Intelligent recommendation method, system and equipment for mountain photovoltaic ecological restoration scheme and medium

Abstract

The invention provides an intelligent recommendation method, system and device for a mountain photovoltaic ecological restoration scheme and a medium, and the method comprises the steps: obtaining geographic data of a target region, the geographic data comprising remote sensing data and a digital elevation model; target identification based on deep learning is carried out based on the geographic data to obtain photovoltaic panel geographic data; performing spatial intelligent deduction and cloud pixel filling on the remote sensing data through a visual large model to obtain reconstructed time sequence reflectivity data, and calculating a normalized vegetation index based on the reconstructed time sequence reflectivity data; performing ecological influence evaluation on the target area based on the photovoltaic panel geographic data and the normalized vegetation index to obtain ecological influence data; and performing ecological restoration scheme matching based on multi-rule constraint according to the ecological influence data to obtain an ecological restoration recommendation scheme. By adopting the method, intelligent conversion from a mountain photovoltaic ecological evaluation result to an executable scheme can be realized.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic ecological restoration technology, and in particular relates to a method, system, equipment and medium for intelligent recommendation of mountain photovoltaic ecological restoration solutions. Background Technology

[0002] With the development of remote sensing technology in the field of ecological monitoring, the photovoltaic identification and vegetation monitoring methods based on satellite imagery have been widely used in ecological restoration technology. They have the characteristics of acquiring surface information over a large area and periodically, which has led to the current mountain photovoltaic ecological management method that mainly relies on a combination of traditional remote sensing monitoring and human experience judgment.

[0003] In current industry practice, ordinary optical satellite imagery is typically used to extract the boundaries of photovoltaic panels, and time-series remote sensing data is used to monitor vegetation growth. This is then combined with expert experience to assess the ecological impact and select restoration solutions. However, the current approach has significant limitations: due to the constraints of complex terrain and climate conditions, the accuracy of photovoltaic panel identification and the continuity of vegetation time-series monitoring are insufficient. Furthermore, there is a lack of scientific quantitative support that is deeply related to the geographical environment between ecological assessment and restoration decisions, resulting in low overall management efficiency and weak decision reproducibility. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, system, equipment, and medium for intelligent recommendation of mountain photovoltaic ecological restoration solutions to address the above-mentioned technical problems. This method can overcome the bottleneck of the disconnect between assessment and restoration measures and realize the intelligent transformation of assessment results into executable solutions.

[0005] Firstly, this application provides an intelligent recommendation method for mountain photovoltaic ecological restoration solutions, including:

[0006] Acquire geographic data for the target area, including remote sensing data and digital elevation models;

[0007] Target recognition based on deep learning is performed on geographic data to obtain geographic data of photovoltaic panels, which includes mountain photovoltaic patches.

[0008] By using a large visual model to perform spatial intelligent extrapolation and cloud pixel filling on remote sensing data, reconstructed temporal reflectance data is obtained, and the normalized vegetation index is calculated based on the reconstructed temporal reflectance data.

[0009] An ecological impact assessment of the target area was conducted based on geographical data of photovoltaic panels and normalized vegetation index, and ecological impact data was obtained.

[0010] Based on ecological impact data, ecological restoration schemes are matched using a multi-rule constraint model to obtain recommended ecological restoration schemes. These recommended schemes indicate the appropriate restoration techniques for each mountain photovoltaic patch.

[0011] In one embodiment, target recognition based on deep learning is performed on geographic data to obtain geographic data of photovoltaic panels, including:

[0012] Based on the solar azimuth and elevation angles at the time of imaging using digital elevation models and remote sensing data, a terrain shadow mask map is generated by simulating the terrain shadow distribution of the target area through an illumination model.

[0013] Spatial registration is performed between remote sensing data and topographic shadow mask, and adaptive spectral enhancement is performed on pixels in the remote sensing data that fall in the shadow area to obtain a topographically calibrated image. The adaptive spectral enhancement is used to reduce the influence of topographic shadow on the spectral characteristics of ground features.

[0014] The terrain-calibrated images are input into the improved U-Net deep learning model, which introduces an attention mechanism in the encoder part to focus on the features of photovoltaic panels in mountainous environments;

[0015] By improving the U-Net deep learning model, pixel-level semantic segmentation is performed on the terrain calibration image, outputting a binarized photovoltaic panel distribution map. The binarized photovoltaic panel distribution map is then vectorized to obtain a vector format of mountain photovoltaic patches.

[0016] In one embodiment, spatial intelligent extrapolation and cloud pixel filling are performed on remote sensing data using a large visual model to obtain reconstructed temporal reflectance data, including:

[0017] The system acquires a multi-temporal satellite image sequence of the target area within a preset time span, and performs cloud and cloud shadow pixel recognition on each image in the multi-temporal satellite image sequence to generate a corresponding binary cloud mask. In the binary cloud mask, the first value represents that the corresponding pixel position is covered by clouds or cloud shadows, and the second value represents that the corresponding pixel position is clear and available.

[0018] Multi-temporal satellite image sequences and binary cloud masks are input into a pre-trained large-scale visual model. The encoder of the large-scale visual model extracts the spatiotemporal joint features of the multi-temporal satellite image sequences. The encoder captures the long-range dependencies between pixels at any spatiotemporal location through a self-attention mechanism to extract spatiotemporal joint features.

[0019] Based on spatiotemporal joint features, the conditional probability inference of all missing pixels identified by the first numerical value in the multi-temporal satellite image sequence is performed by the decoder of the visual big model, and the interpolated reflectance value of the missing pixels is reconstructed pixel by pixel to generate interpolated pixels.

[0020] The pixels identified by the second numerical value and the interpolated pixels are merged based on the binary cloud mask to obtain the reconstructed temporal reflectance data.

[0021] In one embodiment, an ecological impact assessment of the target area is conducted based on geographical data of photovoltaic panels and normalized difference vegetation index (NDVI) to obtain ecological impact data, including:

[0022] Acquire auxiliary geographic environmental data for the target area, including spatial distribution data of soil types, slope and aspect data, and meteorological time series data.

[0023] The geographical data of photovoltaic panels, normalized vegetation index and auxiliary geographical environmental parameters are input into the ecological service function assessment model to simulate and calculate the ecological service function of the target area under the scenarios before and after photovoltaic construction, respectively, and generate ecological spatial data before and after construction.

[0024] Spatial overlay analysis and pixel-level difference calculation were performed on the pre-construction ecological service function spatial data and the post-construction ecological spatial data to obtain ecological impact data.

[0025] In one embodiment, ecological restoration schemes are matched based on ecological impact data using a multi-rule constraint framework to obtain recommended ecological restoration schemes, including:

[0026] Construct a knowledge rule base for restoration measures, in which each restoration technical measure rule is associated with a set of geographical constraints and applicable conditions. The geographical constraints include slope range, soil thickness threshold, and ecological impact level.

[0027] Spatial overlay analysis is performed on ecological impact data and slope and soil thickness data obtained based on digital elevation models to generate a vector layer of ecological restoration assessment units with multiple attribute fields. The attribute fields of each ecological restoration assessment unit include spatial range, average slope value, average soil thickness value, and ecological impact level.

[0028] The attribute fields of each ecological restoration assessment unit are traversed and matched with the rules in the restoration measures knowledge rule base, and all technically feasible combinations of restoration technologies are selected through the rule engine.

[0029] Based on the preset economic optimization objectives, the combinations of restoration technologies are ranked, the optimal combination of restoration technologies is determined for each ecological restoration assessment unit, and a spatialized ecological restoration recommendation scheme is generated.

[0030] Secondly, this application also provides an intelligent recommendation system for mountain photovoltaic ecological restoration schemes, used to implement the method provided in the first aspect, the system comprising:

[0031] The data acquisition module is used to acquire geographic data of the target area, including remote sensing data and digital elevation models.

[0032] The photovoltaic panel identification module is used to perform deep learning-based target identification based on geographic data to obtain photovoltaic panel geographic data, which includes mountain photovoltaic patches.

[0033] The cloud reconstruction module is used to perform spatial intelligent extrapolation and cloud pixel filling on remote sensing data through a large visual model to obtain reconstructed temporal reflectance data, and calculate the normalized vegetation index based on the reconstructed temporal reflectance data.

[0034] The ecological assessment module is used to conduct ecological impact assessments of target areas based on geographical data of photovoltaic panels and normalized vegetation index, and to obtain ecological impact data.

[0035] The restoration plan recommendation module is used to match ecological restoration plans based on multiple rule constraints according to ecological impact data, and obtain ecological restoration recommendation plans. The ecological restoration recommendation plans are used to indicate the restoration technology measures that should be adopted for each mountain photovoltaic patch.

[0036] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-mentioned intelligent recommendation method for mountain photovoltaic ecological restoration schemes.

[0037] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-described intelligent recommendation method for mountain photovoltaic ecological restoration schemes.

[0038] The aforementioned intelligent recommendation method, system, equipment, and medium for mountain photovoltaic ecological restoration schemes acquires geographic data of the target area, including remote sensing data and digital elevation models. Based on this geographic data, deep learning is used for target identification to obtain geographic data of photovoltaic panels, including mountain photovoltaic patches. A large visual model is used to perform spatial intelligent extrapolation and cloud pixel filling on the remote sensing data to obtain reconstructed temporal reflectance data, and the normalized vegetation index is calculated accordingly. Combining the photovoltaic panel geographic data and the normalized vegetation index, an ecological impact assessment is conducted on the target area to obtain ecological impact data. Based on a multi-constraint rule base, the assessment results are intelligently matched with specific geographic environmental conditions to automatically generate ecological restoration technical schemes for each mountain photovoltaic patch. This technical scheme can improve the accuracy of photovoltaic identification, address the problems of discontinuous vegetation monitoring, and the disconnect between ecological assessment and restoration schemes, thereby enhancing the intelligent level of ecological management of mountain photovoltaic projects and providing full-chain technical support for the ecological management of mountain photovoltaic projects, from data acquisition to scheme output. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying 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.

[0040] Figure 1 A flowchart illustrating an intelligent recommendation method for mountain photovoltaic ecological restoration solutions provided in an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram of the structure of an intelligent recommendation system for mountain photovoltaic ecological restoration solutions provided in an embodiment of the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0043] First, a brief introduction to the terms used in the embodiments of this application will be given.

[0044] A Digital Elevation Model (DEM) is a digital model that reflects the undulating shape of the ground, constructed in a computer using limited terrain elevation data. It stores the elevation information of each location within a target area in the form of a regular grid or an irregular triangular mesh. As the core data foundation for terrain analysis, it provides accurate spatial references for terrain shadow simulation, slope and aspect calculation, and terrain undulation feature extraction, serving as crucial data support for solving related technical problems in complex mountainous terrain.

[0045] A geographic feature (GIN) is a discrete representation of the spatial extent of a specific geographic element in a geographic information system (GIS). It refers to a continuous spatial area with the same attribute characteristics (such as land cover type, ecological impact level, and / or restoration scheme type), typically presented as a closed polygon. In mountainous photovoltaic ecological restoration scenarios, GINs can be divided according to different application needs. For example, they can be divided into "mountainous photovoltaic GINs" based on the actual distribution range of photovoltaic panels, or into "severely affected area GINs" and "significantly affected area GINs" based on ecological impact assessment results. Each GIN not only contains clear spatial boundary information but also is associated with corresponding attribute data (such as slope, soil type, and ecological impact level). It is the core spatial unit for achieving spatialized management, accurate assessment, and targeted restoration scheme recommendations. It enables abstract data analysis results to correspond one-to-one with specific geographic areas, providing an intuitive and operable spatial carrier for refined ecological management.

[0046] Ecosystem service function assessment models are quantitative tools built upon ecological principles and mathematical algorithms. By integrating multi-source data such as topography, soil, meteorology, and vegetation, they simulate the service provisioning capacity of ecosystems under specific scenarios and output assessment results in the form of spatial distribution maps or quantitative values. Among these, the InVEST model (Integrated Valuation of Ecosystem Services and Ecosystem Transcription) is an example. Trade-offs is an ecosystem service function assessment tool based on ecological principles and spatial data analysis. Developed jointly by Stanford University and other institutions, it integrates multi-source spatial and temporal data, including topography, soil, vegetation, and meteorology, to quantitatively simulate and assess the core service functions provided by ecosystems. The model includes multiple functional modules such as habitat quality, soil conservation, water supply, and carbon storage. By setting different scenarios (such as before and after project construction), it can calculate and output the spatial distribution and quantitative results of various ecosystem service functions, intuitively presenting the degree of impact of human activities (such as mountain photovoltaic construction) on ecosystems. Its core value lies in transforming abstract ecological processes into quantifiable and visualized spatial data, providing scientific and objective quantitative basis for ecological impact assessment, ecological protection planning, and restoration scheme formulation. It is a key technical tool connecting ecological monitoring data and ecological management decisions.

[0047] Based on the above definitions, the implementation environment of the intelligent recommendation method for mountain photovoltaic ecological restoration schemes provided in this application embodiment will be described. Indicatively, the implementation environment includes: an interactive terminal, a processor, and a memory. The interactive terminal can be a desktop computer, laptop computer, tablet computer, or dedicated touch-screen terminal, etc., used to receive user operation commands, input parameter settings, and display the results of the ecological restoration recommendation scheme; the processor includes, but is not limited to, a central processing unit, a multi-core processor, or an artificial intelligence chip; the memory can be a distributed storage system or a centralized storage system, without limitation here.

[0048] Based on the above definitions and implementation environment, the application scenarios of the embodiments of this application are described. The intelligent recommendation method for mountain photovoltaic ecological restoration solutions provided in the embodiments of this application can be applied to scenarios including but not limited to the following:

[0049] Before the planning or construction of a mountain photovoltaic project, this technical solution can serve as a core tool for ecological pre-assessment and restoration plan development. For example, given the complex terrain and climate conditions of mountainous areas, by accurately identifying the topographic features and potential photovoltaic panel distribution range of the proposed construction area, and combining historical vegetation time series data and geographical environmental parameters, the potential impact of photovoltaic construction on regional ecosystem services (such as soil conservation and habitat quality) can be quantified in advance, and different levels of potential impact areas can be delineated. Based on a multi-constraint rule base, appropriate pre-remediation measures or ecological protection plans can be matched for areas with different site conditions to avoid irreversible ecological damage during construction, achieving forward-looking coordination between photovoltaic project construction and ecological protection, and reflecting the "prevention first" ecological management concept of the solution.

[0050] During the construction of mountain photovoltaic projects, this technical solution can be used for dynamic tracking of ecological conditions and real-time optimization of restoration measures. By regularly acquiring remote sensing and topographic data of the construction area, it continuously updates the actual distribution information of photovoltaic panels. A large visual model ensures the continuity of vegetation growth monitoring and promptly captures changes in ecological impacts caused by construction activities (such as localized vegetation degradation and increased soil erosion). Based on dynamically updated ecological impact data, combined with changes in construction progress and site geographical conditions, the restoration plan is intelligently adjusted to ensure a precise match between restoration measures and the dynamic process of ecological damage. This solves the problem that traditional restoration plans are difficult to adapt to fluctuations in ecological conditions during the construction period, improving the timeliness and targeted nature of ecological restoration.

[0051] After the mountain photovoltaic project is connected to the grid, this technical solution can support post-ecological impact assessment and long-term ecological restoration. By comparing quantitative data on ecological service functions before and after project construction, it can systematically assess the long-term impact of photovoltaic operation on the surrounding ecosystem and identify weak areas and key restoration targets. Based on long-term monitoring of vegetation time-series data, topographic change data, and ecological impact levels, it can recommend long-term sustainable restoration technology combinations for different photovoltaic patches and surrounding buffer zones using a multi-constraint rule base, including measures such as planting native plants, soil improvement, and / or slope reinforcement. At the same time, it can dynamically optimize the restoration plan based on continuous monitoring data of ecological restoration effects, realizing intelligent integrated plan planning from assessment to restoration, and promoting the long-term stable restoration of the ecosystem in the mountain photovoltaic area.

[0052] As an illustration, the intelligent recommendation method for mountain photovoltaic ecological restoration solutions provided in this application embodiment can also be applied to other application scenarios. This is only an example and does not limit the specific application scenarios.

[0053] In one exemplary embodiment, such as Figure 1As shown, a smart recommendation method for mountain photovoltaic ecological restoration schemes is provided. This embodiment illustrates the application of this method to an interactive terminal in the aforementioned implementation environment. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps 101 to 105:

[0054] Step 101: Obtain geographic data for the target area, including remote sensing data and digital elevation models.

[0055] Specifically, the geographic data mainly includes remote sensing data and digital elevation models (DEMs). Remote sensing data can be acquired through high-resolution satellite imagery receiving equipment, airborne remote sensing measurement platforms, or authoritative remote sensing data sharing platforms, such as GF-2 or WorldView series imagery, to provide detailed land cover information, encompassing multispectral band information to meet the needs of land cover identification and environmental monitoring. Digital elevation model (DEM) data can be acquired through lidar measurement technology, topographic mapping engineering, or from geospatial data service agencies. It contains core topographic information such as the altitude and topographic relief of the target area, used to accurately characterize the topographic relief features of the target area. Furthermore, before using the above geographic data, necessary data preprocessing can be performed, including but not limited to radiometric calibration and atmospheric correction of the remote sensing data, and hole filling and format standardization of the DEM data, to provide a high-quality, standardized input foundation for subsequent analysis.

[0056] Step 102: Perform deep learning-based target recognition based on geographic data to obtain photovoltaic panel geographic data, which includes mountain photovoltaic patches.

[0057] Specifically, to overcome the severe shadow interference caused by the complex mountainous terrain, this technical solution first generates a terrain shadow distribution map by simulating the illumination model based on the solar azimuth and elevation angles at the time of imaging, combined with digital elevation model data. Then, the original remote sensing image is spatially registered with this shadow map, and adaptive spectral enhancement processing is performed on pixels located in the shadow areas. For example, algorithms based on Retinex theory or histogram matching methods are used to recover the true spectral features of the ground features suppressed by shadows, thus obtaining a terrain-calibrated image. This terrain-calibrated image is then input into a pre-trained improved U-Net deep learning model for pixel-level classification. For example, this network can adopt an encoder-decoder architecture and integrate an attention mechanism (such as a CBAM module) into the encoder part of a U-Net variant. This network, trained on a large dataset of labeled images containing mountainous photovoltaic scenes, can effectively focus on and learn the unique spectral and spatial texture features of photovoltaic panels in mountainous environments. The network output is a binarized photovoltaic panel distribution raster map, which, after vectorization, yields mountainous photovoltaic patch vector data with precise spatial boundaries and geographic coordinates, i.e., photovoltaic panel geographic data.

[0058] Step 103: Spatial intelligent extrapolation and cloud pixel filling are performed on remote sensing data through a large visual model to obtain reconstructed temporal reflectance data, and the normalized vegetation index is calculated based on the reconstructed temporal reflectance data.

[0059] This method uses spatial intelligent extrapolation results from a large visual model to fill in cloud pixels. The model combines the extrapolated spatial distribution patterns of ground features, the spectral characteristics of adjacent clear areas, and the temporal evolution trends in multi-temporal images to accurately reconstruct the reflectance values ​​of missing pixels covered by clouds. This generates filler pixels that are coordinated with the spectral characteristics of surrounding ground features and conform to spatial distribution logic. The filling process makes full use of global spatial information and the inherent correlation between ground features, avoiding problems such as boundary blurring and numerical distortion that are prone to occur in traditional interpolation methods, so as to obtain complete and continuous reconstructed temporal reflectance data.

[0060] Specifically, for a multi-year, multi-temporal sequence of medium-resolution satellite imagery of the target area before and after the construction of photovoltaic projects, such as Sentinel-2 data, a binary mask identifying clouds and cloud shadow pixels is generated for each image in the sequence using its accompanying quality assessment band or a dedicated cloud detection algorithm (such as Sen2Cor, Fmask). Furthermore, the entire temporal image stack and its corresponding mask sequence are input into a pre-trained large-scale visual model, such as a large-scale model based on the VisionTransformer architecture and adaptively trained for remote sensing temporal analysis. The core of this model lies in its powerful spatiotemporal context understanding capability. Its encoder, through a multi-head self-attention mechanism, can capture the complex nonlinear correlation between clear and missing pixels across time and space dimensions, thereby learning the inherent laws and spatial patterns of the evolution of surface reflectance in the study area over time. The model's decoder, based on this learned global spatiotemporal joint feature, uses all known effective clear pixels as conditions to perform probabilistic inference on each pixel marked as missing by the mask, generating the most likely reasonable reflectance estimate. This completes the intelligent reconstruction of the entire temporal data interruption layer, outputting high-quality, cloud-free reconstructed temporal reflectance data. Based on this, the normalized vegetation index can be calculated pixel-by-pixel and temporally, resulting in a continuous and complete temporal monitoring dataset of vegetation growth status.

[0061] Step 104: Based on the geographical data of photovoltaic panels and the normalized vegetation index, conduct an ecological impact assessment of the target area to obtain ecological impact data.

[0062] Specifically, the aforementioned geographical data of photovoltaic panels (i.e., spatial distribution of disturbance sources), continuous normalized vegetation index time-series data (representing vegetation dynamics), are integrated together with the obtained spatial distribution map of soil types, slope and aspect maps derived from the digital elevation model, and meteorological time-series data such as precipitation and potential evapotranspiration. Further, these data are input into an ecosystem service function quantification model, such as the habitat quality module and soil conservation module of the InVEST model. By configuring model parameters, spatial distribution maps of key ecosystem service function indicators such as habitat degradation and soil erosion in the target area are simulated and output under two scenarios: before and after photovoltaic project construction. Furthermore, pixel-level difference calculations are performed on the spatial distribution maps of the same ecosystem service function before and after construction to obtain a spatial distribution map of the change in ecosystem service function. For example, the degree of change can be divided into several levels based on the statistical distribution characteristics of the change (such as the natural breakpoint method), such as "severe impact", "significant impact", "minor impact" and "no impact", thereby generating a spatially clear ecological impact level distribution map, which serves as quantitative and spatialized ecological impact data.

[0063] Step 105: Based on the ecological impact data, perform ecological restoration scheme matching based on multi-rule constraints to obtain ecological restoration recommended schemes. The ecological restoration recommended schemes are used to indicate the restoration technology measures that should be adopted for each mountain photovoltaic patch.

[0064] For example, a structured knowledge rule base for restoration measures is pre-constructed. This base integrates technical specifications, expert experience, and geographical adaptation requirements in the field of mountain ecological restoration. It encodes various ecological restoration techniques, such as hydroseeding, vegetation carpet laying, and planting combinations of native plants. Each measure is associated with a set of specific geographical applicability rules, such as "applicable to areas with a slope less than 30 degrees, soil thickness greater than 20 cm, and an ecological impact level of severe impact." The method spatially overlays an ecological impact level distribution map with a slope map calculated from a digital elevation model and a soil thickness distribution map obtained through surveys or model inversion, generating ecological restoration assessment units with unified multiple attributes (including spatial location, average slope, average soil thickness, and ecological impact level). Through a rule engine, this method can iterate and match the multiple attributes of each assessment unit with the rules in the knowledge rule base, logically determining whether the unit attributes satisfy all the constraints of each rule. For example, for a unit with the attributes of "slope of 28 degrees, soil thickness of 15 cm, and ecological impact level of severe impact", the rule engine will match all applicable rules and corresponding restoration technologies to form a set of feasible measures. On this basis, cost and benefit coefficients can be further introduced. Through multi-objective optimization algorithms or preset ranking rules (such as optimal cost-benefit ratio), a comprehensive optimal restoration technology or combination of measures will be recommended for each unit from the feasible set. Furthermore, the recommendation results of all units will be spatially summarized and visualized to create an ecological restoration recommendation scheme diagram, along with a list of corresponding restoration technologies.

[0065] Optionally, this technical solution can also drive a language big data model to perform logical reasoning based on ecological impact data and integrate expert database knowledge to match ecological restoration solutions. Specifically, this method first performs structured processing on the ecological impact data and the geographical site condition data of the target area. The ecological impact data originates from the quantitative analysis results of the ecological assessment stage mentioned above, including core information such as ecological impact level and impact type. The geographical site condition data covers slope data obtained based on digital elevation models, soil thickness data obtained from field surveys or geographic data platforms, and regional climate characteristic data. During the processing, these unstructured or semi-structured data need to be converted into textual description information that meets the input requirements of the language big data model, clarifying the attribute category and numerical range of each data point to ensure that the data can be accurately understood by the language big data model.

[0066] For example, this method inputs structured ecological impact data and geographical site condition data into a pre-trained language model. The language model can be a general-purpose model with strong logical reasoning capabilities or a specialized model fine-tuned using data from the ecological restoration field. For instance, a standardized input template can be constructed using prompt word engineering during input, ensuring that the core task of the language model is to reason about suitable ecological restoration techniques based on given data. Furthermore, this method invokes a pre-defined expert database, which integrates the experience of authoritative experts in mountain ecological restoration, typical engineering case solutions, technical specifications, and research results. This database is integrated with the reasoning process of the language model in the form of a structured knowledge base, providing domain-specific knowledge support and preventing the reasoning results from deviating from practical application scenarios.

[0067] For example, this method uses a large-scale language model for logical reasoning. Based on input ecological impact data and geographical site condition data, combined with professional knowledge from an expert database, the model analyzes the applicable scenarios and implementation effects of different restoration technologies and measures, and infers a preliminary candidate set of ecological restoration solutions. During the reasoning process, the model can optimize the reasoning results through a multi-turn dialogue mechanism. For complex geographical environments or areas with high-level ecological impact, the model further refines the data conditions and calls upon corresponding specialized solutions from the expert database to ensure that the restoration technologies and measures in the candidate set match the actual situation of the target area in terms of technical principles and implementation conditions.

[0068] Furthermore, this method can also screen and optimize the candidate set of restoration schemes output by the language big data model. The screening process can combine evaluation indicators such as technical feasibility and ecological compatibility to eliminate schemes that conflict with geographical site conditions or have poor restoration effects. In the optimization stage, a multi-objective decision-making algorithm can be introduced to comprehensively consider factors such as restoration cost, construction period, and long-term maintenance difficulty to prioritize candidate schemes. Finally, the optimal ecological restoration technology measures are determined for each mountain photovoltaic patch, and the restoration schemes of all patches are summarized to generate spatial ecological restoration recommendation schemes. The schemes clearly indicate the combination of restoration technologies and key points of implementation for each region. This embodiment, through the deep integration of the logical reasoning ability of the language big data model and the knowledge of the expert database, can overcome the limitations of traditional rule matching, improve the intelligence and professionalism of ecological restoration scheme matching, and better cope with the diverse ecological restoration needs in complex mountainous environments.

[0069] The aforementioned intelligent recommendation method for mountain photovoltaic ecological restoration schemes acquires geographic data of the target area, including remote sensing data and digital elevation models. Based on this geographic data, deep learning is used for target identification to obtain geographic data of photovoltaic panels, including mountain photovoltaic patches. A large visual model is used to perform spatial intelligent extrapolation and cloud pixel filling on the remote sensing data to obtain reconstructed temporal reflectance data, and the normalized vegetation index is calculated accordingly. Combining the photovoltaic panel geographic data and the normalized vegetation index, an ecological impact assessment is conducted on the target area to obtain ecological impact data. Based on a multi-constraint rule base, the assessment results are intelligently matched with specific geographic environmental conditions to automatically generate ecological restoration technical schemes for each mountain photovoltaic patch. This technical scheme can improve the accuracy of photovoltaic identification, address the problems of discontinuous vegetation monitoring, and the disconnect between ecological assessment and restoration schemes, thereby enhancing the intelligent level of ecological management of mountain photovoltaic projects and providing full-chain technical support for the ecological management of mountain photovoltaic projects, from data acquisition to scheme output.

[0070] In one embodiment, target recognition based on deep learning is performed on geographic data to obtain geographic data of photovoltaic panels, including:

[0071] Based on the solar azimuth and elevation angles at the time of imaging using digital elevation models and remote sensing data, a terrain shadow mask map is generated by simulating the terrain shadow distribution of the target area through an illumination model.

[0072] Specifically, the solar azimuth and elevation angles can be extracted from the metadata of remote sensing data, or calculated using astronomical algorithms combined with the imaging time of the remote sensing image and the latitude and longitude of the target area. The digital elevation model provides the elevation value for each pixel. Combined with the precise solar position parameters at the time of imaging, the shading of direct sunlight by the terrain can be calculated using ray tracing principles or simplified hemispherical field-of-view algorithms. For example, ESRI's Hillshade function or a radiative transfer-based model can be used to simulate which areas of the Earth's surface are in shadow under specific solar azimuth and elevation angle conditions. The terrain shadow mask generated in this step is a binary raster, where shadowed areas are assigned a specific identifier value, and unshadowed areas are assigned another identifier value, thus providing an accurate spatial reference for subsequent image correction.

[0073] Spatial registration is performed between remote sensing data and topographic shadow mask, and adaptive spectral enhancement is applied to pixels in the remote sensing data that fall in the shadow area to obtain a topographically calibrated image. The adaptive spectral enhancement is used to reduce the influence of topographic shadows on the spectral characteristics of ground features.

[0074] Specifically, spatial registration can employ a geometric correction method based on ground control points. By selecting corresponding control points in remote sensing data and topographic shadow mask maps, a spatial transformation relationship is established between the two, ensuring that the location of the shadow area accurately corresponds to that in the remote sensing data. Adaptive spectral enhancement can utilize the Retinex algorithm, multi-scale adaptive histogram equalization algorithm, or mapping enhancement algorithm based on the spectral differences between shadow and non-shadow areas. By analyzing the spectral response characteristics of pixels in the shadow area, brightness, contrast, and spectral curve parameters are dynamically adjusted to effectively reduce the masking effect of topographic shadows on the spectral characteristics of ground objects. This ensures that the spectral information of ground objects within the shadow area remains consistent with that of the non-shadow area. The resulting topographically calibrated image can accurately reproduce the spectral characteristics of photovoltaic panels, laying the foundation for subsequent accurate identification. Furthermore, adaptive spectral enhancement can also employ inversion methods based on physical models. For example, within the non-shaded areas of an image, sample pixels with the same or similar land cover types (such as vegetation or bare soil) to be corrected as those in the shaded areas are selected, and the reflectance statistical characteristics (such as mean and variance) of these sample pixels in each spectral band are analyzed. Based on a physical assumption that similar land features should have inherently consistent spectral reflectance characteristics in shaded and non-shaded areas, with the difference mainly stemming from the attenuation of incident light intensity, a correction model based on band ratio or linear transformation can be constructed. For instance, the reflectance ratio or linear regression relationship between sample pixels in the non-shaded area and the pixels to be corrected in the shaded area in a specific band can be calculated as a spectral adjustment coefficient. This coefficient is then applied to the pixels in the shaded area to scale or transform their original digital values, thereby restoring their approximate spectral response under sufficient illumination conditions. This adaptive method is based on class-by-class land features or local windows, which can effectively weaken the suppression and confusion of key spectral features of land features such as photovoltaic panels by terrain shadows, and improve the visibility and separability of land features in shaded areas.

[0075] The terrain-calibrated images are input into the improved U-Net deep learning model, which introduces an attention mechanism in the encoder part to focus on the features of photovoltaic panels in mountainous environments.

[0076] Specifically, this improved U-Net model integrates attention mechanisms into its encoder path, such as a convolutional block attention module, which sequentially includes a channel attention submodule and a spatial attention submodule. Specifically, the channel attention submodule learns the importance weights of each feature channel through global average pooling and fully connected layers, and recalibrates the feature responses in the channel dimension accordingly, making the network focus more on spectral or deep feature channels that significantly contribute to the identification of photovoltaic panels. The spatial attention submodule generates a two-dimensional spatial weight map by fusing channel information, multiplying it with the input feature map to highlight spatial regions in the image that may contain photovoltaic panels and suppress irrelevant background interference. The introduction of this attention mechanism enables the network to adaptively focus on the discriminative features of photovoltaic panels when dealing with complex mountainous scenes, such as their regular geometry, high reflectivity, and significant spectral contrast with the surrounding environment, thereby improving the model's ability to focus on targets and its feature representation capabilities.

[0077] By improving the U-Net deep learning model, pixel-level semantic segmentation is performed on the terrain calibration image, outputting a binarized photovoltaic panel distribution map. The binarized photovoltaic panel distribution map is then vectorized to obtain a vector format of mountain photovoltaic patches.

[0078] Specifically, the improved U-Net model, which has been fully trained, is used to perform pixel-level semantic segmentation on the terrain calibration image. The model takes the aforementioned terrain calibration image as input, and after multi-level feature extraction by the encoder and feature upsampling and fusion by the decoder, it generates a probability value for each input pixel in the output layer, classifying it as either a photovoltaic panel or the background. By applying a threshold (e.g., 0.5) to this probability output, a binary raster map of photovoltaic panel distribution is obtained, where pixels with a value of 1 represent those identified as photovoltaic panels. Further, to obtain vector data suitable for geospatial analysis and planning management, the binary raster map needs to be vectorized. For example, a connected component analysis algorithm can be used to identify all connected photovoltaic panel pixel regions. Then, contour tracing and polygon simplification algorithms, such as the Douglas-Puk algorithm, are used to convert the boundaries of each connected region into vector polygons. Simultaneously, an area threshold can be applied to filter out excessively small noise polygons, and the polygon boundaries can be smoothed to obtain high-precision, clear-boundary, and attribute-complete vector format mountain photovoltaic patch data. The output mountain photovoltaic location vector can be saved in Shapefile format.

[0079] In one embodiment, spatial intelligent extrapolation and cloud pixel filling are performed on remote sensing data using a large visual model to obtain reconstructed temporal reflectance data, including:

[0080] The system acquires a multi-temporal satellite image sequence of the target area within a preset time span, and performs cloud and cloud shadow pixel recognition on each image in the multi-temporal satellite image sequence to generate a corresponding binary cloud mask. In the binary cloud mask, the first value indicates that the corresponding pixel position is covered by clouds or cloud shadows, and the second value indicates that the corresponding pixel position is clear and available.

[0081] Specifically, the time span of multi-temporal satellite imagery needs to cover both before and after the construction of photovoltaic projects to meet the requirements for ecological change comparison. The imagery needs to include multispectral bands to support subsequent reflectance data extraction. Cloud and cloud shadow pixels are identified for each image in the multi-temporal satellite imagery sequence. The identification process can employ automatic identification algorithms based on image quality assessment bands, threshold segmentation algorithms, or machine learning classification algorithms. By analyzing the spectral characteristics, brightness values, and texture features of pixels, cloud and cloud shadow pixels are distinguished from clear pixels, thereby generating a corresponding binary cloud mask. The first value can be set to 0 to indicate that the corresponding pixel location is covered by clouds or cloud shadows, and the second value can be set to 1 to indicate that the corresponding pixel location is clear and usable. This binary cloud mask can accurately mark the spatial location of missing data in each image, providing a clear target for subsequent interpolation work. Furthermore, for each image, the identification of cloud and cloud shadow pixels can be achieved based on the cloud confidence information provided by the image's built-in quality assessment bands. By setting a threshold (such as a cloud confidence greater than 60%), pixels can be initially classified as suspected cloud areas. Further, spectral feature rules such as high reflectivity in the blue band and specific ratios between the near-infrared and short-wave infrared bands can be combined, or pixel brightness and texture features can be used to refine the identification of suspected cloud areas using machine learning classifiers such as support vector machines and random forests, in order to distinguish between thick clouds, thin clouds and their projected shadows.

[0082] Multi-temporal satellite image sequences and binary cloud masks are input into a pre-trained large-scale visual model. The encoder of the large-scale visual model extracts the spatiotemporal joint features of the multi-temporal satellite image sequences. The encoder captures the long-range dependencies between pixels at any spatiotemporal location through a self-attention mechanism to extract the spatiotemporal joint features.

[0083] Specifically, the large-scale visual model can be a pre-trained model based on the VisionTransformer architecture, the SwinTransformer architecture, or other models with strong spatiotemporal feature extraction capabilities. Before input, the image sequence needs to undergo radiometric calibration, atmospheric correction, and geometric fine correction, and the reflectance values ​​need to be normalized to a specific range (e.g., [0,1]) to eliminate interference from sensor differences and atmospheric effects, ensuring data consistency and comparability. For example, this method extracts the spatiotemporal joint features of multi-temporal satellite image sequences through the encoder part of the large-scale visual model. The encoder is composed of multiple layers of Transformer blocks stacked together, and its core is a self-attention mechanism. This mechanism can calculate the correlation weight between any two elements (i.e., spatiotemporal tokens) in the sequence, thereby capturing long-range spatiotemporal dependencies.

[0084] Furthermore, to adapt to temporal image data, the input image sequence is first reshaped and encoded into a spatiotemporal token sequence, for example, for a sequence of images generated by... A sequence of scene images, in which the first... The image can be represented as a three-dimensional tensor. ,in and These represent the height and width (number of pixels) of the image, respectively. Represents the number of spectral bands, the entire time series. It can be represented as a set of tensors at all time points: Mask sequence It can be represented as Each of them Is with A binary mask tensor with the same spatial dimension. During the encoding process, the model will use the image sequence... With mask information By combining these considerations and learning through a self-attention layer, the feature representation of each spatiotemporal location incorporates effective information from all other spatiotemporal locations (including different times and spaces). This results in a global feature representation that comprehensively characterizes the intrinsic evolutionary patterns and correlation structures of the entire image sequence in both temporal and spatial dimensions—that is, a spatiotemporal joint feature representation. .

[0085] Based on spatiotemporal joint features, the visual big data model decoder performs conditional probability inference on all missing pixels identified by the first numerical value in the multi-temporal satellite image sequence, and reconstructs the interpolated reflectance value of the missing pixels pixel by pixel to generate interpolated pixels.

[0086] Specifically, this method is based on the extracted spatiotemporal joint features. This method uses the decoder part of a large visual model to perform conditional probability inference on all missing pixels in the input image sequence that are marked by a first value (such as 0), in order to reconstruct their reflectance values ​​pixel by pixel. The decoder is usually composed of a multilayer perceptron or transposed convolutional structure, and its function is to reconstruct the reflectance values ​​based on the global spatiotemporal context features provided by the encoder. This process gradually restores or generates complete image data. For example, for any missing pixel in the sequence... (in For spatial location index, It is a band index, and satisfies its mask value. The goal of the model is to predict the most probable reflectance value of a missing pixel based on all known valid (clear) pixel information and learned spatiotemporal patterns. Mathematically, this process can be expressed as calculating the conditional expectation of the missing pixel. The following formula characterizes the model's inference process for a single missing pixel value:

[0087]

[0088] in, This represents the interpolated reflectance value of the missing pixel generated by the model prediction; This represents the conditional expectation operator; the condition part contains two terms: the first term is a set. It represents all points in time. Above, all spatial locations Location, mask value The first term is the effective pixel reflectance value, i.e., all known clear data; the second term is the aforementioned extracted spatiotemporal joint feature. It encodes the evolutionary pattern of the entire sequence. The model approximates the computation of this conditional expectation through its complex nonlinear transformation (i.e., the decoder function). In practice, the decoder utilizes features... And simultaneously consider the information provided by all valid pixels, for each missing location A new pixel value is synthesized. This value not only considers spatially adjacent effective pixels in the same scene image, but more importantly, it makes full use of the reflectance variation patterns of the same geographical location or similar land cover type areas in images from different time phases. This generates a more reasonable interpolation value in terms of spectral characteristics and spatiotemporal evolution logic. By traversing all time points, all spatial locations, and all spectral bands of pixels marked as missing, and repeating this reasoning process, a complete set of interpolated pixel values ​​corresponding to the original missing locations can be generated.

[0089] The pixels identified by the second numerical value and the interpolated pixels are merged based on the binary cloud mask to obtain the reconstructed temporal reflectance data.

[0090] Specifically, this method merges the effective sharp pixels in the original image identified by a second value (such as 1) with the interpolated pixels generated in the previous step to obtain reconstructed temporal reflectance data. The merging operation is performed pixel-by-pixel, temporally, and band-by-band: for any location Check its corresponding mask value ;like The original image will be preserved. The reflectance value; if Then use the interpolated values ​​generated by the model. The replacement is then performed. Through this "following the clues" approach, a new, continuous, and spatially complete reflectance dataset is generated, thus reconstructing the temporal reflectance data. This dataset successfully fills the information gaps caused by cloud cover, providing a reliable data foundation for subsequent calculations of ecological parameters such as the continuous normalized vegetation index.

[0091] In one embodiment, an ecological impact assessment of the target area is conducted based on geographical data of photovoltaic panels and normalized difference vegetation index (NDVI) to obtain ecological impact data, including:

[0092] Obtain auxiliary geographic environmental data for the target area, including spatial distribution data of soil type, slope and aspect data, and meteorological time series data.

[0093] Specifically, spatial distribution data of soil types can be obtained from regional soil survey results databases or geographic information public service platforms. This data is stored in vector or raster format, with each patch or cell encoding the corresponding soil type (e.g., clay, sand, loam) and its key attributes. This data forms the basis for assessing soil erosion sensitivity and habitat suitability. Slope and aspect data can be generated based on the digital elevation model using terrain analysis algorithms. For example, the slope tool in geographic information system software can be used to calculate the maximum rate of elevation change for each cell as the slope, in degrees. The aspect tool can be used to calculate the projection direction of the normal direction of each cell onto the horizontal plane as the aspect, in degrees (clockwise from due north). These topographic factors directly affect surface runoff, soil stability, and light conditions. Meteorological time-series data, especially precipitation and reference evapotranspiration data, can be obtained from long-term observation records or reanalysis meteorological products (e.g., ERA5) of the target area and surrounding meteorological stations. This data needs to be processed into sequences matching the assessment timescale (e.g., months, years) to provide dynamic climate drivers for eco-hydrological process simulation. Before being input into the model, all these auxiliary data need to undergo strict spatial registration, resolution resampling, and temporal alignment with the geographical data of photovoltaic panels and the normalized vegetation index data to ensure that all data layers are consistent in spatial and temporal dimensions.

[0094] The geographic data of photovoltaic panels, normalized vegetation index, and auxiliary geographic environmental parameters are input into the ecosystem service function assessment model to simulate and calculate the ecosystem service function of the target area under the scenarios before and after photovoltaic construction, respectively, and generate ecological spatial data before and after construction.

[0095] For example, this method inputs geographic data of photovoltaic panels, normalized vegetation index, and auxiliary geographic environment data into an ecosystem service function assessment model. The ecosystem service function assessment model can be the InVEST model, the ARIES model, or other mechanistic models with multiple ecosystem service quantification capabilities. Before input, all data must be processed to unify spatial coordinates and standardize data format to ensure data consistency in spatial scale and time dimension. This method sets up pre-construction and post-construction scenarios through the model. The pre-construction scenario uses geographic data, vegetation status, and environmental conditions before photovoltaic construction as input. The post-construction scenario incorporates the impact of photovoltaic panel distribution on the land surface. Through the built-in ecological process algorithm of the model, the ecosystem service function of the target area is simulated and calculated under the two scenarios, generating pre-construction and post-construction ecological spatial data. The ecosystem service function includes core indicators such as soil retention, habitat quality level, and / or vegetation carbon sequestration. The two types of spatial data are presented in raster or vector form, clearly representing the spatial distribution characteristics of ecosystem service functions under different scenarios.

[0096] In one possible embodiment, this technical solution uses the InVEST model as an evaluation tool. The model configuration requires preparing input parameters for both pre-construction and post-construction scenarios. Specifically, in the pre-construction scenario, the geographical data of photovoltaic panels is not included or is considered ordinary, non-threatening land cover. However, in the post-construction scenario, it needs to be converted into habitat threat source data. The conversion process includes assigning specific threat source intensity values ​​to the photovoltaic panel distribution patches (e.g., setting a value between 0 and 1 based on the intensity of their operation and maintenance activities) and setting their spatial attenuation distance (e.g., 500 meters), which characterizes the potential range and intensity of disturbance to the surrounding habitat by the photovoltaic facilities. Normalized Difference Vegetation Index (NDV) time-series data is used to extract parameters characterizing vegetation status. For example, by calculating the annual average NDC for the pre-construction baseline period (e.g., three years before project commencement) and the post-construction monitoring period (e.g., one year after grid connection), and using this as a key input for habitat suitability assessment, combined with soil type data, a spatial distribution map of habitat suitability reflecting the degree of suitability of vegetation cover and soil conditions for biological survival at different times is generated. The core calculations of the model are completed through its built-in mechanism module. For instance, the habitat quality module of the InVEST model calculates the habitat quality index. This module's calculations are based on parameters such as threat sources, habitat suitability, and the sensitivity of threat sources to the habitat. Specifically, the habitat quality index... (corresponding pixel) Land cover type The calculation of ) can be expressed by the following formula:

[0097]

[0098] in, Indicates land cover type Habitat suitability; It is a pixel Regarding land cover types The overall threat level is calculated by taking into account the intensity, distance attenuation, and sensitivity to different habitat types of all threat sources (including photovoltaic threat sources in the post-construction scenario); It is the half-saturation constant; This is a scaling constant; this formula quantifies the degree of habitat degradation caused by human activities. Simultaneously, the soil retention capacity is calculated using the soil conservation module. This module, based on the general soil loss equation, estimates the difference between actual and potential soil erosion using the following formula:

[0099]

[0100] Simplified, soil retention It can be expressed as ,in It refers to soil retention capacity; It is the rainfall erosivity factor, calculated from precipitation data; It is a soil erodibility factor, derived from soil data; It is a topographic factor, calculated based on slope and aspect data; It is a vegetation cover and management factor, whose value is negatively correlated with the normalized vegetation index. It can be obtained by converting the normalized vegetation index. The model is run twice, with the full set of parameters input for the pre-construction and post-construction scenarios respectively. Finally, it outputs the spatial distribution map of habitat quality index and soil retention in the pre-construction scenario, as well as the corresponding results in the post-construction scenario, which serve as the pre-construction ecological space data and the post-construction ecological space data.

[0101] Spatial overlay analysis and pixel-level difference calculation were performed on the pre-construction ecological service function spatial data and the post-construction ecological spatial data to obtain ecological impact data.

[0102] Specifically, this method performs spatial overlay analysis and pixel-level difference calculation on pre-construction and post-construction ecological spatial data to obtain ecological impact data. The spatial overlay analysis employs layer overlay technology in a geographic information system, accurately overlaying the two types of data within the same spatial reference frame to ensure one-to-one matching of ecological service function data at corresponding spatial locations. The pixel-level difference calculation calculates the difference between the post-construction and pre-construction ecological service functions pixel by pixel, clarifying the magnitude and direction of change in ecological service functions for each spatial unit. Areas with negative differences represent ecological service function degradation, while areas with positive differences represent ecological service function improvement. The difference calculation is performed pixel by pixel; for continuous ecological service indicators such as habitat quality or soil retention, it calculates the value of each pixel under the post-construction scenario. Numerical values ​​under the construction prospect scenario The difference between For example, it can be calculated using the following formula:

[0103]

[0104] in, Represents spatial location The change in the ecological service function of pixels. and These represent the ecological service functions (such as habitat quality index or soil conservation) of the pixel before and after construction, respectively. Negative values ​​indicate a decline in ecosystem service functions, while positive values ​​indicate an improvement. Furthermore, this method can be combined with statistical methods to classify the difference results into different levels of ecological impact, forming ecological impact data that includes the degree, scope, and type of impact, providing precise quantitative basis for matching subsequent restoration plans.

[0105] Furthermore, to transform continuous changes into more operationally relevant categorical data, this method can also perform hierarchical classification and geographic mapping of the change results. For example, natural breakpoints, equal-interval grading, or standard deviation-based grading methods can be used. For instance, the natural breakpoint method can be used to find inherent inflection points in the data as boundaries, dividing... The numerical range is divided into several consecutive intervals, each corresponding to an ecological impact level, such as "severely degraded area", "moderately degraded area", "slightly changed area", "basically stable area" and "improved area", generating an ecological impact level distribution map. This map clearly identifies the degree and type of ecological impact experienced by each pixel in the target area due to the construction of the photovoltaic project in a spatially explicit manner. This map and the quantitative change data on which it is based together constitute the ecological impact data output by this embodiment. This data fundamentally changes the limitations of traditional macro-qualitative assessment, realizes the objective quantification, precise positioning and hierarchical expression of ecological impact, and provides direct and reliable input for subsequent restoration decisions for specific spaces.

[0106] In one embodiment, ecological restoration schemes are matched based on ecological impact data using a multi-rule constraint framework to obtain recommended ecological restoration schemes, including:

[0107] A knowledge rule base for restoration measures is constructed, in which each restoration technical measure rule is associated with a set of geographical constraints and applicable conditions. The geographical constraints include slope range, soil thickness threshold, and ecological impact level.

[0108] Specifically, the construction of the knowledge rule base for restoration measures needs to integrate technical standards, scientific research results, and engineering practice experience in the field of mountain ecological restoration, covering various restoration techniques such as vegetation carpet laying, hydroseeding, soil conditioner application, planting of native plants, and gabion slope protection. Each restoration technique rule (e.g., "topsoil hydroseeding," "vegetation carpet laying," or "fish-scale pit + shrub planting") is clearly associated with a set of geographical constraints and applicable conditions. The geographical constraints include slope range, soil thickness threshold, and ecological impact level. The applicable conditions can be set in combination with regional climate characteristics, vegetation growth habits, and other factors. The rule base is stored in a structured data format, supporting the addition, modification, and deletion of rules to ensure good scalability and adaptability, and to provide matching basis for different geographical environments and ecological impact scenarios. For example, geographical constraints and applicable conditions can exist in the form of logical judgments. For example, a rule... The applicable conditions can be described as: slope Between and Between, and soil thickness Greater than the threshold And the ecological impact level Equal to a specific level This rule can be expressed as a logical expression: .in These are the slope, soil thickness, and ecological impact level attributes of the assessment unit, respectively. , , , For rules The preset threshold parameters and rule base are stored in a structured format such as database tables or XML, which supports flexible maintenance and expansion of rules and their parameters, ensuring that they can adapt to the unique site conditions and remediation goals of different regions.

[0109] Spatial overlay analysis is performed on ecological impact data and slope and soil thickness data obtained based on digital elevation models to generate a vector layer of ecological restoration assessment units with multiple attribute fields. The attribute fields of each ecological restoration assessment unit include spatial range, average slope value, average soil thickness value, and ecological impact level.

[0110] Specifically, slope data is obtained by processing a digital elevation model using a slope calculation algorithm. Soil thickness data can be acquired through field sampling and testing, integration of regional soil survey results, or geographic data platforms. Spatial overlay analysis is implemented based on the spatial analysis function of a geographic information system. By unifying the spatial reference system of all data, the impact level represented by the ecological impact data, the topographic slope represented by the slope data, and the soil conditions represented by the soil thickness data are spatially overlaid. For each basic spatial unit formed after overlay (usually the original raster cell, or a homogeneous polygon formed by merging adjacent similar cells), multiple attributes are extracted and calculated. For example, each assessment unit can be calculated using the following formula. average slope value Compared with average soil thickness value :

[0111]

[0112]

[0113] in, It constitutes the evaluation unit The total number of original pixels, and They are the first The slope and soil thickness values ​​of each cell are calculated. Simultaneously, the dominant ecological impact level within the cell (determined through mode statistics) is assigned to the cell as its ecological impact level attribute. Ultimately, each evaluation unit They all possess a spatial geometric information and its attributes. , , The records show that all units together form an ecological restoration assessment unit vector layer. In the ecological restoration assessment unit vector layer, each assessment unit contains attribute fields such as a clear spatial range, average slope value, average soil thickness value, and ecological impact level, realizing the spatial correlation between ecological impact and geographical site conditions.

[0114] The attribute fields of each ecological restoration assessment unit are iterated and matched with the rules in the restoration measures knowledge rule base, and all technically feasible combinations of restoration technologies are selected through the rule engine.

[0115] Specifically, this method iterates and matches the attribute fields of each ecological restoration assessment unit with the rules in the restoration measures knowledge rule base, and uses a rule engine to filter out all technically feasible combinations of restoration measures. The rule engine employs a forward reasoning mechanism, using the attribute fields of the assessment unit as input conditions, comparing each rule in the rule base one by one to determine whether the geographical constraints of the assessment unit meet the applicability requirements of the rule. If they do, the corresponding restoration measures are included in the candidate combinations. For example, for a given assessment unit… Its attribute set is The rule engine iterates through every rule in the knowledge base. This checks whether the cell attributes satisfy all the constraints of the rule. For example, a matching function can be defined. Its return value is a Boolean value (true or false). The decision logic can be formalized as: check , as well as Do all three sub-conditions hold true? If all three hold true, then... If true, it means the rule is true. Applicable to this unit, corresponding repair techniques The set of feasible measures included in this unit After traversing all rules, the unit can be obtained. All technically feasible remedial measures group Furthermore, a conflict resolution mechanism can be set up during the matching process. When multiple rules are matched simultaneously, they are filtered according to the priority of technical adaptation to ensure the technical feasibility and rationality of candidate combinations, providing a basis for subsequent optimization and selection.

[0116] Based on the preset economic optimization objectives, the combinations of restoration technologies are ranked, the optimal combination of restoration technologies is determined for each ecological restoration assessment unit, and a spatialized ecological restoration recommendation scheme is generated.

[0117] Specifically, the economic optimization objectives can be set as lowest restoration cost, highest input-output ratio, and lowest long-term maintenance cost. The ranking process can incorporate multi-objective decision-making methods such as the analytic hierarchy process (AHP) and entropy weight method, comprehensively considering factors such as the construction difficulty, material cost, maintenance cycle, and ecological restoration effect of restoration technologies. Candidate restoration technology combinations are quantitatively scored and ranked, and the combination with the highest score is selected as the optimal restoration technology combination for that evaluation unit. The optimal combinations of all evaluation units are then associated with their corresponding spatial ranges to generate spatialized ecological restoration recommendation schemes. These schemes are presented in vector graphics, clearly marking the recommended restoration technologies for each area, along with a list of technologies and implementation guidelines. Furthermore, each restoration technology in the knowledge base can be... Predefined implementation cost per unit area and long-term ecological benefits Among them, cost Covers direct costs such as materials, labor, and machinery; benefits Quantification can be based on expert scoring or assessments of recovery effectiveness in literature (e.g., setting a score of 1-10), for an assessment unit. and its set of feasible measures It is necessary to calculate each measure in the set. Overall score For example, it can be calculated using the following formula:

[0118]

[0119] in, It is a measure Implementation cost per unit area; It is a measure The long-term ecological benefit coefficient; and Describing the feasible sets respectively The maximum cost and maximum benefit of all measures in the process, this operation aims to normalize and eliminate the influence of dimensions; It is a cost weighting factor ( Its value can be set according to project budget constraints or decision-making preferences; a larger value indicates a greater emphasis on cost control, while a smaller value indicates a greater emphasis on ecological benefits. Based on the calculated... For all Sort in descending order and select the measure with the highest score. The optimal restoration technology measures selected for the assessment unit are matched and optimized through the matching and optimization process for all assessment units. The optimal measures for each unit are then associated with its spatial geometry to generate a spatialized ecological restoration recommendation map. A list of technical measures, including measure type, location, area, and estimated cost, can be directly exported.

[0120] The above embodiments, by constructing a formalized knowledge rule base and explicit mathematical matching logic, deeply integrate abstract ecological impact assessment with specific geographical site conditions, and introduce a quantitative multi-objective decision-making model, realizing automatic, transparent, and reproducible intelligent recommendations from complex environmental constraints to optimal remediation technical measures. This completely changes the traditional decision-making model that relies on the subjective experience of experts, and improves the scientific nature, pertinence, and implementation efficiency of remediation plans.

[0121] In summary, the intelligent recommendation method for mountain photovoltaic ecological restoration provided in this application acquires geographic data of the target area, including remote sensing data and digital elevation models. Based on this, it utilizes data fusion and intelligent terrain calibration technologies, combined with an improved deep learning model incorporating an attention mechanism, to achieve high-precision automatic identification of mountain photovoltaic panels, obtaining geographic data of photovoltaic panels containing mountain photovoltaic patches. Simultaneously, it leverages the spatiotemporal reasoning capabilities of a large visual model to perform spatial intelligent extrapolation and cloud pixel filling on remote sensing data obscured by clouds, reconstructing complete temporal reflectance data and inverting the normalized vegetation index to achieve vegetation restoration. The process involves continuous monitoring of the growth status of the photovoltaic (PV) panels; then coupling PV panel geographic data, vegetation time-series data, and multi-source auxiliary geographic environmental data such as soil and meteorology; using ecosystem service function assessment models such as InVEST to quantitatively analyze the changes in ecosystem service functions of the target area before and after PV construction, forming spatialized ecological impact data; based on the constructed knowledge rule base for restoration measures, the ecological impact data is spatially matched with geographic site conditions such as slope and soil thickness under multiple constraints; through rule engine reasoning and economic optimization, the optimal combination of restoration technologies is intelligently recommended for each mountain PV patch, generating a spatialized ecological restoration recommendation scheme.

[0122] The above-mentioned technical solution constructs a complete technical closed loop from precise perception to intelligent decision-making. Specific advantages include: 1) Significantly enhancing the perception capabilities of mountain photovoltaic ecological management. Through data fusion and intelligent terrain calibration, it effectively overcomes shading, achieving high-precision automatic identification of photovoltaic panels. It also reconstructs missing remote sensing data using spatiotemporal reasoning based on a large visual model, enabling continuous monitoring of vegetation growth status; 2) Achieving a shift from macroscopic description to spatial quantitative assessment. By coupling multi-source spatiotemporal data with mechanistic models, it quantifies ecological impacts and accurately implements them at specific spatial locations; 3) Overcoming the bottleneck of disconnect between assessment and restoration measures. Through a restoration knowledge rule base and multi-constraint spatial matching, it achieves intelligent transformation of assessment results into executable solutions, automatically combining abstract assessment levels with specific geographical site conditions to recommend the optimal technical combination for each patch requiring restoration.

[0123] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0124] Based on the same inventive concept, this application also provides an intelligent recommendation system 10 for mountain photovoltaic ecological restoration schemes, which implements the intelligent recommendation method for mountain photovoltaic ecological restoration schemes described above. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the intelligent recommendation system 10 for mountain photovoltaic ecological restoration schemes provided below can be found in the limitations of the intelligent recommendation method for mountain photovoltaic ecological restoration schemes described above, and will not be repeated here.

[0125] In one exemplary embodiment, such as Figure 2 As shown, a smart recommendation system 10 for mountain photovoltaic ecological restoration schemes is provided to implement the smart recommendation method for mountain photovoltaic ecological restoration schemes as described above. The system includes:

[0126] Data acquisition module 11 is used to acquire geographic data of the target area, including remote sensing data and digital elevation model;

[0127] The photovoltaic panel identification module 12 is used to perform deep learning-based target identification based on geographic data to obtain photovoltaic panel geographic data, which includes mountain photovoltaic patches.

[0128] The cloud reconstruction module 13 is used to perform spatial intelligent extrapolation and cloud pixel filling on remote sensing data through a large visual model to obtain reconstructed temporal reflectance data, and calculate the normalized vegetation index based on the reconstructed temporal reflectance data.

[0129] Ecological assessment module 14 is used to conduct ecological impact assessment of the target area based on the geographical data of photovoltaic panels and the normalized vegetation index, and obtain ecological impact data.

[0130] The restoration scheme recommendation module 15 is used to match ecological restoration schemes based on multi-rule constraints according to ecological impact data to obtain ecological restoration recommendation schemes. The ecological restoration recommendation schemes are used to indicate the restoration technology measures that should be adopted for each mountain photovoltaic patch.

[0131] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the intelligent recommendation method for a mountain photovoltaic ecological restoration scheme as described above.

[0132] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0133] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0134] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for intelligently recommending mountain photovoltaic ecological restoration solutions, characterized in that, The method includes: Acquire geographic data of the target area, including remote sensing data and digital elevation models; Based on the geographic data, deep learning-based target recognition is performed to obtain photovoltaic panel geographic data, wherein the photovoltaic panel geographic data includes mountain photovoltaic patches; The remote sensing data is spatially intelligently extrapolated and cloud pixel filled using a large visual model to obtain reconstructed temporal reflectance data, and the normalized vegetation index is calculated based on the reconstructed temporal reflectance data. An ecological impact assessment of the target area is conducted based on the geographical data of the photovoltaic panels and the normalized vegetation index to obtain ecological impact data. Based on the ecological impact data, an ecological restoration scheme is matched using a multi-rule constraint to obtain a recommended ecological restoration scheme. The recommended ecological restoration scheme is used to indicate the restoration technology measures that should be adopted for each of the mountain photovoltaic patches.

2. The method according to claim 1, characterized in that, The process of performing deep learning-based target recognition based on the geographic data to obtain photovoltaic panel geographic data includes: Based on the digital elevation model and the solar azimuth and elevation angles at the time of remote sensing data imaging, the terrain shadow distribution of the target area is simulated by the illumination model to generate a terrain shadow mask map. Spatially register the remote sensing data with the terrain shadow mask, and perform adaptive spectral enhancement on the pixels in the remote sensing data that fall in the shadow area to obtain a terrain calibration image. The adaptive spectral enhancement is used to reduce the influence of terrain shadow on the spectral characteristics of ground features. The terrain calibration image is input into an improved U-Net deep learning model, which introduces an attention mechanism in the encoder part to focus on the features of photovoltaic panels in mountainous environments; The improved U-Net deep learning model is used to perform pixel-level semantic segmentation on the terrain calibration image, outputting a binarized photovoltaic panel distribution map. The binarized photovoltaic panel distribution map is then vectorized to obtain the mountain photovoltaic patch in vector format.

3. The method according to claim 1, characterized in that, The process of using a large visual model to perform spatial intelligent extrapolation and cloud pixel filling on the remote sensing data to obtain reconstructed temporal reflectance data includes: A multi-temporal satellite image sequence of the target area within a preset time span is acquired, and cloud and cloud shadow pixel identification is performed on each image in the multi-temporal satellite image sequence to generate a corresponding binary cloud mask; wherein, in the binary cloud mask, the first value represents that the corresponding pixel position is covered by cloud or cloud shadow, and the second value represents that the corresponding pixel position is clear and usable. The multi-temporal satellite image sequence and the binary cloud mask are input into a pre-trained large visual model. The encoder of the large visual model extracts the spatiotemporal joint features of the multi-temporal satellite image sequence. The encoder captures the long-range dependencies between pixels at any spatiotemporal location through a self-attention mechanism to extract the spatiotemporal joint features. Based on the spatiotemporal joint features, the decoder of the visual big model performs conditional probability inference on all missing pixels identified by the first numerical value in the multi-temporal satellite image sequence, and reconstructs the interpolated reflectance value of the missing pixels pixel by pixel to generate interpolated pixels. The pixels identified by the second numerical value and the interpolated pixels are merged based on the binary cloud mask to obtain the reconstructed temporal reflectance data.

4. The method according to claim 1, characterized in that, The ecological impact assessment of the target area based on the geographical data of the photovoltaic panels and the normalized vegetation index yields ecological impact data, including: Acquire auxiliary geographic environment data for the target area, including soil type spatial distribution data, slope and aspect data, and meteorological time series data; The geographical data of the photovoltaic panel, the normalized vegetation index, and the auxiliary geographical environment parameters are input into the ecosystem service function assessment model to simulate and calculate the ecosystem service function of the target area under the scenarios before and after photovoltaic construction, respectively, and generate ecological space data before construction and ecological space data after construction. Spatial overlay analysis and pixel-level difference calculation are performed on the pre-construction ecological service function spatial data and the post-construction ecological spatial data to obtain the ecological impact data.

5. The method according to claim 1, characterized in that, The step of matching ecological restoration schemes based on the ecological impact data using multiple rule constraints to obtain recommended ecological restoration schemes includes: Construct a knowledge rule base for remediation measures, in which each remediation technical measure rule is associated with a set of geographical constraints and applicable conditions, including slope range, soil thickness threshold, and ecological impact level; The ecological impact data and the slope data and soil thickness data obtained based on the digital elevation model are spatially overlaid to generate a vector layer of ecological restoration assessment units with multiple attribute fields; wherein, the attribute fields of each ecological restoration assessment unit include spatial range, average slope value, average soil thickness value and ecological impact level; The attribute fields of each ecological restoration assessment unit are traversed and matched with the rules in the restoration measures knowledge rule base, and all technically feasible combinations of restoration technologies are selected through the rule engine. The combinations of restoration technologies are ranked based on preset economic optimization objectives. The optimal combination of restoration technologies is determined for each ecological restoration assessment unit, and a spatialized ecological restoration recommendation scheme is generated.

6. A smart recommendation system for mountain photovoltaic ecological restoration schemes, used to implement the method as described in any one of claims 1 to 5, characterized in that, The system includes: The data acquisition module is used to acquire geographic data of the target area, including remote sensing data and digital elevation models; A photovoltaic panel identification module is used to perform deep learning-based target identification based on the geographic data to obtain photovoltaic panel geographic data, wherein the photovoltaic panel geographic data includes mountain photovoltaic patches; The cloud reconstruction module is used to perform spatial intelligent extrapolation and cloud pixel filling on the remote sensing data through a large visual model to obtain reconstructed temporal reflectance data, and calculate the normalized vegetation index based on the reconstructed temporal reflectance data. The ecological assessment module is used to conduct an ecological impact assessment of the target area based on the geographical data of the photovoltaic panel and the normalized vegetation index, and obtain ecological impact data. The restoration plan recommendation module is used to match ecological restoration plans based on multi-rule constraints according to the ecological impact data to obtain ecological restoration recommendation plans. The ecological restoration recommendation plans are used to indicate the restoration technology measures that should be adopted for each of the mountain photovoltaic patches.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 5.

Images