A system for evaluating the ecological restoration effect of an open-pit mine

Abstract

The application discloses an open-pit mine ecological restoration effect evaluation system and belongs to the field of ecological restoration evaluation. The system comprises a multi-source heterogeneous data fusion module, a space-time coupling evaluation module, a multi-scale hierarchical evaluation module and a restoration trajectory prediction module. The multi-source heterogeneous data fusion module performs space-time benchmark unification and pixel-level registration fusion on satellite multispectral remote sensing images, unmanned aerial vehicle hyperspectral images and data of a ground Internet of Things sensor network, and generates a multi-dimensional ecological parameter data set aligned in space-time. The space-time coupling evaluation module constructs an ecological element space correlation graph model based on a graph attention network, learns the space coupling weight of five elements of vegetation-soil-terrain-hydrology-microclimate through a multi-head graph attention mechanism, and outputs a space coupling evaluation value. The application realizes ecological element space coupling modeling, multi-scale hierarchical evaluation and restoration trajectory prediction, and the evaluation result is scientific and accurate, which can provide a scientific decision basis for mine ecological restoration.

Description

Technical Field

[0001] This invention relates to ecological restoration assessment, and more particularly to an assessment system for the effectiveness of ecological restoration in open-pit mines. Background Technology

[0002] Open-pit mining causes severe damage to the regional ecological environment, leading to multiple ecological problems such as topographic disturbance, soil degradation, vegetation destruction, hydrological imbalance, and biodiversity loss. Ecological restoration after mine closure is a crucial link in the sustainable development of mining areas, and scientifically assessing the effectiveness of ecological restoration is the core foundation for measuring the quality of restoration projects and guiding subsequent restoration strategies.

[0003] Currently, mine ecological restoration assessments mainly rely on the following methods: First, macro-monitoring assessment based on remote sensing images, using satellite remote sensing to extract indicators such as vegetation indices and land use types for evaluation. However, due to limitations in spatial resolution and temporal frequency, it is difficult to obtain deep ecological information such as soil and hydrology. Second, point-based assessment based on ground sampling and laboratory analysis, which has high accuracy but insufficient spatial representativeness and cannot reflect the overall condition of the restoration area. Third, traditional multi-indicator comprehensive evaluation methods such as the Analytic Hierarchy Process (AHP), which use expert weighting to assign weighted scores to multi-dimensional indicators. However, the indicators are considered to be independent of each other, ignoring the spatial coupling relationship between ecological elements such as vegetation, soil, topography, hydrology, and microclimate.

[0004] Chinese invention patent application CN2024110747081 discloses a method for evaluating the benefits of ecological protection and restoration in mining areas. This method employs multi-source data fusion and dynamic weight adjustment via machine learning, combined with fuzzy comprehensive evaluation and multi-objective decision analysis. However, this method has the following shortcomings: First, the evaluation model treats each indicator as an independent variable, failing to establish a spatial correlation model between ecological elements and thus unable to reflect the synergistic and antagonistic effects of ecological processes within the restoration area. Second, the assessment is conducted only at a single spatial scale, neglecting scale differences and result transmission between the pit-level, mining area-level, and regional levels. Third, it only evaluates the current restoration status, lacking the ability to predict future restoration trends and failing to provide forward-looking guidance for long-term restoration planning. Fourth, the assessment threshold is fixed and not dynamically adjusted according to the restoration progress and climate zoning, resulting in low scores in the early stages of restoration and a loss of discriminatory power.

[0005] In summary, existing technologies suffer from technical deficiencies such as lack of spatial coupling analysis of ecological elements, discontinuity in multi-scale assessment, gaps in restoration trend prediction, and rigid assessment thresholds. There is an urgent need for an open-pit mine ecological restoration effectiveness assessment system that can achieve deep fusion of multi-source data, spatial coupling modeling of ecological elements, multi-scale hierarchical assessment, and restoration trajectory prediction. Summary of the Invention

[0006] Purpose of the invention: The purpose of this invention is to overcome the shortcomings of existing technologies in ecological restoration assessment, such as ignoring spatial coupling of elements, lacking multi-scale assessment and trend prediction capabilities, and rigid assessment thresholds. It provides an open-pit mine ecological restoration effectiveness assessment system that integrates multi-source heterogeneous data, constructs a spatial correlation graph model of ecological elements, and realizes multi-scale hierarchical assessment and restoration trajectory prediction.

[0007] Technical solution: A system for evaluating the effectiveness of ecological restoration in open-pit mines, comprising:

[0008] The multi-source heterogeneous data fusion module is used to acquire satellite multispectral remote sensing images, UAV hyperspectral images and monitoring data from ground IoT sensor networks. It performs spatiotemporal benchmark unification and pixel-level registration fusion on the multi-source data to generate a spatiotemporally aligned multidimensional ecological parameter dataset. The multidimensional ecological parameter dataset includes spatiotemporally continuous data in five dimensions: vegetation cover, soil physicochemical indicators, topographic slope, hydrological parameters and microclimate parameters.

[0009] The spatiotemporal coupling assessment module constructs a spatial correlation graph model of ecological elements based on a graph attention network. It uses each sampling unit in the restoration area as a graph node and the spatial adjacency relationship between nodes as graph edges. The parameters of each dimension in the multidimensional ecological parameter dataset are used as node feature vector inputs. The module learns the spatial coupling weights between different ecological elements through a multi-head graph attention mechanism and outputs the spatial coupling assessment value of each sampling unit. The spatial coupling assessment value reflects the spatial coordination state of the five elements of vegetation, soil, topography, hydrology and microclimate.

[0010] The multi-scale hierarchical assessment module calculates the comprehensive ecological restoration score at each of the three spatial scales: pit unit, mining area unit, and regional unit. It also uses scale aggregation and decomposition methods to achieve upward aggregation and downward decomposition calibration of cross-scale assessment results.

[0011] The restoration trajectory prediction module, based on an attention-enhanced long short-term memory network, takes the temporal data of the spatiotemporally aligned multidimensional ecological parameter dataset as input, learns the temporal patterns of ecological restoration evolution, and predicts the changing trends of various ecological parameters and restoration effectiveness scores within a preset future time period.

[0012] Furthermore, the multi-source heterogeneous data fusion module includes a spatiotemporal reference unification submodule and a pixel-level registration and fusion submodule. The spatiotemporal reference unification submodule unifies satellite imagery, UAV imagery, and ground sensor data to the same spatial coordinate system and time reference. The pixel-level registration and fusion submodule uses a mutual information-based registration algorithm to resample data with different spatial resolutions to a unified grid, and fuses heterogeneous observations at the same spatiotemporal location using Kalman filtering to generate the multidimensional ecological parameter dataset with consistent spatial resolution.

[0013] Furthermore, in the spatiotemporal coupling evaluation module, the calculation process of the multi-head graph attention mechanism is as follows: for each graph node, calculate its attention coefficient with all neighboring nodes, and obtain the updated feature of the node by weighted summation of the attention coefficient and the feature vector of the neighboring nodes. After iteration through multiple graph attention layers, the final node feature is mapped to the spatial coupling evaluation value through a fully connected layer. The calculation of the attention coefficient introduces an ecological distance constraint term, so that the attention weight between nodes that are closer in spatial distance is higher.

[0014] Furthermore, in the multi-scale hierarchical evaluation module, the scale aggregation method is as follows: based on the comprehensive score of the mining pit unit, the area-weighted average is aggregated to obtain the mining area unit score, and then aggregated to obtain the regional unit score according to the ecological weight of the mining area unit; the decomposition and calibration method is as follows: the regional unit score is decomposed according to the ecological sensitivity coefficient of each mining pit unit to obtain the calibrated mining pit unit score, which is used to correct local evaluation bias.

[0015] Furthermore, in the repair trajectory prediction module, the attention-enhanced long short-term memory network includes an encoder and a decoder; the encoder is composed of multiple LSTM units, which encode the input time-series data to extract time-series features; the decoder adds a self-attention layer on the basis of LSTM, which performs weighted aggregation on the hidden state sequence output by the encoder to predict the changing trend of each ecological parameter and the repair effectiveness score within a preset time period in the future.

[0016] Furthermore, the system also includes a dynamic threshold calibration module. Based on the climate zone and mine type of the restored mine, the dynamic threshold calibration module matches the corresponding undisturbed ecosystem parameters from the reference ecosystem database as the evaluation benchmark, and dynamically adjusts the threshold range in combination with the restoration duration, so that the evaluation standard adapts to the restoration process.

[0017] Furthermore, the system also includes an evaluation result visualization and early warning module. The evaluation result visualization and early warning module outputs the comprehensive score, spatial coupling evaluation value and predicted restoration effect score at each scale in the form of heat map, time series curve and three-dimensional terrain overlay map. When the spatial coupling evaluation value of any sampling unit is lower than the preset early warning threshold or the predicted restoration effect score shows a downward trend, early warning information is generated and the location of abnormal area is marked.

[0018] Furthermore, the ground-based IoT sensor network includes soil moisture sensors, soil pH sensors, soil temperature sensors, surface runoff monitors, and micro weather stations; the spatial resolution of the satellite multispectral remote sensing image is not less than 2 meters, the spatial resolution of the UAV hyperspectral image is not less than 0.05 meters, and the data acquisition interval of the ground-based IoT sensor network does not exceed 1 hour.

[0019] Furthermore, in the aforementioned multidimensional ecological parameter dataset, vegetation cover is obtained by fusing the normalized vegetation index from satellite multispectral imagery with the chlorophyll inversion results from UAV hyperspectral imagery; soil physicochemical indicators are obtained by fusing real-time monitoring data from ground-based IoT sensors with the soil spectral inversion results from UAV hyperspectral imagery; topographic slope is calculated using a digital elevation model generated by UAV oblique photogrammetry; hydrological parameters are obtained by fusing surface runoff monitoring data with the hydrological analysis results of the digital elevation model; and microclimate parameters are obtained by fusing micro-weather station data with satellite remote sensing surface temperature products.

[0020] Beneficial effects:

[0021] (1) Spatial coupling modeling of ecological elements breaks through the limitation of traditional assessment that treats each indicator as an independent variable. This invention constructs a spatial correlation graph model of ecological elements based on graph attention network. Through the multi-head graph attention mechanism, it automatically learns the spatial coupling weights between the five elements of vegetation-soil-topography-hydrology-microclimate. It can reveal the synergistic and antagonistic effects of ecological processes within the restoration area. The assessment results are more in line with the real laws of multi-element interaction in the ecosystem.

[0022] (2) Multi-scale hierarchical assessment to achieve spatial scale connectivity from pit to region. This invention conducts hierarchical assessment according to three spatial scales: pit unit, mining area unit and regional unit. It also achieves the transmission and correction of cross-scale assessment results through scale aggregation and decomposition calibration methods, which solves the problems of insufficient spatial representativeness of single-scale assessment and disconnect between scale assessment results.

[0023] (3) Predicting restoration trajectory with forward-looking assessment capabilities. This invention is based on an attention-enhanced LSTM network to learn the temporal patterns of restoration evolution. It can predict the changing trends of ecological parameters and restoration effectiveness scores in the future, providing a scientific basis for long-term restoration planning and avoiding passive responses only when degradation occurs during the restoration process.

[0024] (4) Dynamic threshold calibration, the assessment standard is adaptively adjusted with the restoration process. The present invention matches the ecological system benchmark with the climate zone and mine type of the mine, and dynamically adjusts the threshold range in combination with the restoration time, so that the assessment standard is more in line with the actual situation of different restoration stages and avoids the assessment distortion caused by fixed threshold.

[0025] (5) Deep fusion of multi-source heterogeneous data to improve the accuracy and completeness of assessment data. This invention performs pixel-level registration and fusion and Kalman filtering fusion of three types of data sources: satellite multispectral remote sensing, UAV hyperspectral data and ground IoT data. This achieves comprehensive data coverage from macro to micro and from space to ground. The generated multidimensional ecological parameter dataset has high spatial resolution, high spectral information and high timeliness. Attached Figure Description

[0026] Figure 1 This is a block diagram of the open-pit mine ecological restoration effectiveness evaluation system of the present invention;

[0027] Figure 2 This is a flowchart of the multi-source heterogeneous data fusion module of the present invention;

[0028] Figure 3 This is a schematic diagram of the spatiotemporal coupling evaluation module of the present invention. Detailed Implementation

[0029] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] Example 1

[0031] This embodiment provides a system for evaluating the effectiveness of ecological restoration in open-pit mines, such as... Figure 1-3 As shown, it includes a multi-source heterogeneous data fusion module, a spatiotemporal coupling evaluation module, a multi-scale hierarchical evaluation module, and a restoration trajectory prediction module, and is applied to the evaluation of the ecological restoration effectiveness of an open-pit coal mine in a temperate semi-arid climate zone in northern China.

[0032] The spatiotemporal reference unification submodule of the multi-source heterogeneous data fusion module adopts the CGCS2000 national geodetic coordinate system as the unified spatial reference frame, and the time reference is unified to Beijing time (UTC+8). Through projection transformation, the WGS84 coordinates of satellite imagery, the local coordinate system of UAV imagery, and the latitude and longitude coordinates of ground sensors are unified to the CGCS2000 coordinate system. The pixel-level registration and fusion submodule adopts a registration algorithm based on mutual information. Using satellite multispectral imagery as the reference, UAV hyperspectral imagery is resampled to a unified grid with a spatial resolution of 2m. Ground sensor point data is allocated to the corresponding grid cells using an inverse distance weighted interpolation method. When there are multiple source observations at the same grid location, Kalman filtering is used for fusion. Satellite remote sensing inversion values ​​are used as prior estimates, UAV hyperspectral inversion values ​​are used as observation updates, and ground sensor data are used for real-time verification. The state transition matrix of the Kalman filter is set as the identity matrix, and the observation matrix is ​​determined according to the uncertainty covariance matrix of each data source. After fusion, a spatiotemporally aligned multidimensional ecological parameter dataset with a spatial resolution of 2m and a temporal resolution of 1 day is output.

[0033] In this embodiment, the satellite multispectral remote sensing imagery uses data from the Gaofen-2 satellite, with a spatial resolution of 0.8m (panchromatic) / 3.2m (multispectral). After panchromatic sharpening, the multispectral band resolution is improved to 0.8m, exceeding the minimum requirement of 2m; the revisit period is 4 days. The UAV hyperspectral imagery is acquired using a DJI M350 RTK equipped with a DJI L2 LiDAR and a custom pushbroom hyperspectral camera, with a spatial resolution of 0.03m (better than the 0.05m requirement), a spectral range of 400-1000nm, and 256 bands. The ground-based IoT sensor network includes: FDR soil moisture sensor (range 0-100%, accuracy ±2%), pH-200 soil pH sensor (range pH 3-10, accuracy ±0.1%), PT100 soil temperature sensor (range -40℃ to 80℃, accuracy ±0.3℃), SWL-1 surface runoff monitor (range 0-500L / min, accuracy ±1%), and WS-500 mini weather station (wind speed 0-60m / s, wind direction 0-360°, temperature -40℃ to 60℃, humidity 0-100%RH, rainfall 0-500mm / h), with a data acquisition interval of 30 minutes (better than the 1-hour requirement).

[0034] In the multidimensional ecological parameter dataset, vegetation cover is obtained by fusing the Normalized Difference Vegetation Index (NDVI) (the ratio of the difference between near-infrared and red band reflectance to their sum) calculated from satellite multispectral imagery with the chlorophyll a / b inversion results from UAV hyperspectral imagery (using a 685nm / 730nm dual-band ratio model). The fusion weights are adaptively determined by Kalman filtering. Soil physicochemical indicators include soil moisture content, pH value, and organic matter content. Moisture content and pH value are provided in real-time by ground-based IoT sensors, while organic matter content is obtained through partial least squares regression modeling in the short-wave infrared band (1700nm-2500nm) of UAV hyperspectral imagery. The topographic slope was obtained by inversion of the model, and the sensor measured values ​​and the spectral inversion values ​​were fused by Kalman filtering. The slope was calculated by using a digital elevation model (DEM, resolution 0.5m) generated by UAV oblique photogrammetry. The slope was calculated by fitting the plane within a 3×3 window to obtain the gradient using the Horn method. The hydrological parameters included surface runoff and catchment area. The surface runoff was measured by a surface runoff monitoring instrument, and the catchment area was calculated by the D8 flow direction algorithm of the DEM. The microclimate parameters included air temperature, humidity, wind speed and rainfall, which were measured by a micro weather station and fused with the daily surface temperature product of MODIS satellite (resolution 1km, downscaled to 2m grid by bilinear interpolation).

[0035] The spatiotemporal coupling assessment module constructs a spatial correlation graph model of ecological elements based on a graph attention network. The restoration area is divided into sampling units using a 2m×2m grid, with each sampling unit corresponding to a node in the graph model. The node feature vector is 5-dimensional, consisting of vegetation cover, comprehensive values ​​of soil physicochemical indicators (standardized mean values ​​of water content, pH value, and organic matter content), topographic slope, comprehensive values ​​of hydrological parameters (standardized mean values ​​of runoff and catchment area), and comprehensive values ​​of microclimate parameters (standardized mean values ​​of air temperature, humidity, wind speed, and rainfall). The values ​​of each dimension are standardized to the [0,1] interval using Min-Max. Graph edges are established based on the 8-neighborhood spatial adjacency relationship, that is, each node is connected to its adjacent nodes in the top, bottom, left, right, and four diagonal directions.

[0036] The multi-head graph attention mechanism uses 4-head attention, with each head having 8 dimensions. The attention coefficient is calculated using the formula: α ij =softmax j (LeakyReLU(a T [Wh i ||Wh j ]+λ·exp(-d ij / σ))), where W is the linear transformation matrix, h i h j These are the feature vectors of nodes i and j, respectively, where a is the attention vector and d is the feature vector of node i. ij Let be the Euclidean spatial distance between nodes i and j, σ be the distance decay parameter (set to 50m in this embodiment), λ be the ecological distance constraint weight coefficient (set to 0.3 in this embodiment), and exp(-d ij / σ) is an ecological distance constraint term, which makes the attention weight between nodes that are closer in space higher. Each node is spliced ​​together after being weighted and aggregated by 4 heads of attention to obtain a 32-dimensional update feature. After two layers of graph attention layer iteration, it is mapped to a spatial coupling evaluation value (the value range is [0,1], where 1 indicates the best spatial coordination state of the five elements and 0 indicates the worst).

[0037] The multi-scale hierarchical assessment module evaluates at three spatial scales: pit unit, mining area unit, and regional unit. A pit unit is a single pit or slope unit, typically ranging from 0.5 to 5 hectares in area; a mining area unit is the entire mine restoration area, containing multiple pit units; and a regional unit is the mine and its surrounding impact area, typically ranging from tens to hundreds of square kilometers in area. The comprehensive score calculation method for each scale is as follows: the pit unit score equals the area-weighted average of all spatially coupled assessment values ​​within that unit; the mining area unit score equals the area-weighted average of all pit unit scores within that mining area; and the regional unit score equals the ecological weighted average of all mining area unit scores within that region. The ecological weights are determined by normalizing the mean vegetation cover and biodiversity index of each mining area unit. The decomposition calibration method is as follows: the deviation between the regional unit score and the mining area unit score is allocated according to the ecological sensitivity coefficient of each mining pit unit (determined by the standardized weighted sum of three indicators: slope, soil erosion modulus and vegetation vulnerability), and added to the original score of the mining pit unit to obtain the calibrated score. The calibration range does not exceed ±10% of the original score.

[0038] The restoration trajectory prediction module is based on an attention-enhanced Long Short-Term Memory (LSTM) network. The encoder uses a two-layer LSTM with 128 hidden units per layer, encoding multidimensional ecological parameter time-series data from the past N time steps (N=36 in this embodiment, i.e., 36 months) into a hidden state sequence. The decoder adds a self-attention layer to a one-layer LSTM (128 hidden units) to weight and aggregate the hidden state sequence output by the encoder, calculating the contribution weight of each time step's hidden state to the prediction result. It then predicts the changing trends of vegetation cover, soil physicochemical indicators, hydrological parameters, and restoration effectiveness scores for the next M time steps (M=12 in this embodiment, i.e., 12 months). The training data consists of monthly monitoring data of the restoration area over the past three years, using the Adam optimizer with a learning rate of 0.001, a batch size of 32, and 200 training epochs. The training and validation sets are divided in an 8:2 ratio, and the early stopping strategy tolerates 10 epochs.

[0039] This embodiment is applied to a remediation area (3.2 km²) of an open-pit coal mine. 2 The remediation of this area has been underway for three years. The system output a time-series curve of the remediation effectiveness score over 36 months, with the score rising from 0.32 to 0.71, and predicting that the score will continue to rise to 0.83 in the next 12 months. It was found that the spatial coupling assessment values ​​of two mining pit units in the northwest of the mining area were significantly lower than those of the surrounding area (below 0.45). Analysis revealed that the soil pH value in this area was low (4.8), which hindered vegetation restoration. It is recommended to add lime improvement measures.

[0040] Example 2

[0041] This embodiment has the same core structure as Embodiment 1, the difference being that this embodiment is applied to the evaluation of the ecological restoration effectiveness of an open-pit metal mine in a subtropical humid climate zone in southern China, and a dynamic threshold calibration module is added.

[0042] The dynamic threshold calibration module, based on the climate zone and mine type of the restored mine, matches the parameters of the corresponding undisturbed ecosystem from the reference ecosystem database as the evaluation benchmark. In this embodiment, the climate zone is a subtropical humid region (annual average temperature 16-22℃, annual rainfall 1200-2000mm), the mine type is a metal mine (mainly copper mine, with a risk of heavy metal pollution), and the reference ecosystem is a secondary shrub-grassland ecosystem undisturbed by mining within the same climate zone. Its reference parameters are: vegetation cover 85%-95%, soil pH 5.5-7.0, organic matter content 3.0%-5.0%, area proportion of slope ≤25° ≥80%, and surface runoff coefficient ≤0.3. The reference ecosystem database is constructed from long-term fixed-location observation data of the China Ecosystem Research Network (CERN) and national forest resource inventory data, and is stored cross-categorized by climate zone (temperate semi-arid, temperate semi-humid, subtropical humid, tropical humid, plateau and alpine) and mine type (coal mine, metal mine, non-metal mine).

[0043] The method for adjusting the dynamic threshold range is as follows: the lower limit V of the threshold range corresponding to the repair duration t years. min (t)=V ref ×(1-exp(-kt)), where V ref To reference the baseline values ​​of the corresponding parameters in the ecosystem, k is the restoration rate coefficient (0.3 in this embodiment), and t is the restoration period. The lower limit of the vegetation cover threshold is 0.85×(1-exp(-0.3))=0.22 in the first year of restoration, 0.85×(1-exp(-0.9))=0.52 in the third year, 0.85×(1-exp(-1.5))=0.66 in the fifth year, and 0.85×(1-exp(-3.0))=0.81 in the tenth year. The threshold gradually increases with the restoration process to avoid the loss of differentiation due to excessively high standards in the early stage of restoration leading to low scores.

[0044] In this embodiment, the satellite multispectral remote sensing imagery uses Sentinel-2 data with a spatial resolution of 10m (upgraded to 5m after super-resolution reconstruction) and a revisit period of 5 days; the UAV hyperspectral imagery has a spatial resolution of 0.05m, a spectral range of 400-2500nm, and 400 bands; the ground-based IoT sensor network, based on Embodiment 1, adds a heavy metal content sensor (measuring the concentration of Cu, Zn, Pb ions, etc., with a range of 0-100mg / L and an accuracy of ±5%), with a data acquisition interval of 15 minutes.

[0045] In the graph attention network of this embodiment, the node feature vector is expanded to 6 dimensions, adding the heavy metal pollution index dimension to the 5-dimensional structure of Embodiment 1; the distance decay parameter σ in the ecological distance constraint term is adjusted to 30m (subtropical mountainous terrain is more fragmented, and spatial correlation decays faster); the prediction duration M of the repair trajectory prediction module is adjusted to 6 months (subtropical rainy season rainfall is concentrated, ecological parameters fluctuate greatly, long-term prediction uncertainty is high, shortening the prediction window improves reliability).

[0046] This embodiment was applied to a copper mine remediation area (1.8 km²). 2 The remediation has been underway for two years. The system identified three mining pit units with high heavy metal pollution indices (Cu concentration > 15 mg / L), and the spatial coupling assessment value was below the lower limit of the dynamic threshold (0.35 < 0.39). It is predicted that the remediation effectiveness score of this area will remain between 0.38 and 0.42 in the next 6 months without a significant upward trend. It is recommended to add passivation remediation measures.

[0047] Example 3

[0048] This embodiment has the same core structure as Embodiment 1, but the difference is that this embodiment is applied to the collaborative evaluation of multiple mining pits in a large open-pit mine (including 5 mining areas), and adds a visualization and early warning module for the evaluation results.

[0049] The evaluation results visualization and early warning module outputs the comprehensive score, spatial coupling evaluation value and predicted restoration effect score at each scale in three visualization forms: (1) Heat map, with the orthophoto of the restoration area as the base map, the spatial coupling evaluation value is mapped to the range of 0-1 and displayed as a green-yellow-red gradient, green indicates good synergy and red indicates poor synergy; (2) Time series curve, the time change trend of parameters such as comprehensive score, vegetation coverage and soil physicochemical index in each month since restoration is displayed in a line graph, and the extension curve of the prediction period and the 95% confidence interval are marked with dashed lines; (3) Three-dimensional terrain overlay map, the heat map is overlaid on the three-dimensional terrain model generated by the UAV, and the spatial distribution of restoration effect under different perspectives can be rotated to view.

[0050] The early warning mechanism is as follows: when the spatial coupling evaluation value of any sampling unit is lower than the preset early warning threshold (set to 0.4 in this embodiment), the system generates a yellow early warning and marks the location of the abnormal area with a flashing border on the heat map; when the predicted repair effectiveness score shows a continuous downward trend in the next 6 months (average monthly decrease > 0.02), the system generates a red early warning and outputs a quantitative analysis report of the downward trend, including the magnitude of the decrease, the area of ​​the affected region, and possible causes of degradation (inferred by analyzing the changing trends of parameters in each dimension).

[0051] In this embodiment, the cross-scale assessment result transfer process of the multi-scale hierarchical assessment module is as follows: Each of the five mining areas contains 8-15 mining pit units. First, the area-weighted average score of each mining pit unit is calculated to obtain the mining area-level score. The ecological weights of the five mining areas are 0.18, 0.22, 0.25, 0.20, and 0.15 respectively (determined based on the average vegetation cover and biodiversity index of each mining area). Weighted aggregation yields a regional unit score of 0.68. During decomposition calibration, the ecological sensitivity coefficients of three mining pit units in mining area 2 are relatively high (slope > 35°, erosion modulus > 5000 t / km²). 2 •a) The regional scoring bias was assigned a calibration amount of +0.05 to +0.08 to these three units, and the calibrated scores more accurately reflected the difficulty of repairing these highly sensitive areas.

[0052] This embodiment is applied to the remediation area of ​​a large open-pit coal mine (area 12.5 km²). 2 The area comprises five mining zones, and restoration has been underway for five years. The system outputs a three-dimensional topographic overlay map of the region, visually demonstrating the spatial differences in restoration effectiveness across the five mining zones. Three pit units in mining zone 2 triggered a yellow alert (spatial coupling assessment value 0.35-0.39). The prediction module shows that one of these units will experience a downward trend in its score over the next six months (average monthly decrease of 0.025), triggering a red alert. Analysis suggests that the degradation is due to poor slope drainage in the area, leading to persistently high soil moisture content and restricted root growth due to oxygen deficiency. It is recommended to construct additional drainage ditches.

[0053] Example 4

[0054] This embodiment provides a complete application scheme for an open-pit mine ecological restoration effectiveness evaluation system, including all modules of Embodiment 1, the dynamic threshold calibration module of Embodiment 2, and the evaluation result visualization and early warning module of Embodiment 3. It is applied to the ecological restoration effectiveness evaluation of an open-pit limestone mine in a northern temperate semi-humid climate zone.

[0055] The overall system architecture of this embodiment is as follows: Figure 1 As shown, the workflow is as follows: The multi-source heterogeneous data fusion module first performs spatiotemporal benchmark unification and pixel-level registration and fusion on the three types of data sources to generate a multi-dimensional ecological parameter dataset (such as...). Figure 2 As shown); the spatiotemporal coupling evaluation module performs spatial coupling modeling on the five-dimensional parameters in the dataset based on a graph attention network, and outputs the spatial coupling evaluation value of each sampling unit (e.g. Figure 3 (As shown in the figure) The multi-scale hierarchical assessment module calculates the comprehensive score and performs cross-scale calibration at three scales: pit, mining area, and region; the repair trajectory prediction module predicts the future repair trend based on attention-enhanced LSTM; the dynamic threshold calibration module dynamically adjusts the assessment threshold according to climate zoning and repair duration; and the assessment result visualization and early warning module outputs visualization charts and early warning information.

[0056] In this embodiment, the satellite multispectral remote sensing imagery uses Planet SkySat data with a spatial resolution of 0.5m and a revisit period of 1 day; the UAV hyperspectral imagery has a spatial resolution of 0.02m, a spectral range of 400-1700nm, and 300 bands; the ground-based IoT sensor network includes a soil moisture sensor (range 0-100%, accuracy ±1.5%), a soil pH sensor (range pH 3-10, accuracy ±0.05%), a soil temperature sensor (range -30℃ to 70℃, accuracy ±0.2℃), a surface runoff monitor (range 0-300L / min, accuracy ±0.5%), and a micro weather station (wind speed 0-50m / s, wind direction 0-360°, temperature -30℃ to 50℃, humidity 0-100%RH, rainfall 0-200mm / h), with a data acquisition interval of 10 minutes.

[0057] The graph attention network employs a 6-head attention mechanism, with each head having 8 dimensions. After two layers of graph attention iterations, it outputs 48-dimensional node features, which are then mapped to spatial coupling evaluation values ​​via a two-layer fully connected network (128-dimensional → 64-dimensional → 1-dimensional). The distance decay parameter σ in the ecological distance constraint term is set to 40m, and the constraint weight coefficient λ is set to 0.25. The encoder of the restoration trajectory prediction module uses a 3-layer LSTM (256 hidden units per layer), and the decoder uses a 1-layer LSTM (256 hidden units) plus a self-attention layer to predict the restoration trend over the next 24 months. In the dynamic threshold calibration module, the restoration rate coefficient k is set to 0.25 (limestone mine soil remediation is slower, with a lower restoration rate than coal mines), and the reference ecosystem is a limestone mountain shrubland in a temperate semi-humid climate zone.

[0058] This embodiment was applied to a limestone mine remediation area (area 5.6 km²). 2 The restoration has been underway for four years. System assessment results show: the overall regional score is 0.72 (good), the scores for four mining pit units are 0.81-0.88, the scores for two slope units are 0.52-0.59, and the score for one spoil heap unit is 0.38 (triggering a yellow alert). It is predicted that the regional score will rise to 0.79 in 24 months, and the spoil heap unit score will rise to 0.48, but will still be below the dynamic threshold of 0.50. It is recommended that additional soil covering and vegetation replanting measures be implemented for the spoil heap units. The 3D topographic overlay map output by the visualization module clearly shows the spatial distribution comparison between high-scoring areas (mining pit backfill areas) and low-scoring areas (exposed spoil heap areas), providing an intuitive basis for subsequent restoration resource allocation.

[0059] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A system for evaluating the effectiveness of ecological restoration in open-pit mines, characterized in that, include: The multi-source heterogeneous data fusion module is used to acquire satellite multispectral remote sensing images, UAV hyperspectral images and monitoring data from ground IoT sensor networks. It performs spatiotemporal benchmark unification and pixel-level registration fusion on the multi-source data to generate a spatiotemporally aligned multidimensional ecological parameter dataset. The multidimensional ecological parameter dataset includes spatiotemporally continuous data in five dimensions: vegetation cover, soil physicochemical indicators, topographic slope, hydrological parameters and microclimate parameters. The spatiotemporal coupling assessment module constructs a spatial correlation graph model of ecological elements based on a graph attention network. It uses each sampling unit in the restoration area as a graph node and the spatial adjacency relationship between nodes as graph edges. The parameters of each dimension in the multidimensional ecological parameter dataset are used as node feature vector inputs. The module learns the spatial coupling weights between different ecological elements through a multi-head graph attention mechanism and outputs the spatial coupling assessment value of each sampling unit. The spatial coupling assessment value reflects the spatial coordination state of the five elements of vegetation, soil, topography, hydrology and microclimate. The multi-scale hierarchical assessment module calculates the comprehensive ecological restoration score at each of the three spatial scales: pit unit, mining area unit, and regional unit. It also uses scale aggregation and decomposition methods to achieve upward aggregation and downward decomposition calibration of cross-scale assessment results. The restoration trajectory prediction module, based on an attention-enhanced long short-term memory network, takes the temporal data of the spatiotemporally aligned multidimensional ecological parameter dataset as input, learns the temporal patterns of ecological restoration evolution, and predicts the changing trends of various ecological parameters and restoration effectiveness scores within a preset future time period.

2. The open-pit mine ecological restoration effectiveness evaluation system according to claim 1, characterized in that, The multi-source heterogeneous data fusion module includes a spatiotemporal reference unification submodule and a pixel-level registration and fusion submodule. The spatiotemporal reference unification submodule unifies satellite imagery, UAV imagery, and ground sensor data to the same spatial coordinate system and time reference. The pixel-level registration and fusion submodule uses a mutual information-based registration algorithm to resample data with different spatial resolutions to a unified grid, and fuses heterogeneous observations at the same spatiotemporal location through Kalman filtering to generate the multidimensional ecological parameter dataset with consistent spatial resolution.

3. The open-pit mine ecological restoration effectiveness evaluation system according to claim 1, characterized in that, In the spatiotemporal coupling evaluation module, the calculation process of the multi-head graph attention mechanism is as follows: for each graph node, calculate its attention coefficient with all neighboring nodes, and sum the attention coefficient with the feature vectors of the neighboring nodes in a weighted manner to obtain the updated feature of the node. After iteration through multiple graph attention layers, the final node feature is mapped to the spatial coupling evaluation value through a fully connected layer. The calculation of the attention coefficient incorporates an ecological distance constraint, which makes the attention weight higher between nodes that are spatially closer.

4. The open-pit mine ecological restoration effectiveness evaluation system according to claim 1, characterized in that, In the multi-scale hierarchical assessment module, the scale aggregation method is as follows: based on the comprehensive score of the mining pit unit, the area-weighted average is aggregated to obtain the mining area unit score, and then aggregated to obtain the regional unit score according to the ecological weight of the mining area unit; the decomposition and calibration method is as follows: the regional unit score is decomposed according to the ecological sensitivity coefficient of each mining pit unit to obtain the calibrated mining pit unit score, which is used to correct local assessment bias.

5. The open-pit mine ecological restoration effectiveness evaluation system according to claim 1, characterized in that, In the repair trajectory prediction module, the attention mechanism-enhanced long short-term memory network includes an encoder and a decoder; the encoder is composed of multi-layer LSTM units, which encode the input temporal data to extract temporal features; The decoder adds a self-attention layer to the LSTM to perform weighted aggregation of the hidden state sequence output by the encoder, and predicts the changing trend of each ecological parameter and the restoration effectiveness score within a preset time period.

6. The open-pit mine ecological restoration effectiveness evaluation system according to claim 1, characterized in that, The system also includes a dynamic threshold calibration module. Based on the climate zone and mine type of the restored mine, the dynamic threshold calibration module matches the corresponding undisturbed ecosystem parameters from the reference ecosystem database as the evaluation benchmark, and dynamically adjusts the threshold range in combination with the restoration time, so that the evaluation standard adapts to the restoration process.

7. The open-pit mine ecological restoration effectiveness evaluation system according to claim 1, characterized in that, The system also includes an evaluation result visualization and early warning module. The evaluation result visualization and early warning module outputs the comprehensive score, spatial coupling evaluation value and predicted restoration effect score at each scale in the form of heat map, time series curve and three-dimensional terrain overlay map. When the spatial coupling evaluation value of any sampling unit is lower than the preset early warning threshold or the predicted restoration effect score shows a downward trend, early warning information is generated and the location of the abnormal area is marked.

8. The open-pit mine ecological restoration effectiveness evaluation system according to claim 1, characterized in that, The ground-based IoT sensor network includes soil moisture sensors, soil pH sensors, soil temperature sensors, surface runoff monitors, and micro weather stations; the spatial resolution of the satellite multispectral remote sensing imagery is no less than 2 meters, the spatial resolution of the UAV hyperspectral imagery is no less than 0.05 meters, and the data acquisition interval of the ground-based IoT sensor network is no more than 1 hour.

9. The open-pit mine ecological restoration effectiveness evaluation system according to claim 1, characterized in that, In the multidimensional ecological parameter dataset, vegetation cover is obtained by fusing the normalized vegetation index of satellite multispectral imagery with the chlorophyll inversion results of UAV hyperspectral imagery; soil physicochemical indicators are obtained by fusing real-time monitoring data from ground IoT sensors with the soil spectral inversion results of UAV hyperspectral imagery; topographic slope is calculated using a digital elevation model generated by UAV oblique photogrammetry; hydrological parameters are obtained by fusing surface runoff monitoring data with the hydrological analysis results of the digital elevation model; and microclimate parameters are obtained by fusing micro-weather station data with satellite remote sensing surface temperature products.

Images