Degenerated ecosystem restoration method

Abstract

The embodiment of the invention discloses a degraded ecosystem restoration method, and the method comprises the steps: obtaining the original multi-modal data of a monitoring region, the original multi-modal data comprising point cloud data, hyperspectral image data, soil data, and water body data; acquiring real-time hydrometeorological data; preprocessing the original multi-modal data to obtain structured data, and aligning the structured data to obtain a fusion feature map; determining a degradation type of the monitoring area based on the fused feature map and the real-time hydro-meteorological data; querying candidate schemes from a pre-established knowledge base based on the degradation type; based on environmental constraint conditions, performing constraint condition check, multi-objective optimization and reward function screening on the candidate schemes to obtain an optimization scheme; generating long-term effect prediction data and risk assessment data based on the optimization scheme and derivative data obtained by processing the original multi-modal data; and executing the optimization scheme when the long-term effect prediction data and the risk assessment data meet requirements.

Description

Technical Field

[0001] This application relates to the field of ecological environment restoration technology, and relates to, but is not limited to, a method for restoring degraded ecosystems. Background Technology

[0002] With the diverse types of degradation in the Yunnan Plateau ecosystem, such as rocky desertification, lake eutrophication, and mine damage, traditional methods have three major drawbacks. First, the governance is fragmented: traditional methods divide governance units according to administrative boundaries (such as the four-category zoning of rocky desertification in eastern Yunnan), ignoring the material cycle correlation between mountains and lakes. Second, the dynamic response is lagging: relying on manual monitoring (such as quarterly vegetation surveys) cannot capture sudden environmental changes in real time (such as water depletion during the dry season). Finally, the technology is linearly superimposed: in mine restoration, the backfilling of topsoil and the vegetation configuration are separated, resulting in a large secondary collapse rate. Summary of the Invention

[0003] In view of this, embodiments of this application provide a method for restoring degraded ecosystems, which at least addresses the problem of lag in dynamic response.

[0004] The technical solution of this application embodiment is implemented as follows: In a first aspect, embodiments of this application provide a method for restoring a degraded ecosystem, the method comprising: Acquire raw multimodal data of the monitoring area, including point cloud data collected by the lidar sensor carried by the UAV, hyperspectral image data collected by the hyperspectral imager carried by the UAV, soil data collected by the soil moisture meter in the IoT sensor network, and water body data measured by the water quality buoy in the IoT sensor network. Acquire real-time hydrological and meteorological data collected by real-time environmental sensors in the monitored area; The original multimodal data is preprocessed to obtain structured data, and the structured data is aligned using a spatiotemporal registration algorithm to obtain a fused feature map. Based on the fused feature map and the real-time hydrological and meteorological data, the degradation type of the monitoring area is determined; Based on the degradation type, at least one corresponding candidate solution is queried from the pre-established knowledge base; based on the environmental constraints of the monitoring area, the at least one candidate solution is subjected to constraint checks, multi-objective optimization, and reward function screening to obtain an optimized solution; Based on the optimization scheme and the derived data obtained from the original multimodal data, long-term effect prediction data and risk assessment data corresponding to the optimization scheme are generated. When the long-term effect prediction data meets the preset ecological goals and the risk assessment data meets the preset safety threshold, the optimization scheme is executed to restore the degraded ecosystem.

[0005] The beneficial effects of the technical solutions provided in this application include at least the following: By real-time sensing of multimodal raw data, including drone lidar point clouds, hyperspectral images, and IoT soil and water data, combined with real-time hydrological and meteorological data, covering key dimensions such as topography, vegetation, soil, and hydrology, comprehensive data support is provided for subsequent analysis, avoiding the limitations of single data sources. Real-time sensing of multimodal raw data and real-time hydrological and meteorological data improves the dynamic response rate, enabling a comprehensive reflection of the ecological status of the entire monitoring area, rather than being limited to administrative boundaries. In subsequent analysis and scheme formulation, considering this comprehensive data allows for a full understanding of the material cycle correlations between different ecological elements such as mountains and lakes. A complete closed loop is constructed, encompassing data preprocessing, degradation type identification, candidate scheme screening, effect and risk prediction, and scheme implementation. Combined with environmental constraints, multi-objective optimization, and reward function screening, this ensures that optimized schemes balance feasibility, economy, and ecological benefits, reducing subjective decision-making bias. Long-term effect prediction data and risk assessment data are generated in advance before scheme implementation, ensuring execution only when ecological goals and safety thresholds are met, avoiding secondary ecological damage or resource waste caused by blind implementation, and guaranteeing the reliability and safety of the restoration process. Attached Figure Description

[0006] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 A schematic flowchart illustrating a method for restoring a degraded ecosystem, provided as an embodiment of this application; Figure 2 This is a flowchart illustrating an intelligent planning method for restoration paths of degraded ecosystems, provided in an embodiment of this application. Detailed Implementation

[0007] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0008] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0009] It should be noted that the terms "first, second, and third" used in the embodiments of this application are merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0010] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of this application pertain. It should also be understood that terms such as those defined in general dictionaries should be understood to have a meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0011] This application provides a method for restoring degraded ecosystems, applied to electronic devices. These electronic devices include, but are not limited to, mobile phones, laptops, tablets and handheld internet devices, multimedia devices, streaming media devices, mobile internet devices, wearable devices, or other types of electronic devices. The functions implemented by this method can be achieved by a processor in the electronic device calling program code. The program code can be stored in a computer storage medium; therefore, the electronic device includes at least a processor and a storage medium. The processor can be used to perform processing based on a degraded ecosystem restoration process, and the memory can be used to store data required and generated during the degraded ecosystem restoration process.

[0012] Figure 1 This is a flowchart illustrating a method for restoring a degraded ecosystem, as provided in an embodiment of this application. Figure 1 As shown, the method includes at least the following steps: Step S110: Obtain raw multimodal data of the monitoring area. The raw multimodal data includes point cloud data collected by the lidar sensor mounted on the UAV, hyperspectral image data collected by the hyperspectral imager mounted on the UAV, soil data collected by the soil moisture meter in the IoT sensor network, and water body data measured by the water quality buoy in the IoT sensor network; obtain real-time hydrological and meteorological data collected by the real-time environmental sensor in the monitoring area. This system can collect multi-type and multi-modal data related to the ecosystem within the monitoring area. The lidar sensor, also known as a LiDAR sensor, can penetrate the vegetation canopy to characterize micro-topography with an accuracy of ±5 cm. The point cloud data is used to obtain information such as terrain and elevation. The hyperspectral imager can identify 132 spectral bands, and the hyperspectral image data can be used to analyze vegetation types. Various sensors in the IoT sensor network can transmit the collected monitoring data to the backend data center or platform in real time. It can also collect hydrological and meteorological data closely related to the ecosystem through dedicated real-time environmental sensors. The hydrological and meteorological data includes data such as precipitation, evaporation, wind speed, humidity, and runoff, which have real-time changing characteristics.

[0013] Step S120: Preprocess the original multimodal data to obtain structured data, and align the structured data using a spatiotemporal registration algorithm to obtain a fused feature map; One approach is to use a spatiotemporal registration algorithm to fuse structured data from different sources and generate a unified fusion feature map.

[0014] Step S130: Based on the fused feature map and the real-time hydrological and meteorological data, determine the degradation type of the monitoring area; The degradation type may include rocky desertification, lake eutrophication, mine damage, etc.; in some embodiments, the degradation type includes degradation level and geographic coordinates.

[0015] Step S140: Based on the degradation type, query at least one corresponding candidate solution from the pre-established knowledge base; based on the environmental constraints of the monitoring area, perform constraint checks, multi-objective optimization, and reward function screening on the at least one candidate solution to obtain an optimized solution; Among them, the corresponding candidate solutions in the knowledge base can be called according to the degradation type. For example, the candidate solution for rocky desertification area is "drone seedling transport and fish scale pit water collection". Then, dynamic constraint check, multi-objective optimization and reward function screening are carried out, and finally the optimized solution is output, such as "sesbania planting combined with photovoltaic water collection". Compared with the traditional restoration solution, the unit cost per hectare of this optimized solution is reduced by 15% to 30%, and the water retention rate is increased by 30%.

[0016] Step S150: Based on the optimization scheme and the derived data obtained from the original multimodal data, generate long-term effect prediction data and risk assessment data corresponding to the optimization scheme; The long-term effect prediction data may include the increase in vegetation coverage over 10 years, changes in soil fertility, etc., and the risk assessment data may include slope collapse efficiency, cyanobacterial bloom risk, etc.

[0017] Step S160: When the long-term effect prediction data meets the preset ecological goals and the risk assessment data meets the preset safety threshold, the optimization scheme is executed to restore the degraded ecosystem.

[0018] The ecological goal can be to increase vegetation coverage by 30%, and the safety threshold can be that the probability of slope collapse is less than 5%. If the long-term effect prediction data meets the preset ecological goal and the risk assessment data meets the preset safety threshold, the optimization plan will be officially implemented and the ecological restoration process will be carried out.

[0019] In the above embodiments, multimodal raw data such as real-time sensing of UAV lidar point clouds, hyperspectral images, and IoT soil and water data, combined with real-time hydrological and meteorological data, are used to cover key dimensions such as topography, vegetation, soil, and hydrology. This provides comprehensive data support for subsequent analysis, avoiding the limitations of single data sources. By sensing multimodal raw data and real-time hydrological and meteorological data in real time, the dynamic response rate is improved, enabling a comprehensive reflection of the ecological status of the entire monitoring area, rather than being limited to administrative boundaries. In subsequent analysis and scheme formulation, considering these comprehensive data allows for a full understanding of the material cycle correlations between different ecological elements such as mountains and lakes. A complete closed loop is constructed, encompassing data preprocessing, degradation type identification, candidate scheme screening, effect and risk prediction, and scheme execution. Combined with environmental constraints, multi-objective optimization, and reward function screening, this ensures that the optimized scheme balances feasibility, economy, and ecological benefits, reducing subjective decision-making bias. Long-term effect prediction data and risk assessment data are generated in advance before scheme execution, and execution is only carried out when ecological goals and safety thresholds are met, avoiding secondary ecological damage or resource waste caused by blind implementation and ensuring the reliability and safety of the restoration process.

[0020] In some embodiments, the method further includes: Step S161: Obtain the actual implementation effect data of the optimization scheme monitored by the soil moisture meter, the water quality buoy, the real-time environmental sensor and the vegetation monitoring equipment in the IoT sensor network within the target time period; The actual implementation effect can be real data on vegetation survival rate or soil erosion resistance.

[0021] Step S162: Based on the long-term effect prediction data, the first prediction data within the target duration, and the actual implementation effect data, a deviation analysis is performed to obtain the deviation analysis results. This involves collecting monitoring data from various devices within the IoT sensor network over a target duration. This data is related to the actual implementation effect of the optimization plan. The target duration can be 1 year, 3 years, etc., matching the time dimension of the long-term effect prediction data. Deviation analysis can be performed from dimensions such as numerical deviation, spatial deviation, and temporal deviation to obtain the deviation analysis results.

[0022] Step S163: Based on the deviation analysis results, determine the source of error; Step S164: Based on the source of error, update the technical parameters in the optimization scheme in the knowledge base.

[0023] Based on the deviation analysis results, the specific reasons for the deviation can be traced. In one embodiment, if the actual vegetation coverage is lower than the predicted value, the source of error may be unreasonable scheme parameters (such as too low seedling planting density), deviation of the preset model assumptions (such as the model assuming 800 mm of precipitation, but the actual precipitation is only 600 mm), external interference (such as seedling death caused by pests and diseases), etc.

[0024] In another embodiment, the first prediction data may include a predicted seabuckthorn survival rate of 80%, and the actual landing effect data may include an actual seabuckthorn survival rate of 70%. If the analysis shows that the source of error is that the soil pH value is greater than 9.0, which exceeds the tolerance range, then the pH range in the "seabuckthorn combined with biochar" scheme can be modified to 6.5 to 8.5.

[0025] In the above embodiments, by acquiring the actual implementation effect data of the optimized solution monitored by the IoT sensor network, and performing deviation analysis and error source location with the long-term effect prediction data, dynamic optimization of the solution can be achieved, avoiding the decline in adaptability caused by the solidification of the solution; at the same time, the technical parameters of the optimized solution in the knowledge base are updated based on the error source, and effective experience is continuously accumulated to form a virtuous cycle of implementation, feedback and iteration, which improves the accuracy of subsequent design of similar degraded area solutions and further ensures the success rate of degraded ecosystem restoration.

[0026] In some embodiments, the soil data includes soil pH and organic matter content, and the water data includes total nitrogen concentration and total phosphorus concentration. Step S120, "preprocessing the original multimodal data to obtain structured data," includes: Step S1201: Use a point cloud classification algorithm to classify the point cloud data to obtain the bare rock ratio of the monitoring area; One approach is to use point cloud classification algorithms to categorize point cloud data collected by UAV LiDAR, and finally calculate the bare rock rate. The point cloud classification algorithm will classify point cloud data into categories such as bare rock points, soil points, and vegetation points based on laser reflection intensity and spatial distribution characteristics. Then, by statistically analyzing the proportion of the surface area corresponding to the bare rock points to the total area of ​​the monitored area, the bare rock rate can be obtained.

[0027] Step S1202: Using a digital elevation model generation algorithm, the slope of the monitoring area is generated based on the point cloud data; Point cloud data includes elevation information for each point. The digital elevation model (DEM) generation algorithm grids these discrete elevation points to form a continuous digital elevation model. The algorithm then calculates the tilt angle of each grid cell to obtain the slope distribution data of the entire monitoring area.

[0028] Step S1203: Based on the hyperspectral image data, determine the vegetation type and vegetation growth status of the monitoring area; Among them, by analyzing hyperspectral image data, the vegetation type of the monitoring area can be identified and the vegetation growth status can be determined. Since different vegetation has different spectral characteristics, the spectral curve of each pixel in the hyperspectral image data can be compared with the spectral library of known vegetation to determine the vegetation type. At the same time, the vegetation growth status can be determined by the vegetation index in the hyperspectral image data. In one embodiment, pioneer plants and invasive species can be distinguished. The pioneer plants can be plants that can first settle and grow in degraded, barren or extreme environments (such as bare rock, slag heaps, and land after fire), such as Rumex acetosa.

[0029] Step S1204: Integrate the bare rock ratio, slope, vegetation type, vegetation growth status, pH value, organic matter content, total nitrogen concentration, and total phosphorus concentration to form the structured data.

[0030] After obtaining the structured data, the structured data can be aligned to obtain a 200×200×138-dimensional gridded fusion feature map. Channels 0 to 2 correspond to terrain slope or bare rock ratio; channels 3 to 134 correspond to 132 spectral bands of hyperspectral image data; and channels 135 to 137 correspond to soil pH, total nitrogen (TN) concentration and total phosphorus (TP) concentration of water.

[0031] In the above embodiments, the preprocessing of the original multimodal data uses point cloud classification algorithms, digital elevation model generation algorithms, and hyperspectral image analysis to accurately extract core ecological indicators such as bare rock ratio, slope, vegetation type, and growth status. At the same time, it integrates structured data such as soil pH value, organic matter content, and water total nitrogen and total phosphorus concentrations to eliminate format differences and redundant information from different sources, forming unified structured data. This facilitates subsequent spatiotemporal registration and fusion feature map generation, and also focuses on key data dimensions, improving the efficiency and accuracy of subsequent degradation type identification and scheme selection.

[0032] In some embodiments, step S130, "determining the degradation type of the monitoring area based on the fused feature map and the real-time hydrological and meteorological data," includes the following steps: Step S1301: Perform differential convolution operation on each channel of the fused feature map to obtain an intermediate feature map; Differential convolution operations can be performed on each channel of the fused feature map to extract backbone features. For example, the fused feature map can be reduced to an intermediate feature map of 100×100×256 by using a 7×7 convolution (stride 2) in MobileNetV3.

[0033] Step S1302: Perform weight adaptive enhancement on the height and width dimensions of the intermediate feature map to obtain the reconstructed feature map; In this process, adaptive weights can be calculated for the height and width dimensions of the intermediate feature map, and then the weights can be used to enhance the intermediate feature map to obtain the reconstructed feature map.

[0034] Step S1303: Based on the real-time hydrological and meteorological data, generate an environmental data vector; concatenate the reconstructed feature map and the environmental data vector to obtain a concatenated feature map; The real-time hydrological and meteorological data can be real-time precipitation, real-time evaporation, etc. The real-time hydrological and meteorological data can be encoded into a three-dimensional environmental data vector. The reconstructed feature map and the environmental data vector are spliced ​​together to obtain a spliced ​​feature map.

[0035] Step S1304: Generate a dynamic weight matrix based on the stitched feature map; generate a degradation heatmap based on the dynamic weight matrix and the stitched feature map; Specifically, a 1×1 convolution operation can be performed on the stitched feature map to generate a dynamic weight matrix that matches the spatial dimension of the stitched feature map. The dynamic weight matrix is ​​then normalized using a Sigmoid activation function to obtain a normalized weight matrix. The stitched feature map is then multiplied element-wise with the normalized weight matrix to generate a 200×200×5 degradation heatmap representing the degradation state of the monitoring area. The 5 values ​​correspond to the degradation probability distributions of 5 preset degradation types, and each channel stores the probability value of each grid cell in the monitoring area belonging to the corresponding degradation type.

[0036] Step S1305: Analyze the degradation heat map to obtain the degradation type of the monitoring area.

[0037] Among them, grid cells with probability values ​​exceeding the preset threshold can be selected from 5 types of degradation probability distributions and regarded as high-risk areas. Feature matching and classification of high-risk areas of degradation heatmap can be performed to determine the specific degradation type of the monitoring area.

[0038] In the above embodiments, when determining the degradation type of the monitoring area, key information of the fusion feature map is enhanced by differential convolution, and the reconstructed feature map is generated by adaptively combining the weights of height and width dimensions to highlight the spatial features related to degradation. At the same time, real-time hydrological and meteorological data are integrated to generate environmental data vectors, and a degradation heat map is constructed through a dynamic weight matrix. This allows the degradation type identification to take into account both static features and dynamic environmental factors, and the results are more consistent with the actual degradation state of the area. Furthermore, the visualization of the heat map facilitates rapid analysis of the degradation type and reduces the difficulty of subsequent scheme matching.

[0039] In some embodiments, step S1302, "performing weight adaptive enhancement on the height and width dimensions of the intermediate feature map to obtain a reconstructed feature map," includes the following steps: Step S13021: Perform global pooling on the height dimension of the intermediate feature map to obtain a highly compressed feature map; generate a height weight matrix based on the highly compressed feature map and a one-dimensional convolution. Specifically, global pooling is performed on the height dimension of the 100×100×256 intermediate feature map to compress it into a 1×100×256 height compressed feature map. A one-dimensional convolution operation with a kernel size of 3 is performed on the height compressed feature map to obtain an initial height weight matrix. The initial height weight matrix is ​​then normalized using the Sigmoid activation function to finally generate a 1×100×1 height weight matrix.

[0040] Step S13022: Perform global pooling on the width dimension of the intermediate feature map to obtain a width-compressed feature map; generate a width weight matrix based on the width-compressed feature map and one-dimensional convolution. Similarly, global pooling of the width dimension of the 100×100×256 intermediate feature map can compress the intermediate feature map into a 100×1×256 width compressed feature map. A one-dimensional convolution operation with a kernel size of 3 is performed on the width compressed feature map to obtain an initial width weight matrix. The initial width weight matrix is ​​then normalized by the Sigmoid activation function to finally generate a 100×1×1 width weight matrix.

[0041] Step S13023: Based on the height weight matrix and the width weight matrix, perform feature reconstruction on the intermediate feature map to obtain the reconstructed feature map.

[0042] The intermediate feature map can be reconstructed by weighting it element-by-element using the normalized height weight matrix and width weight matrix.

[0043] In the above embodiments, global pooling and one-dimensional convolution are performed on the height and width dimensions of the intermediate feature map to generate targeted weight matrices. This enables specific enhancement of different spatial locations (such as steep slope areas and vegetation degradation areas) and effectively suppresses interference from irrelevant spatial features. By reconstructing the intermediate feature map element by element through the height and width weight matrices, key features related to degradation are further amplified, providing high-quality feature input for subsequent splicing feature map generation and degradation heatmap construction. This improves the accuracy of degradation type identification and adapts to the spatial differences of different monitoring areas, enhancing the algorithm's versatility.

[0044] In some embodiments, step S140, "based on the environmental constraints of the monitoring area, performing constraint checks, multi-objective optimization, and reward function screening on the at least one candidate solution to obtain an optimized solution," includes the following steps: Step S1401: Based on the environmental constraints of the monitoring area, perform constraint checks on the at least one candidate scheme, eliminate candidate schemes that do not meet the constraints, and obtain at least one sub-candidate scheme. Among these measures, infeasible solutions can be eliminated through constraint checks. These constraints may include prohibiting terraced field construction when the slope is greater than 25°, or prohibiting construction within nature reserves.

[0045] Step S1402: Based on the at least one sub-candidate scheme, use a multi-objective optimization algorithm to generate at least one target candidate scheme with the optimization objectives of minimum cost, maximum ecological benefit and minimum recovery period; The multi-objective optimization algorithm can be the NSGA-Ⅲ algorithm with "Min (cost), Max (ecological benefit), Min (recovery cycle)" as the optimization objectives.

[0046] Step S1403: With the goal of maximizing the reward function value, adjust the technical parameters of each target candidate scheme based on the initial strategy, monitor the changing trend of the reward function value of each target candidate scheme in real time and iterate continuously until the reward function value of each target candidate scheme reaches the optimal value under the current environment, and use the technical parameters at this time as the initial technical parameters of the target candidate scheme. The initial strategy is determined based on the structured data and the real-time hydrological and meteorological data. Step S1404: Based on the structured data, the real-time hydrological and meteorological data, and the initial technical parameters in each target candidate scheme, calculate the actual cost, actual ecological benefits, and actual recovery cycle of each target candidate scheme; Step S1405: Based on the actual cost, actual ecological benefits, and actual recovery period of each target candidate scheme, and the weights corresponding to cost, ecological benefits, and recovery period in the preset reward function, determine the reward function value of each target candidate scheme. Step S1406: Based on the reward function values ​​of each target candidate solution, select an optimized solution from the at least one target candidate solution.

[0047] The technical parameters of each target candidate scheme can be adjusted based on the structured data and the real-time hydrological and meteorological data. For example, if the precipitation is less than 800 mm, the photovoltaic water lifting power can be increased by 30%, and the optimal scheme can be output in the end.

[0048] In the above embodiments, during the candidate scheme screening and optimization process, infeasible schemes are first eliminated through environmental constraint checks to reduce ineffective optimization inputs; then, multi-objective optimization is carried out with the goal of "minimum cost, maximum ecological benefit, and minimum recovery period" as the objective. Combining the initial strategy based on structured data and real-time hydrological and meteorological data, the technical parameters are iteratively adjusted to the optimal reward function value to ensure that the parameters are adapted to the current environment; finally, the balance between economic, ecological, and time objectives is achieved through actual cost, benefit, and cycle calculations and reward function scoring, ensuring the feasibility and comprehensive benefits of the final optimized scheme.

[0049] In some embodiments, the derived data includes soil data, flood data, plant data, and watershed geographic data; the risk assessment data includes a risk heat map; and step S150, "based on the optimization scheme and the derived data obtained from the original multimodal data, generating long-term effect prediction data and risk assessment data corresponding to the optimization scheme," includes the following steps: Step S1501: Based on the optimization scheme, the soil data, and the watershed geographic data, the first model is used to determine the predicted data for spatial changes in organic matter. The first model can be a CENTURY model, which uses input scheme parameters to simulate the 10-year change in organic matter.

[0050] Step S1502: Based on the optimization scheme, the flood data, and the watershed geographic data, the second model is used to determine the spatial variation prediction data of soil erosion. The second model can be an LSTM hydrological model used to predict soil erosion.

[0051] Step S1503: Based on the optimization scheme, the plant data, and the watershed geographic data, the third model is used to determine the predicted data for spatial changes in vegetation cover. The third model can be a vegetation growth model or a soil erosion resistance model, used to predict changes in vegetation cover or soil erosion resistance.

[0052] Step S1504: Based on the predicted data of spatial changes in organic matter, the predicted data of spatial changes in soil and water loss, the predicted data of spatial changes in vegetation cover, and the preset risk level assessment rules, generate the risk heat map. The risk heat map indicates the probability of cyanobacterial blooms or slope collapses in the monitored area.

[0053] The risk level assessment rules may include a vegetation coverage rate of less than 40% and a soil erosion rate greater than 50 t / km². 2 There is a high risk of slope collapse, and there is a high risk of cyanobacterial blooms when the organic matter content is less than 2% and the water temperature is greater than 25℃.

[0054] Therefore, potential risks can be assessed based on the forecast results and risk level assessment rules.

[0055] In the above embodiments, based on the optimized scheme and derived data, multiple models are used to predict the spatial changes in organic matter, soil erosion, and vegetation cover changes, comprehensively covering the three core dimensions of restoration effect: soil quality, soil and water conservation, and vegetation restoration. At the same time, combined with preset risk level assessment rules, the multi-dimensional prediction data are integrated into a risk heat map that marks the probability of cyanobacterial blooms and slope collapse. This not only clarifies the long-term ecological effects of the scheme in advance, but also provides an intuitive warning of potential risks, providing a basis for safety verification before the scheme is implemented and avoiding ecological risks after implementation.

[0056] In one embodiment, this application provides a method for intelligent planning of restoration paths for degraded ecosystems, such as... Figure 2 As shown, the method includes the following steps: Step S1: Multimodal data perception and fusion; The target area is the steep slope rocky desertification area in Dongchuan, Yunnan (slope 35°, bare rock rate 70%). The area suffers from severe soil erosion during the rainy season (topsoil loss rate reaches 5t / hectare·month). The vegetation is mainly composed of Rumex japonicus (coverage rate 18%), and the natural recovery cycle is over 10 years. The core issues are the need to solve the problems of low vegetation survival rate (<60% with traditional methods) and the difficulty of dynamically adapting technical solutions to environmental changes.

[0057] First, the raw data of the area is received, including raw point cloud from UAV LiDAR (point density 200 points / ㎡), UAV hyperspectral image (132 bands, resolution 0.5m), and real-time data stream from IoT sensors (20 soil moisture meters + 5 weather stations, sampling frequency 10 minutes / time). Next, the LiDAR point cloud data was processed. The slope was calculated using the moving surface fitting method (window radius 5m) (because this algorithm can accurately capture the micro-topography of steep slopes). The bare rock ratio was extracted using the bare rock identification algorithm (based on the point cloud reflection intensity threshold of 0.8). The output slope value was 35.2° and the bare rock ratio was 72%. Then, the hyperspectral image was processed, and a continuum removal method (which can eliminate background noise interference) was performed on the 132-band image. The characteristic peak of Rumex japonicus (reflectance 0.58) was detected in the 73-band (wavelength 680nm), and its coverage was calculated to be 18% by maximum likelihood classification. Further processing of IoT sensor data involves applying a sliding window mean filter (10-minute window size to smooth high-frequency noise) to soil pH (8.0-9.5) and organic matter (1.2%-2.5%) data, outputting the current values: pH=8.9, organic matter 1.8%. Finally, data fusion was performed. Spatial alignment was achieved using WGS84 coordinate transformation (error < 0.5m), and time matching was achieved using a timestamp synchronization algorithm (deviation < 1 minute). The three source data were integrated into a 200×200 grid (grid size 10m×10m). The output is a 200×200×138 dimensional gridded feature map (channel 0: slope; channel 1: bare rock ratio; channels 2-133: reflectance of 132 spectral bands; channel 134: soil pH; channel 135: soil organic matter; channel 136: real-time precipitation; channel 137: wind speed).

[0058] Step S2: Dynamic classification of degradation types; First, receive the 138-channel feature map (including geographic coordinate information) output by S1. Next, backbone features were extracted. A 3×3 depth-separable convolution was applied to the terrain channel (0-1) (output channel 64, stride 1, to enhance the gully edge features of the 35° steep slope), a 1×1 pointwise convolution was applied to the vegetation channel (2-133) (output channel 128, used to amplify the spectral response of Rumex acetosa in the 73 band), and a 5×5 dilated convolution was applied to the soil and meteorological channel (134-137) (dilution rate 2, to correlate the spatial correlation between pH value and organic matter loss). Then, dimension-specific enhancement is performed. Global height pooling (compressed along the y-axis) is applied to the fused feature map (intermediate feature map) to generate a 256×100 vector. This vector is then convolved with 1D (kernel size 3, output channels 256) to generate a height weight matrix. Simultaneously, global width pooling (compressed along the x-axis) is applied to generate a 256×100 vector. This vector is then convolved with 1D (with the same parameters) to generate a width weight matrix. Finally, the features are reconstructed using "original feature map × Sigmoid (height weight) × Sigmoid (width weight)" (preserving the spatial orientation of the steep slope crack). Then, environmental coupling weighting is performed, and the real-time precipitation data (3mm) is encoded into a three-dimensional vector [0,0,1] (corresponding to "no rain / light rain / moderate rain"). After being concatenated with the reconstructed feature map, the resulting concatenated feature map is generated by a 1×1 convolution (output channel 1) to generate dynamic weights. By "concatenated feature map × Sigmoid (dynamic weights)", the weight of the soil channel (134-135) is increased from 0.3 to 0.8 (because precipitation may exacerbate soil erosion, it is necessary to strengthen soil features). The output is a 200×200×5 degradation heatmap (probability distribution of 5 degradation types), in which the area with coordinates (X:102.3°E, Y:26.5°N) is marked as "severe rocky desertification" (probability 92%), and is accompanied by a confidence heatmap with pixel-level error <3% (i.e. degradation heatmap).

[0059] Step S3, adaptive optimization of the recovery path; First, receive the coordinates and attributes of the "severely rocky desertification" area (slope 35.2°, pH 8.9, precipitation 3mm) output by S2. Next, knowledge base matching is performed. The "severe rocky desertification" entry in the technology combination knowledge base is called to match the basic technology "drone seedling transport + fish scale pit water collection". At the same time, dynamic constraint checks are performed: because the slope 35.2°>25°, the rule "disable terraced field project" is triggered (to avoid secondary collapse caused by steep slope terraces), and candidate schemes containing terraces are excluded. Then, multi-objective optimization was performed, with "Min (cost), Max (3-year vegetation survival rate), Min (recovery cycle)" as the objectives. The NSGA-III algorithm (population size 50, 30 iterations) was used to generate 50 candidate schemes. Then, a reinforcement learning agent (state variables: real-time precipitation 3mm, soil pH 8.9) selected the schemes according to the reward function (0.6×survival rate + 0.3×cost savings - 0.1×cycle) and adjusted the photovoltaic water pumping power to 1.3 times the rated value (due to low precipitation, irrigation needs to be enhanced). The output result is the Pareto optimal solution - "Senecio planting (density 2 plants / ㎡) + fish scale pit water collection (pit depth 0.5m, spacing 2m) + photovoltaic irrigation (power 300W / hectare)", cost ¥78,000 / hectare, predicted 3-year survival rate 85%, recovery period 1.5 years.

[0060] Step S4, digital twin watershed verification; First, receive the optimization scheme output by S3 and the Dongchuan precipitation database from 1993 to 2023 (annual average precipitation of 700 mm, including 32 events with daily precipitation >100 mm). Next, soil evolution simulation was performed. The CENTURY model was loaded, and the root depth of sesbania was 1.2m and the annual biomass was 3.5t / hectare (to simulate the effect of vegetation on soil improvement). The simulation simulated the change of soil organic matter over 10 years and the output showed that the organic matter content increased to 2.8% in the 5th year (baseline 1.8%). Then, hydrological risk prediction was carried out. An LSTM hydrological model was trained using 30 years of precipitation data (64 neurons in the input layer, 32 neurons in the hidden layer, and 100 iterations to accurately capture the temporal correlation between precipitation and soil erosion). The predicted soil erosion rate under extreme rainstorms (150 mm of daily precipitation) was 2.4 t / hectare·time (4.0 t / hectare·time in the traditional scheme). Regenerate a risk heat map, overlay it with a slope >30° area and a soil erosion resistance model (erosion resistance index K value decreased from 0.4 to 0.25), and output a GeoTIFF format risk map (resolution 10m×10m), marking the slope collapse probability of 12% in the area (X:102.3°E, Y:26.5°N) (28% in the traditional scheme). The output results include long-term effect predictions (85% vegetation survival rate and 55% increase in soil organic matter in 3 years) and risk heat maps with 3 levels of risk zones (low <5% / medium 5%-15% / high >15%).

[0061] Step S5: Dynamic feedback and knowledge base update; First, we receive S4's predicted 3-year survival rate of 85% and first-quarter (3-month) IoT monitoring data (average vegetation height 42cm, survival rate 72%, soil pH still 8.9). Next, a bias analysis was performed to calculate the predicted-measured survival rate bias (85%-72%=13%). Since it exceeded the threshold of 10%, root cause tracing was triggered. Then, root cause tracing was conducted, and the tolerance pool of sesbania species (pH tolerance upper limit 8.5) was compared with the measured pH of 8.9, confirming that "excessive soil alkalinity led to a low survival rate"; The knowledge base has been updated again, with the addition of the rule "When soil pH > 8.5, supplement with topsoil (dosage 0.5m). 3 / hectare, pH adjusted to 7.5), the revised plan is the original plan + topsoil improvement, and the knowledge base and rule base of the technology combination are updated simultaneously; The output is the updated technology combination knowledge base (version V1.2). The measured vegetation survival rate in the second season (months 4-6) increased to 86%, and the soil pH decreased to 8.2.

[0062] After one year of treatment, the bare rock rate in the Dongchuan rocky desertification area has decreased from 60% to 35%, and the vegetation coverage has increased to 70%. Compared with traditional methods, the restoration efficiency has increased by 40%, the cost has decreased by 33%, and the solution can be dynamically adapted to environmental changes during the dry and rainy seasons.

[0063] In this embodiment of the application, the cross-system collaborative model of "mountain-lake-mine" is used to solve the problem of fragmented material circulation caused by traditional governance based on administrative units. For example, the linkage between the governance of rocky desertification in Dongchuan and the control of eutrophication in Dianchi Lake reduces the transmission of non-point source pollution and breaks through the limitations of fragmented governance. The combination of real-time multimodal data perception and reinforcement learning optimization has shortened the solution update cycle from the traditional quarter to the minute level, effectively responding to emergencies such as water depletion during the dry season and erosion by torrential rain, and achieving an improvement in dynamic response capabilities. Digital twin platforms replace 30% of physical experiments, reducing trial-and-error costs. Meanwhile, multi-objective optimization algorithms improve ecological benefits (such as vegetation survival rate) by 20%-30% and shorten the recovery cycle by 15%-20%, achieving a balance between cost and efficiency. The coupling mechanism between engineering measures (such as UAV micro-terrain modification) and biological measures (such as the selection and mating of indigenous species) solves the secondary damage caused by the linear superposition of traditional technologies (such as reducing the collapse rate of topsoil backfill from 25% to 8% in mine restoration), and realizes collaborative technological innovation.

[0064] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0065] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0066] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0067] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected to achieve the purpose of the embodiments of this application according to actual needs. In addition, each functional unit in the embodiments of this application may be fully integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the integrated unit may be implemented in hardware or in the form of hardware plus software functional units.

[0068] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause the device automatic test line to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0069] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined to obtain new method embodiments without conflict. The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined to obtain new method embodiments or device embodiments without conflict.

[0070] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for restoring a degraded ecosystem, the method comprising: obtaining original multi-modal data of a monitoring area, the original multi-modal data comprising point cloud data collected by a laser radar sensor carried by a drone, hyperspectral image data collected by a hyperspectral imager carried by the drone, soil data collected by a soil moisture sensor in an IOT sensor network, and water body data measured by a water quality buoy in the IOT sensor network; obtaining real-time hydro-meteorological data collected by real-time environmental sensors in the monitoring area; preprocessing the original multi-modal data to obtain structured data, and aligning the structured data using a spatio-temporal registration algorithm to obtain a fusion feature map; determining a degradation type of the monitoring area based on the fusion feature map and the real-time hydro-meteorological data; based on the degradation type, querying at least one candidate solution corresponding to the degradation type from a pre-established knowledge base, and performing constraint condition checking, multi-objective optimization, and reward function screening on the at least one candidate solution based on environmental constraint conditions of the monitoring area to obtain an optimized solution; based on the optimized solution and derived data processed from the original multi-modal data, generating long-term effect prediction data and risk assessment data corresponding to the optimized solution; when the long-term effect prediction data meets a preset ecological target and the risk assessment data meets a preset safety threshold, executing the optimized solution to restore the degraded ecosystem.

2. The method of claim 1, wherein, The method further comprises: obtaining actual landing effect data of the optimized solution monitored by the soil moisture sensor, the water quality buoy, the real-time environmental sensors, and a vegetation monitoring device in the IOT sensor network within a target time period; based on first prediction data of the long-term effect prediction data within the target time period and the actual landing effect data, performing bias analysis to obtain a bias analysis result; based on the bias analysis result, determining an error source; based on the error source, updating technical parameters in the optimized solution in the knowledge base.

3. The method of claim 1, wherein, The soil data comprises pH value and organic matter content of the soil, and the water body data comprises total nitrogen concentration and total phosphorus concentration of the water body, and the preprocessing of the original multi-modal data to obtain the structured data comprises: classifying the point cloud data using a point cloud classification algorithm to obtain a bare rock rate of the monitoring area; generating a slope of the monitoring area based on the point cloud data using a digital elevation model generation algorithm; determining a vegetation type and a vegetation growth state of the monitoring area based on the hyperspectral image data; integrating the bare rock rate, the slope, the vegetation type, the vegetation growth state, the pH value, the organic matter content, the total nitrogen concentration, and the total phosphorus concentration to form the structured data.

4. The method of claim 1, wherein, The determination of the degradation type of the monitoring area based on the fusion feature map and the real-time hydro-meteorological data comprises: performing a differential convolution operation on each channel in the fusion feature map to obtain an intermediate feature map; performing weight self-adaptive reinforcement on a height dimension and a width dimension of the intermediate feature map respectively to obtain a reconstructed feature map; generate an environmental data vector based on the real-time hydro-meteorological data; and concatenate the reconstructed feature map and the environmental data vector to obtain a concatenated feature map; generate a dynamic weight matrix based on the concatenated feature map; and generate a degeneration heat map based on the dynamic weight matrix and the concatenated feature map; analyze the degeneration heat map to obtain a degeneration type of the monitoring area.

5. The method of claim 4, wherein, The height dimension and the width dimension of the intermediate feature map are respectively subjected to weight self-adaptive reinforcement to obtain a reconstructed feature map, including: perform global pooling on the height dimension of the intermediate feature map to obtain a height-compressed feature map; and generate a height weight matrix based on the height-compressed feature map and one-dimensional convolution; perform global pooling on the width dimension of the intermediate feature map to obtain a width-compressed feature map; and generate a width weight matrix based on the width-compressed feature map and one-dimensional convolution; perform feature reconstruction on the intermediate feature map based on the height weight matrix and the width weight matrix to obtain a reconstructed feature map.

6. The method of claim 1, wherein, The at least one candidate scheme is subjected to constraint condition checking, multi-objective optimization, and reward function screening based on the environmental constraint conditions of the monitoring area to obtain an optimized scheme, including: The at least one candidate scheme is subjected to constraint condition checking based on the environmental constraint conditions of the monitoring area to exclude candidate schemes that do not meet the constraints, thereby obtaining at least one sub-candidate scheme; At least one target candidate scheme is generated based on the at least one sub-candidate scheme by using a multi-objective optimization algorithm, with minimum cost, maximum ecological benefit, and minimum recovery period as optimization objectives; The technical parameters of each target candidate scheme are adjusted based on an initial strategy, the change trend of the reward function value of each target candidate scheme is monitored in real time and continuously iterated until the reward function value of each target candidate scheme reaches an optimal value under the current environment, the technical parameters at this time are taken as the initial technical parameters of the target candidate scheme, and the initial strategy is determined based on the structured data and the real-time hydro-meteorological data; The actual cost, actual ecological benefit, and actual recovery period of each target candidate scheme are calculated based on the structured data, the real-time hydro-meteorological data, and the technical parameters in each target candidate scheme; The reward function value of each target candidate scheme is determined based on the actual cost, actual ecological benefit, and actual recovery period of each target candidate scheme, and the weights corresponding to the cost, ecological benefit, and recovery period in the preset reward function, respectively; An optimized scheme is selected from the at least one target candidate scheme based on the reward function value of each target candidate scheme.

7. The method of claim 1, wherein, The derivative data includes soil data, flood data, plant data, and watershed geographic data, the risk assessment data includes a risk heat map, and the long-term effect prediction data and risk assessment data corresponding to the optimized scheme are generated based on the optimized scheme and the derivative data processed from the original multi-modal data, including: organic matter spatial variation prediction data is determined based on the optimized scheme, the soil data, and the watershed geographic data by using a first model; determine water and soil loss amount spatial variation prediction data based on the optimization scheme, the flood data, and the watershed geographic data by using a second model; determine vegetation coverage spatial variation prediction data based on the optimization scheme, the plant data, and the watershed geographic data by using a third model; generate the risk heat map based on the organic matter spatial variation prediction data, the water and soil loss amount spatial variation prediction data, the vegetation coverage spatial variation prediction data, and a preset risk level evaluation rule; wherein the risk heat map indicates the cyanobacteria bloom or the slope collapse probability of the monitoring area.

Images