A method and system for precise grouping and self-evolution repair of desert grassland based on seed region division, grass variety trait database and multi-source perception

Abstract

The application discloses a kind of desert grassland precision group configuration and self-evolution repair method and system based on seed area division grass variety character database and multi-source perception, comprising the following steps: S1, construct the digital grass variety characteristic library covering native grass species cold tolerance, drought tolerance, salt tolerance and other physical and biological characteristics, extract multi-dimensional germplasm characteristic vector Vseed;At the same time, through the Internet of Things node and space remote sensing platform, the multi-source environmental data of the repair area are obtained, the environmental characteristic vector Venv is extracted, and the two-dimensional input space is constructed;Realize germplasm resource assetization and group configuration decision quantification, break through the community stress resistance bottleneck from the source;Crack the rigid problem of process means, greatly improve the survival rate of construction with the adaptive mechanism of "one grid one prescription";Create a closed-loop iteration architecture covering the micro-ecological gene level, and give the system the ability of cross-cycle "self-evolution".

Description

Technical Field

[0001] This invention relates to the field of ecological restoration and smart forestry and grassland technology, specifically to a method and system for the precise combination and self-evolutionary restoration of desert grassland based on a seed-regionalized grass variety trait database and multi-source sensing. Background Technology

[0002] The Three-North Shelterbelt Project is a landmark project in my country's ecological civilization construction. However, in arid, semi-arid, and high-altitude cold and windy sand areas, traditional ecological restoration technologies generally face technical bottlenecks such as low survival rates, monotonous community structures, and unsustainable ecological functions. Existing technologies often suffer from deficiencies such as a lack of scientific quantitative basis for grass species selection, rigid process parameters, and a lack of systematic self-evolution capabilities. Grass species selection often relies on manual experience or blindly introduces single cultivated grass species, ignoring the ecological niche complementarity of native grass species in specific habitats, resulting in poor stress resistance and easy degradation. Moreover, planting methods are often limited to fixed species ratios and standardized construction parameters, ignoring the differentiated stress effects of different climatic zones and micro-topographic units on the plant establishment process. Existing ecological restoration projects are often one-off constructions, lacking a systematic update and improvement mechanism to quantify the ecological feedback after restoration (such as changes in vegetation cover, changes in the multifunctionality and multi-service nature of the ecosystem, positive succession of soil microbial communities, and restoration and enhancement of core functions) and guide subsequent restoration work. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for precise combination and self-evolutionary restoration of desert grassland based on a seed-regionalized grass variety trait database and multi-source sensing, so as to solve the problems mentioned in the background art.

[0004] By adopting the above technical solutions, the assetization of germplasm resources and the quantification of combination decisions have been realized, breaking through the bottleneck of community stress resistance from the source; the problem of rigid process methods has been solved, and the adaptive mechanism of "one grid, one prescription" has greatly improved the survival rate of planting; and a closed-loop iterative architecture covering the microecological gene level has been created, giving the system the ability to "self-evolve" across cycles.

[0005] Compared with existing technologies, the beneficial effects of this invention are as follows: by constructing a digital grass variety trait database through "multi-source heterogeneous data fusion", physiological stress resistance thresholds (such as salt tolerance limit EC50) and ecological functions that are difficult to measure in the field are transformed into calculable high-dimensional feature vectors Vseed. Combined with the latent semantic mapping of deep neural networks and the principle of ecological niche complementarity ("leguminous nitrogen fixation + grass sand fixation + shrub windbreak"), the precise quantitative matching of "specific habitats" and "stress resistance functional groups" is achieved, which completely solves the problem of community degradation caused by traditional restoration that relies heavily on human experience and blindly introduces single species. Based on the environmentally dominant stress factors (such as wind erosion threshold θwind and salinity threshold θsaline), the optimal collaborative planting mode is automatically matched, and the macro-meteorological base (precipitation P, altitude ALT) and micro-physical characteristics (top soil pressure coefficient εvigor) are coupled into the mathematical correction model of soil cover thickness Tfinal and irrigation quota. Through the air-ground integrated intelligent hardware to execute "one grid, one prescription", the environmental filtration pressure such as drought dehydration and physical crusting is accurately resolved, which greatly improves the initial survival rate under extreme habitats. A comprehensive evaluation matrix encompassing vegetation, soil, and microecology was constructed. Metagenomics technology was creatively introduced to quantitatively analyze the core functional genes of the underlying CNP cycle, enabling precise calculation of the Ecosystem Multifunctionality Index (EMF). This index was then transformed into a reward signal Rt within a reinforcement learning framework, driving a gradient algorithm that automatically updates the adaptation weight matrix between habitat and grass species. This endows the decision-making model with "lifelong learning" capabilities, achieving a complete shift from "human experience-based pre-setting" to "data-driven evolution," providing intelligent assurance for the long-term maintenance of desert grassland restoration. Attached Figure Description

[0006] Figure 1 This is an overall flowchart of the present invention; Figure 2 This is a schematic diagram of the module structure and data flow of a seed-regional-based grass variety trait database and a multi-source sensing desert grassland precise combination and self-evolution repair system in this invention. Figure 3 This is a topology diagram of the germplasm-environment environment filtering and niche complementarity algorithm based on deep neural networks in this invention. Figure 4 This is a block diagram of the differential construction mode decision-making and dynamic parameter correction logic for spatiotemporal heterogeneity in this invention. Figure 5 This is a diagram of the closed-loop self-evolutionary iterative architecture based on ecosystem multifunctionality (EMF) and reinforcement learning (RL) in this invention. Detailed Implementation

[0007] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0008] Please see Figure 1 , Figure 3 , Figure 4 and Figure 5This invention provides a technical solution: a method for precise matching and self-evolutionary restoration of desert grassland based on a seed-regionalized grass variety trait database and multi-source sensing, characterized by the following steps: S1. Construct a digital grass variety feature library covering the physical and biological characteristics of native grass species, such as cold resistance, drought resistance, and salt tolerance, and extract multidimensional germplasm feature vectors (Vseed). Simultaneously, multi-source environmental data of the restoration area is acquired through IoT nodes and aerospace remote sensing platforms, and environmental feature vectors (Venv) are extracted to construct a two-dimensional input space. Features include conventional phenotypic and morphological data (such as thousand-grain weight, root-shoot ratio, and seed size): mainly extracted automatically by connecting to national or regional forestry and grassland germplasm resource banks API and historical survey data; core physiological stress resistance thresholds (such as seedling topsoil pressure coefficient εvigor, salt tolerance half-lethal threshold EC50, and critical germination temperature Tmin): pre-determined through standardized gradient stress experiments in controlled environments (such as artificial climate chambers, texture analyzers, and greenhouse hydroponic systems), and converted into dimensionless digital features for database storage and ecological function evaluation feature data (water use efficiency, specific leaf area, carbon and nitrogen fixation capacity): supplemented by integrating publicly published ecological literature data and field standard plot measurement results; complex biological attributes that are difficult to measure in real time at the engineering site are transformed into high-frequency structured data assets that the system can call at any time, thus providing computing power support for efficient and intelligent front-end configuration. By deploying surface IoT nodes (TDR sensors, EC sensors) and aerospace remote sensing nodes, the most decision-indicative topography, environmental physicochemical properties, and meteorological indicators of the restoration area are acquired in real time, forming a low-dimensional and efficient environmental feature vector Venv = [v1, v2, ..., vn]T; which consists of three parts: meteorological and topographical base (such as altitude, precipitation, extreme low temperature, and average wind speed / wind erosion potential), acquired through the regional GIS geographic information system meteorological station API interface; real-time soil conditions (such as shallow soil moisture content, electrical conductivity, pH value, and shallow soil temperature), acquired by TDR sensors and four-electrode EC sensors deployed on the surface; and surface remote sensing features (such as initial vegetation index, percentage of surface sand content, surface albedo and slope, aspect micro-topography index), acquired through UAV multispectral flight inspection and satellite remote sensing data interpretation.

[0009] S2. Using a deep neural network (DNN), the Venv and Vseed input latent semantic space mapping layers are matched and calculated. Based on the principles of environmental filtering and niche complementarity, the optimal hybrid weight vector W is output. The signal transmission of the hidden layer nodes of the deep neural network adopts an activation function. Weight updates and nonlinear feature extraction are performed; the feature vector Vseed includes the critical germination temperature, average seed size μsize, and seedling top pressure coefficient εvigor; Vseed and Vven are projected into the same latent semantic space using a deep neural network, and the cosine similarity or fitness score between the genetic traits and environmental characteristics of each grass species is calculated to eliminate inferior species with insufficient stress resistance. Neural network algorithms: Data enters from the input layer, undergoes layers of "linear transformation" and "non-linear activation," and finally outputs a feature vector (A[l]). In the formula, A[l-1] is the output of the previous layer (if it is the first layer, it is the original input data, such as precipitation, altitude, etc.), W[l] is the weight matrix, which is the parameter learned by the system during training and determines the importance of different indicators (for example, in high-altitude areas, the weight of the indicator "altitude" will be amplified), and b[l] is the bias term, used to shift the triggering critical point of the activation function. Cosine similarity: Wherein, Consine Similarity is the cosine similarity, Venw is the environmental feature vector, and Vseed is the grass variety resource feature vector; the algorithm is based on the community niche complementarity principle of “leguminous nitrogen fixation + grass sand fixation + shrub windbreak”, automatically calculates the synergistic effect coefficient between species, and outputs the multi-species combination scheme with the strongest stress resistance and the highest comprehensive score, as well as the optimal mixed sowing weight vector W.

[0010] S3: Based on a multi-condition decision tree algorithm, precipitation P, altitude ALT, wind erosion modulus θwind, and electrical conductivity EC are used as input conditions. It matches differentiated planting patterns to address spatiotemporal heterogeneity and uses a dynamic correction module for process parameters to calculate the final construction parameters, generating a gridded construction prescription map. The decision tree algorithm includes the following decision branches: when the judgment condition "strong wind erosion θwind and drought" is met, the output matching mode is mechanical sand barrier and hole-seeding water collection mode; when the judgment condition "altitude > 3000m and cold" is met, the output matching mode is full film coverage and furrow mode. The final construction parameters include the soil cover thickness Tfinal, which is calculated using a soil cover correction function based on altitude ALT, annual precipitation P, and seed size and top soil pressure coefficient. The dynamic formula was calculated and corrected to obtain the result; S4: Based on the prescription map, the integrated air-ground automated equipment performs operations and integrates the restored multidimensional ecological feedback data to carry out cross-cycle system closed-loop self-evolution; the specific steps of closed-loop self-evolution include: monitoring apparent vegetation, core physicochemical and micro-ecological metagenomic data; calculating the ecological multifunctionality index to extract the EMF score; calculating the Reward value Rt through the reinforcement learning reward mechanism; and dynamically updating the adaptation weight matrix Wseed corresponding to the optimal hybrid seeding weight vector using the policy gradient algorithm.

[0011] Please see Figure 2 This invention provides a technical solution: a seed-regionalized grass varietal trait database and a multi-source sensing system for precise matching and self-evolutionary restoration of desert grassland; comprising the following architecture: a perception layer: a multi-source environmental perception and data input layer including IoT soil sensors, meteorological station API nodes, and UAV aerospace remote sensing platforms, used to generate multi-dimensional feature vectors; by deploying surface IoT nodes (TDR sensors, EC sensors) and aerospace remote sensing nodes, the most decision-indicative topography, environmental physicochemical properties, and meteorological indicators of the restoration area are acquired in real time, forming a low-dimensional and efficient environmental feature vector Venv = [v1, v2, ..., vn]T; consisting of three parts: meteorological and topographical base (such as altitude, precipitation, extreme low temperature, and average wind speed / wind erosion potential), acquired through the regional GIS geographic information system meteorological station API interface; and real-time soil conditions (such as shallow soil moisture content, electrical conductivity, pH value, and shallow soil temperature), acquired by TDR sensors and four-electrode EC sensors deployed on the surface; In addition, surface remote sensing features (such as initial vegetation index, percentage of surface sand content, surface albedo, slope, and aspect micro-topography index) are obtained through UAV multispectral flight inspection and satellite remote sensing data interpretation. Decision-Making Hub: An AI decision-making and combination hub containing a grass variety trait database and a self-evolutionary reinforcement learning engine, used to receive multi-dimensional feature vectors and output gridded construction prescription maps; using multi-criteria decision analysis (MCDA) or fuzzy comprehensive evaluation methods, it calculates the fit between the environmental feature vector Venv and the pattern feature space M; the system identifies the dominant stress factors in Venv (such as wind erosion intensity, surface evaporation rate, extreme temperature) and executes matching logic such as the following: High wind erosion-extremely arid habitat mode: When the wind erosion modulus > θwind (critical threshold for wind erosion) and the annual precipitation < 150mm, the system automatically matches the "mechanical sand barrier + hole seeding for water collection + application of water-retaining agent + microbial fertilizer + shade net" mode. This mode focuses on physical wind protection and in-situ water enrichment. High-altitude, low-temperature habitat mode: When the altitude is > 3000m or the extreme temperature is < -20℃, the system is matched with the "full film mulching + furrow planting" mode. This mode focuses on raising the soil temperature and isolating frost damage through film mulching; Saline-alkali habitat model: When the soil electrical conductivity EC > θsaline (critical threshold for saline-alkali toxicity), the system is matched with the "ditching for salt drainage + salt barrier layer construction + salt-tolerant shrub and grass mixed sowing" model; Soil cover depth model and irrigation compensation model are used to solve the dynamic adaptive adjustment of process parameters. Dynamic correction model for seeding and covering depth (Tfinal): Determining the covering depth requires balancing between "preventing seed wind erosion / dehydration" and "reducing soil breaking resistance"; T0: The baseline sowing depth for grass seeds under these geological conditions; Es / P: Evapotranspiration ratio, reflecting the degree of drought. The higher the evapotranspiration ratio, the greater the depth is needed to utilize deeper water. α is the evaporation regulation coefficient. ALT - ALTbase: Altitude increment. High-altitude areas are often accompanied by insufficient ground temperature, so it is necessary to adjust the depth appropriately by using the β coefficient to obtain surface heat, or to adjust the depth to avoid the cold (depending on the specific season). (μsize, εvigor) is the soil-jacking correction function based on the seed particle size μsize and the soil-jacking pressure coefficient εvigor. It is defined as follows: The larger the μsize (seed size), the more energy it carries; the larger the εvigor (overhead pressure coefficient), the stronger its ability to penetrate the physical crust. Intelligent Irrigation Quota Calculation Model (Qirr) During the initial management phase after planting, the system calculates irrigation compensation in real time based on environmental sensing: ETc: Potential transpiration of plants under current meteorological conditions; Ks: Drought resistance response coefficient calculated based on Vseed characteristics (Ks is lower for grass species with strong drought resistance); Peff: Effective precipitation; Ssoil: Soil texture correction factor (sandy soils infiltrate quickly, requiring small amounts multiple times to increase the correction frequency). Execution Layer: This layer comprises an integrated air-ground intelligent execution layer containing unmanned variable seeders and intelligent irrigation and maintenance equipment. It receives prescription maps and executes differentiated construction and maintenance. The variable execution mechanism based on prescription maps enables precise air-ground integrated operations. The Ecological Multifunctionality (EMF) quantification model enables multi-dimensional data monitoring and full-element evaluation. The weight optimization based on the reward mechanism completes reinforcement learning and self-evolutionary iteration. The optimal grass species combination weight W, planting mode, and dynamic process parameters (such as soil cover thickness Tfinal and irrigation quota) output by the system decision are transformed into a gridded variable construction prescription map with spatial geographic coordinates (GIS). Execution logic: Link up with intelligent air-ground equipment (such as unmanned variable seeders and drone spraying matrices) equipped with Beidou / RTK high-precision navigation modules to perform precise "one grid, one prescription" operations in the restoration area with heterogeneous micro-topography, and realize real-time servo adjustment of sowing depth and mixed sowing ratio; At the macro-level of vegetation restoration, multispectral and hyperspectral image inversion technology from unmanned aerial vehicles (UAVs) is used to dynamically extract vegetation cover, aboveground / underground biomass, and community species diversity indices, quantitatively assessing the survival rate of newly established plants and the level of primary productivity recovery in the early stages of restoration. At the meso-level of soil habitat, in-situ IoT sensing and periodic sampling are combined to monitor the degree of improvement in core physicochemical properties such as soil saturated hydraulic conductivity, pH, electrical conductivity, and bulk density at high frequencies, and to measure the dynamic changes in soil organic carbon pool and available nitrogen and phosphorus to assess the nutrient cycling retention and transformation efficiency of the ecosystem. At the micro-level of ecology, metagenomics technology is introduced to deeply analyze the structural evolution of the soil micro-ecological network, and the abundance of core functional genes involved in the carbon-nitrogen-phosphorus (CNP) cycle (such as the nitrogen-fixing gene nifH) is quantitatively extracted, ultimately accurately characterizing the directed positive succession of soil microbial communities and the overall improvement of core metabolic functions mediated by the habitat. EMF Quantification Formula: After standardizing the acquired n ecological parameters, the system calculates the Ecosystem Multifunctionality (EMF) index of the restoration grid. Wherein, Fi,t is the measured value of the i-th ecological function indicator in the t-th period (such as vegetation cover, organic carbon, available phosphorus, microbial nitrogen, microbial community diversity, microbial functional gene abundance, etc.). and These are the baseline mean and standard deviation of the indicator within the region, respectively. By integrating the full-element evaluation results into the reinforcement learning (RL) framework of the AI ​​decision-making center, self-evolution across cycles can be achieved. Reward calculation: The system compares the actual EMFt after repair with the target expected value EMFexp to calculate the environmental reward value Rt. Model parameter adaptive update: Based on the generated reward value Rt, the system uses the gradient ascent algorithm to automatically correct the "implicit adaptation weight matrix Wseed" of the corresponding grass species in the grass trait database under specific habitats. in, The historical weights before the update; The system's learning rate; This represents the policy gradient.

[0012] Co-evolution: This establishes a data-driven system update mechanism. If a certain planting pattern or grass species ratio receives continuous positive rewards (Rt > 0) in a specific extreme habitat, the system will automatically increase its recommendation priority in future similar habitat tasks, thereby achieving a complete transformation of the decision-making model from "human experience-based pre-set" to "data-driven evolution".

Claims

1. A method for precise matching and self-evolutionary restoration of desert grasslands based on a seed-regionalized grass varietal trait database and multi-source sensing, characterized in that: Includes the following steps: S1. Construct a digital grass variety feature library covering the physical and biological characteristics of native grass species, such as cold resistance, drought resistance, and salt resistance, and extract the multi-dimensional germplasm feature vector Vseed; at the same time, acquire multi-source environmental data of the restoration area through IoT nodes and aerospace remote sensing platform, extract the environmental feature vector Venv, and construct a two-dimensional input space. S2. Using a deep neural network (DNN), the Venv and Vseed input latent semantic space mapping layers are matched and calculated. Based on the principles of environmental filtering and niche complementarity, the optimal hybrid weight vector W is output. S3: Based on the multi-condition decision tree algorithm, precipitation P, altitude ALT, wind erosion modulus θwind and electrical conductivity EC are used as input conditions. Differentiated planting patterns are matched according to spatiotemporal heterogeneity. The final construction parameters are calculated using the process parameter dynamic correction module to generate a gridded construction prescription map. S4: Based on the prescription map, the air-ground integrated automated equipment performs operations and integrates the repaired multi-dimensional ecological feedback data to carry out cross-cycle system closed-loop self-evolution.

2. The method for precise combination and self-evolutionary restoration of desert grassland based on a seed-regionalized grass varietal trait database and multi-source sensing, as described in claim 1, is characterized in that: In step S1, the features include: conventional phenotypic and morphological feature data (such as thousand-grain weight, root-shoot ratio, and seed size), core physiological stress resistance thresholds (such as seedling top soil pressure coefficient εvigor, salt tolerance half-lethal threshold EC50, and critical germination temperature Tmin), and ecological function evaluation feature data (water use efficiency, specific leaf area, and carbon and nitrogen fixation capacity).

3. The method for precise matching and self-evolutionary restoration of desert grassland based on a seed-regionalized grass varietal trait database and multi-source sensing, as described in claim 1, is characterized in that: In step S2, the signal transmission of the hidden layer nodes in the deep neural network uses an activation function. Perform weight updates and nonlinear feature extraction; The feature vector Vseed includes the critical germination temperature, the average seed size μsize, and the seedling top pressure coefficient εvigor.

4. The method for precise combination and self-evolutionary restoration of desert grassland based on a seed-regionalized grass varietal trait database and multi-source sensing, as described in claim 1, is characterized in that: The decision tree algorithm in step S3 includes the following decision branches: when the judgment condition is "strong wind erosion and drought", the output matching mode is mechanical sand barrier and hole seeding water collection mode; when the judgment condition is "altitude > 3000m and cold", the output matching mode is full film coverage and furrow mode.

5. The method for precise combination and self-evolutionary restoration of desert grassland based on a seed-regionalized grass varietal trait database and multi-source sensing, as described in claim 1, is characterized in that: The final construction parameters in step S3 include the topsoil thickness Tfinal, which is determined by a topsoil correction function that includes altitude ALT, annual precipitation P, and topsoil pressure coefficient. The dynamic formula was calculated and corrected.

6. The method for precise combination and self-evolutionary restoration of desert grassland based on a seed-regionalized grass varietal trait database and multi-source sensing, as described in claim 1, is characterized in that: The specific steps of the closed-loop self-evolution in step S4 include: monitoring apparent vegetation, core physicochemical and microecological metagenomic data; calculating the ecological multifunctionality index to extract the EMF score; calculating the Reward value Rt through a reinforcement learning reward mechanism; and dynamically updating the adaptation weight matrix Wseed corresponding to the optimal hybrid seeding weight vector using a policy gradient algorithm.

7. A system for precise combination and self-evolutionary restoration of desert grassland based on a seed-regionalized grass phenotypic database and multi-source sensing, employing the method for precise combination and self-evolutionary restoration of desert grassland based on a seed-regionalized grass phenotypic database and multi-source sensing as described in claims 1-5, characterized in that: The architecture includes the following: Perception layer: This layer contains a multi-source environmental perception and data input layer, including IoT soil sensors, weather station API nodes, and UAV aerospace remote sensing platforms, used to generate multi-dimensional feature vectors; Decision-making center: An AI decision-making and combination center containing a grass variety trait database and a self-evolutionary reinforcement learning engine, used to receive multi-dimensional feature vectors and output gridded construction prescription maps; Execution layer: This is an integrated air-ground intelligent execution layer that includes unmanned variable seeders and intelligent irrigation and maintenance equipment. It is used to receive prescription maps and execute differentiated construction and maintenance.

Images