water conservation capacity. The model is then optimized based on the actual water conservation values. Specifically, please refer to the aforementioned training method for graph neural networks targeting arable land yield; this embodiment will not elaborate further.

[0160] It should be noted that the above is an illustrative example of training a graph neural network for arable land and water resources. It should also be noted that for other types of natural resources, those skilled in the art can refer to the above training process to train the graph neural network, which will not be elaborated here.

[0161] Based on this, by obtaining the interaction relationships between various natural resources and obtaining ecological analysis results based on these interaction relationships, the ecological analysis results can be used to analyze the situation of various natural resources and the entire ecosystem as a whole, so as to achieve a more comprehensive monitoring effect.

[0162] For example, such as Figure 15 As shown, it is displayed as based on Figure 3 The results of the distribution of natural resource quantities shown in the middle Figure 9 The displayed natural resource quality results and acquired ecological analysis results show that water conservation capacity has been affected by changes in the quantity and quality of arable land and water resources. Specifically, green areas represent areas where water conservation capacity remains unchanged, yellow areas represent areas where water conservation capacity has weakened, and red areas represent areas where water conservation capacity has significantly decreased. S500 outputs natural resource quantity distribution results, natural resource quality results, and ecological analysis results as natural resource remote sensing monitoring results.

[0163] Based on this, the natural resource remote sensing monitoring results of this embodiment can comprehensively analyze the situation of natural resources in terms of quantity, quality and ecological conditions, thereby achieving a more refined and comprehensive natural resource monitoring effect.

[0164] like Figure 16 As shown in the figure, the natural resource remote sensing integrated monitoring device provided in this embodiment includes a preprocessing module 61, a quantity analysis module 62, a quality analysis module 63, an ecological analysis module 64, and an output module 65.

[0165] The preprocessing module 61 is used to acquire time-series remote sensing data of the target area and perform preprocessing.

[0166] The quantity analysis module 62 is used to obtain the quantity information of each natural resource in the target area based on time-series remote sensing data and through a natural resource identification model, as the result of natural resource quantity distribution.

[0167] The quality analysis module 63 is used to construct a set of remote sensing feature indices; based on time-series remote sensing data and combined with the set of remote sensing feature indices, the remote sensing feature index data of the land parcels is calculated and obtained by a step-by-step calculation method; the remote sensing feature index data of the land parcels is input into a trained regression-physical model to obtain the quality information of each natural resource in the target area, which is used as the natural resource quality result; the remote sensing feature index set includes several types of remote sensing feature indices.

[0168] The ecological analysis module 64 is used to obtain initial graph structure data based on the results of natural resource quantity distribution and natural resource quality; extract the interaction relationships between various natural resources based on the initial graph structure data; update the initial graph structure data based on the interaction relationships between various natural resources to obtain updated graph structure data; use a graph structure neural network to extract features based on the updated graph structure data to obtain several global and local features; and obtain ecological analysis results based on the global and local features.

[0169] Output module 65 is used to output the results of natural resource quantity distribution, natural resource quality, and ecological analysis as the results of natural resource remote sensing monitoring.

[0170] Based on the same technical concept, the natural resource remote sensing integrated monitoring method provided in the embodiments of the present invention can be implemented on the terminal side or the server side.

[0171] like Figure 17 The diagram illustrates an optional hardware structure of a terminal according to an embodiment of the present invention. The terminal 70 can be a mobile phone, computer device, tablet device, personal digital processing device, factory back-end processing device, etc. The terminal 70 includes at least one processor 71, a memory 72, at least one network interface 74, and a user interface 73. The various components in the device are coupled together via a bus system 75. It is understood that the bus system 75 is used to realize communication between these components. In addition to a data bus, the bus system 75 also includes a power bus, a control bus, and a status signal bus.

[0172] The user interface 73 may include a monitor, keyboard, mouse, trackball, clicker, button, touchpad, or touch screen.

[0173] It is understood that memory 72 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM) or programmable read-only memory (PROM), used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memory characterized in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable categories of memory.

[0174] In this embodiment of the invention, the memory 72 is used to store various types of data to support the operation of the terminal. Examples of this data include: any executable program for operation on the terminal 70, such as the operating system 721 and application programs 722; the operating system 721 contains various system programs, such as the framework layer, core library layer, driver layer, etc., for implementing various basic services and handling hardware-based tasks. The application program 722 may contain various applications, such as media players, browsers, etc., for implementing various application services. The implementation of the natural resource remote sensing integrated monitoring method provided in this embodiment of the invention can be included in the application program 722.

[0175] The methods disclosed in the above embodiments of the present invention can be applied to processor 71, or implemented by processor 71. Processor 71 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 71 or by instructions in the form of software. The processor mentioned above may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 71 can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present invention. Processor 71 may be a microprocessor or any conventional processor, etc. The steps of the accessory optimization method provided in the embodiments of the present invention can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium, which is located in a memory. The processor reads the information in the memory and combines it with its hardware to complete the steps of the aforementioned method.

[0176] In an exemplary embodiment, terminal 70 may be used by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs) to execute the aforementioned method.

[0177] This invention also provides a computer-readable storage medium storing a computer program that, when invoked by a processor, implements the natural resource remote sensing integrated monitoring method provided by this invention.

[0178] Computer-readable storage media can be tangible devices capable of holding and storing instructions used by an instruction execution device. Computer-readable storage media can be, for example, (but not limited to) electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, and mechanical encoding devices.

[0179] The computer-readable program represented herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network, to an external computer or external storage device. A network adapter card or network interface in each computing / processing device receives computer-readable program instructions from the network and forwards these instructions to the computer-readable storage medium in the respective computing / processing device.

[0180] In summary, this application obtains quantitative information on various natural resources, yields results on the distribution of natural resource quantities, obtains quality information on natural resources, and, based on the quantitative and quality results, identifies the interaction relationships between various natural resources to reflect their correlation and mutual influence. Furthermore, it obtains ecological analysis results based on these interaction relationships, thereby facilitating a holistic analysis of the ecological situation of the entire target area and achieving a more refined and comprehensive monitoring effect.

[0181] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0182] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for integrated remote sensing monitoring of natural resources, comprising: Acquire temporal remote sensing data of the target area and perform preprocessing; Based on the aforementioned time-series remote sensing data, the quantity information of each natural resource within the target area is obtained through a natural resource identification model, which serves as the result of the natural resource quantity distribution. Construct a set of remote sensing feature indices; based on the time-series remote sensing data and the set of remote sensing feature indices, calculate and obtain the remote sensing feature index data of the land parcel using a step-by-step calculation method; The remote sensing feature index data of the land parcels are input into a trained regression-physical model to obtain the quality information of each natural resource within the target area, which serves as the natural resource quality result. The remote sensing feature index set includes several types of remote sensing feature indices. The regression-physical model includes a regression sub-model and a physical neural network sub-model. The regression sub-model and the physical neural network sub-model are connected together by a loss function. The regression sub-model is based on a CNN framework, and the physical neural network sub-model is based on a PINN framework. Based on the results of the quantity distribution and quality of natural resources, initial graph structure data is obtained; based on the initial graph structure data, the interaction relationships between the natural resources are extracted; based on the interaction relationships between the natural resources, the initial graph structure data is updated to obtain updated graph structure data; based on the updated graph structure data, feature extraction is performed to obtain several global and local features; based on the global and local features, ecological analysis results are obtained through a graph structure neural network; wherein, the step of extracting the interaction relationships between the natural resources based on the initial graph structure data includes: performing temporal convolution based on the initial graph structure data to obtain several node change feature vectors; performing graph convolution based on the node change feature vectors to obtain edge feature vectors between nodes; and using the edge feature vectors as the interaction relationships; The results of the quantity distribution of natural resources, the quality of natural resources, and the ecological analysis are output as the results of remote sensing monitoring of natural resources.

2. The method according to claim 1, characterized in that, The process of calculating and obtaining land parcel remote sensing feature index data based on the time-series remote sensing data and the remote sensing feature index set, using a step-by-step calculation method, includes: Based on the time-series remote sensing data, calculate the values ​​of the remote sensing feature indices of various classes corresponding to each pixel, and obtain pixel remote sensing feature index data; Pixel segmentation is performed based on the pixel remote sensing feature index data to obtain the remote sensing features of each object, and the object remote sensing feature index data is also obtained. Based on the aforementioned quantity distribution results, the remote sensing features of each object are clustered to obtain the remote sensing features of each land parcel, and the land parcel remote sensing feature index data is obtained.

3. The method according to claim 1, characterized in that, The construction of the remote sensing feature index set includes: Based on the quantity information of each natural resource in the natural resource quantity distribution results, each sampling grid is obtained; based on each sampling grid, field sampling is carried out to obtain the measured data corresponding to each sampling grid; Several types of initial remote sensing feature indices are selected; based on the time-series remote sensing data and combined with each of the sampling grids, the values ​​of the initial remote sensing feature indices of each type corresponding to each sampling grid are calculated; Based on the measured data, the correlation degree of each type of initial remote sensing feature index is analyzed to select the initial remote sensing feature indices of each type that meet the correlation degree requirements as each remote sensing feature index, and the remote sensing feature index set is formed based on the selected remote sensing feature indices.

4. The method according to claim 1, characterized in that, The process of updating the initial graph structure data based on the interaction relationships between the various natural resources to obtain updated graph structure data includes: Obtain ecological monitoring data; For each node in the initial graph structure data, the corresponding ecological information is extracted from the ecological monitoring data, and the ecological information is used as the node feature of the corresponding node for updating. Based on the interaction relationships between the natural resources, the relationships between the nodes in the initial graph structure data are obtained, and the edge features between the corresponding nodes are updated based on the relationships between the nodes.

5. The method according to claim 1, characterized in that, After obtaining the quality information of each of the natural resources within the target area, the method further includes: Acquire the quality change driving information of the target area, and input the quality change driving information and the quality information of each of the natural resources into the geographic interpretation model; The geographic interpretation model analyzes the causes of quality changes in each of the natural resources based on the quality change driving information, so as to update and output the quality information of each of the natural resources. The updated output quality information of each of the natural resources is taken as the natural resource quality result.

6. A comprehensive remote sensing monitoring device for natural resources, characterized in that, It includes a preprocessing module, a quantitative analysis module, a quality analysis module, an ecological analysis module, and an output module; The preprocessing module is used to acquire time-series remote sensing data of the target area and perform preprocessing. The quantity analysis module is used to obtain the quantity information of each natural resource in the target area based on the time-series remote sensing data and through a natural resource identification model, as the result of natural resource quantity distribution. The quality analysis module is used to construct a set of remote sensing feature indices; based on the time-series remote sensing data and in combination with the set of remote sensing feature indices, the remote sensing feature index data of the land parcel is calculated and obtained by a step-by-step calculation method. The remote sensing feature index data of the land parcels are input into a trained regression-physical model to obtain the quality information of each natural resource within the target area, which serves as the natural resource quality result. The remote sensing feature index set includes several types of remote sensing feature indices. The regression-physical model includes a regression sub-model and a physical neural network sub-model. The regression sub-model and the physical neural network sub-model are connected together by a loss function. The regression sub-model is based on a CNN framework, and the physical neural network sub-model is based on a PINN framework. The ecological analysis module is used to: obtain initial graph structure data based on the results of the quantity distribution of natural resources and the results of the quality of natural resources; extract the interaction relationships between the natural resources based on the initial graph structure data; update the initial graph structure data based on the interaction relationships between the natural resources to obtain updated graph structure data; perform feature extraction based on the updated graph structure data to obtain several global and local features; and obtain ecological analysis results through a graph structure neural network based on the global and local features. The step of extracting the interaction relationships between the natural resources based on the initial graph structure data includes: performing temporal convolution based on the initial graph structure data to obtain several node change feature vectors; performing graph convolution based on the node change feature vectors to obtain edge feature vectors between nodes; and using the edge feature vectors as the interaction relationships. The output module is used to output the results of the quantity distribution of natural resources, the results of the quality of natural resources, and the results of the ecological analysis, as the results of remote sensing monitoring of natural resources.

7. A terminal, characterized in that, include: A processor and a memory, wherein the memory and the processor are communicatively connected; The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory, so that the terminal performs the integrated remote sensing monitoring method for natural resources as described in any one of claims 1 to 5.

8. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the natural resource remote sensing integrated monitoring method as described in any one of claims 1 to 5.

Images