Regional ecological restoration state evaluation method and system based on image recognition

Abstract

The invention discloses a regionalized ecological restoration state evaluation method and system based on image recognition, and relates to the technical field of ecological restoration analysis, and the method comprises the steps: determining an ecological reference region, a plurality of to-be-restored regions, and a plurality of ecological monitoring points; the method comprises the following steps: establishing a rule engine in an ecological reference area, generating a first succession curve, issuing the first succession curve to a plurality of to-be-restored areas, loading environment data to execute parameter compensation, generating a second succession curve, extracting trend characteristics, combining multispectral image data, setting a multi-condition encoder, determining an initial restoration state, and establishing a plurality of succession change partitions. Synchronizing to a preset ecological association topology network output spatio-temporal index table, introducing a regression model, analyzing a succession process of a second succession curve, and predicting a succession change partition where ecological severe degradation occurs; according to the method, overall management of ecological benchmark, restoration and environment is realized, and the prediction accuracy of severe ecological degradation is improved.

Description

Technical Field

[0001] This invention relates to the field of ecological restoration analysis technology, specifically to a regional ecological restoration status assessment method and system based on image recognition. Background Technology

[0002] Ecological restoration projects are a key means to address regional ecological degradation and enhance environmental carrying capacity. They are widely used in areas such as mine revegetation, wetland reconstruction, and desertification control. In the full life cycle management of ecological restoration, continuous and accurate monitoring and evaluation of the restored area are the core links to ensure restoration effectiveness and prevent secondary degradation.

[0003] When facing large-scale, complex ecological restoration scenarios, the following problems exist: On the one hand, existing technologies often rely on a single baseline succession curve, ignoring the influence of geological and meteorological data of the area to be restored, making it difficult to correctly analyze the collected image data and greatly reducing the accuracy of restoration status assessment; on the other hand, existing technologies usually process the image semantic features of multispectral images and the temporal trend features of sensor data separately, lacking a deep fusion coding mechanism, and are unable to adaptively divide the time partition granularity according to the current succession status level, which easily leads to sparse sampling in a certain time window and makes it difficult to identify early degradation in a timely manner; at the same time, existing systems only perform image recognition and independent evaluation on a certain partition, and cannot analyze the succession change process between regions. For example, upstream soil erosion causes related downstream ecological degradation, resulting in alarms only being triggered after degradation has occurred and caused significant consequences, reducing the efficiency and accuracy of ecological restoration. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a regionalized ecological restoration status assessment method and system based on image recognition. It identifies an ecological baseline region, multiple regions to be restored, and multiple ecological monitoring points. A first succession curve is generated within the ecological baseline region, and a second succession curve adapted to the regions to be restored is constructed through environmental compensation. A multi-condition encoder is used to determine the initial restoration state and adaptively partition the succession changes. A topological network and regression model are used to predict the next succession change zone to experience severe degradation, thereby improving the efficiency of ecological restoration and solving the problems mentioned in the background technology.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, this application provides a regional ecological restoration status assessment method based on image recognition, the method comprising: determining an ecological baseline area, multiple areas to be restored, and multiple ecological monitoring points; Specifically, a rule engine is built within the ecological baseline area to generate the first succession curve. The first succession curve is then distributed to multiple areas to be restored to perform cascade calibration. Environmental data is loaded to perform parameter compensation on the first succession curve, thereby generating the second succession curve. Based on the second succession curve, trend features are extracted, and multispectral image data is combined to set up a multi-condition encoder to determine the initial repair state. Multiple succession change partitions are established in each area to be repaired according to the time series. Based on the succession change partitions and synchronized to the pre-defined ecological association topology network output spatiotemporal index table, a regression model is introduced to analyze the succession process of the second succession curve and predict the next succession change partitions where ecological degradation will occur.

[0006] Furthermore, ecological baseline areas, multiple areas requiring restoration, and multiple ecological monitoring points were identified, including: Upon receiving a monitoring command, the system acquires the geographical location, spectral code, and sampling frequency of the monitoring terminal, and generates a unique hash value. The monitoring terminal is equipped with a multispectral imaging sensor, a soil sensor, and a meteorological sensor. Meanwhile, node attributes are determined at the topological hierarchy in the preset ecological association topology network; among them, the root node is the ecological baseline area, the intermediate cascade nodes are the areas to be restored, and the monitoring terminal acquisition nodes are the ecological monitoring points, which are used to collect spectral image data and environmental data; among them, the spectral image data includes image data of topography, soil, and vegetation cover, and the environmental data includes meteorological parameters and geological parameters.

[0007] Furthermore, a rules engine is built to generate the first succession curve, including: Image data within the ecological baseline area were acquired, and image recognition algorithms were used to extract the normalized vegetation index and reflectance values ​​at N spectral band positions. Volatility is calculated based on the normalized vegetation index, and the volatility is compared with the standard volatility threshold to determine the succession steady state level. For the N spectral band positions, perform iterative cross-sparsening to divide the spectral bands into Class I bands and Class II bands; where Class I bands represent the odd-numbered positions of the spectral bands arranged in order of their center wavelengths, and Class II bands represent the even-numbered positions of the spectral bands arranged in order of their center wavelengths. Reconstruction processing is performed on Class I and Class II bands to screen spectral bands that meet the corresponding succession steady-state level, including core bands and related bands; the reflection gradient difference between related bands and corresponding core bands is calculated to determine the succession coefficients of each spectral band position; A blank ecological succession trajectory flow is preset, and the reflection value of the core band is weighted and calibrated based on the succession coefficient to obtain the first succession curve.

[0008] Furthermore, by comparing volatility with the standard volatility threshold, the succession steady-state level is determined, including: If the volatility is less than or equal to the standard volatility threshold, mark the current volatility as 'a' and edit it as a first-level character. Combine 'a' and the first-level character to generate a high steady-state level. If the volatility is greater than the standard volatility threshold, the current volatility is marked as 'b' and edited as a secondary character. The 'b' and the secondary character are then combined to generate a low steady-state level.

[0009] Furthermore, loading environmental data performs parameter compensation on the first succession curve, including: Meteorological and geological parameters were retrieved from multiple areas to be repaired. The meteorological parameters included atmospheric conductivity, ambient temperature and humidity, and solar radiation intensity, while the geological parameters included soil salinity, soil moisture content, and soil pH. Establish the first correlation between meteorological parameters and the first successional curve, establish the second correlation between geological parameters and the first successional curve, and construct a compensation factor generation model based on the first and second correlations. The compensation factors for each area to be repaired are generated based on the compensation factor generation model, including: determining the fluctuation deviation of meteorological parameters and determining the ratio deviation of geological parameters; Based on the time series, a three-dimensional change graph between fluctuation deviation and ratio deviation is established and overlaid with a standard three-dimensional change graph to obtain the corresponding horizontal and vertical screenshots. The overlapping total area and non-overlapping total area are then compared to generate the graph. The ratio of the non-overlapping total area to the sum of the overlapping and non-overlapping total areas is marked as a compensation factor. The first succession curve is then compensated and corrected to obtain the second succession curve.

[0010] Furthermore, a multi-condition encoder is set up to determine the initial repair state, including: A multi-condition encoder includes a timing encoder, a visual encoder, and a decoding unit; The amplitude and succession rate of the second succession curve are extracted using a time-series encoder as trend features. Semantic and texture features are extracted by performing convolution operations on multispectral image data using a visual encoder. Feature fusion is performed on trend features, semantic features, and texture features to obtain a joint feature tensor. The joint feature tensor is then solved by the decoding unit to output the initial repair state. The value of the initial repair state is represented by a probability value.

[0011] Furthermore, multiple successional change partitions are established in each region to be repaired according to the time series, including: Based on the initial restoration state, obtain the collection cycle timestamps and sampling frequency ranges of the ecological monitoring points, and determine the minimum sampling interval based on the sampling frequency range; Standard time granularity is extracted based on the second succession curve; The dynamic partitioning step size is determined based on the minimum sampling interval and the standard time granularity. The timestamp corresponding to the initial repair state is used as the starting anchor point. The acquisition cycle is discretized according to the dynamic partitioning step size to obtain multiple successive change partitions.

[0012] Furthermore, each successional change partition is mapped to a temporal attribute of a topological node.

[0013] Furthermore, the evolution process of the second succession curve is analyzed, including: Calculate the deviation trajectory of the second succession curve relative to the first succession curve, identify the positive and negative slopes of the deviation trajectory, and extract the mean and standard deviation to construct multiple succession monitoring vectors; The current compensation factor is retrieved, and a regression model is used to establish multiple correlation expressions between each succession monitoring vector and the compensation factor. The coefficients of determination of the correlation expressions are extracted, and the cases of severe degradation are screened by sorting in descending order. Combined with the spatiotemporal index table, the next succession change partition to experience severe ecological degradation is predicted. The spatiotemporal index table includes a set of node attributes, a number of topological levels, and a set of associated succession change partitions.

[0014] Secondly, this application provides a regional ecological restoration status assessment method and system based on image recognition. The system includes: a region determination module: determining an ecological baseline region, multiple regions to be restored, and multiple ecological monitoring points; Specifically, a rule engine is built within the ecological baseline area to generate the first succession curve. The first succession curve is then distributed to multiple areas to be restored to perform cascade calibration. Environmental data is loaded to perform parameter compensation on the first succession curve, thereby generating the second succession curve. State assessment module: Based on the second succession curve, trend features are extracted, multispectral image data is combined, a multi-condition encoder is set to determine the initial repair state, and multiple succession change partitions are established in each area to be repaired according to the time series. Degradation prediction module: Based on succession change partitions and synchronized to the preset ecological association topology network output spatiotemporal index table, a regression model is introduced to analyze the succession process of the second succession curve and predict the next succession change partition where ecological degradation will occur.

[0015] (III) Beneficial Effects This invention provides a method and system for regional ecological restoration status assessment based on image recognition, which has the following beneficial effects: 1. This invention combines the rule engine of the ecological benchmark area, determines the corresponding succession coefficient by the ratio of volatility to standard volatility threshold, and constructs a first succession curve; by loading environmental data, it performs parameter compensation on the first succession curve to generate a second succession curve that is adapted to multiple areas to be restored, thereby realizing the overall management of ecological benchmarks, restoration, and environmental elements. 2. This invention solves the problem of incomplete representation of single-modal data by setting up a multi-condition encoder to couple the semantic features of multispectral images with the trend features of the second succession curve and output a numerical initial restoration state. At the same time, based on the quality of the initial restoration state, the acquisition cycle is dynamically adjusted and multiple succession change partitions are divided. This adaptive partitioning mechanism effectively balances monitoring accuracy and computing resources, reduces the missed detection rate or computing power waste caused by improper sampling intervals, and significantly improves the timeliness of ecological restoration assessment. 3. This invention uses a regression model to establish multiple correlation expressions between each succession monitoring vector and compensation factor, screens out situations where ecological degradation occurs, and predicts the next succession change zone where ecological degradation will occur through a spatiotemporal index table, thereby improving the accuracy and robustness of prediction. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a regionalized ecological restoration status assessment method according to an exemplary embodiment; Figure 2 This is a schematic diagram of a module of a regionalized ecological restoration status assessment system according to an exemplary embodiment. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] The core of this invention lies in constructing a multi-regional ecological assessment system for ecological restoration scenarios. A rule engine is built within the ecological benchmark area to construct a first succession curve. By loading environmental data and performing parameter compensation, a second succession curve adapted to multiple areas to be restored is generated. Combined with a multi-condition encoder, the initial restoration state is determined and the succession change partition is adaptively configured. The generated succession change partitions are synchronously mapped to a preset ecological association topology network. A regression model is introduced to deeply analyze the succession process of the second succession curve in each partition, predict the next succession change partition where ecological degradation will occur, and improve the success rate of ecological restoration and operation and maintenance efficiency.

[0019] Example 1: This invention provides a method for regional ecological restoration status assessment based on image recognition; Figure 1 This is a flowchart illustrating a regionalized ecological restoration status assessment method according to an exemplary embodiment; please refer to [link / reference]. Figure 1 The method includes the following steps: S1: Identify the ecological baseline area, multiple areas to be restored, and multiple ecological monitoring points; Specifically, a rule engine is built within the ecological baseline area to generate the first succession curve. The first succession curve is then distributed to multiple areas to be restored to perform cascade calibration. Environmental data is loaded to perform parameter compensation on the first succession curve to generate the second succession curve. The ecological baseline area, multiple areas to be restored, and multiple ecological monitoring points were identified, including: Upon receiving a monitoring command, the system acquires the monitoring terminal's geographical location, spectral code, and sensor-configured sampling frequency. It then concatenates these strings to form an original message. Using a pre-defined hash algorithm, such as MD5, it performs a hash operation on the original message to generate a unique hash value. The monitoring terminal is equipped with a multispectral imaging sensor, a soil sensor, and a meteorological sensor. It should be noted that the geographical location refers to the latitude, longitude, and altitude information of the deployed sensors; the spectral code is the hardware identification code of the built-in photosensitive device in the multispectral imaging sensor; and the sensor-configured sampling frequency refers to the sampling frequency of the multispectral imaging sensor, the soil sensor, and the meteorological sensor. Simultaneously, node attributes are determined at the topological hierarchy of the preset ecological association topology network; among them, the root node is the ecological baseline area, the intermediate cascade nodes are the areas to be restored, and the monitoring terminal acquisition nodes are the ecological monitoring points, which are used to collect spectral image data and environmental data; among them, the spectral image data includes image data of topography, soil, and vegetation cover, and the environmental data includes meteorological parameters and geological parameters. Specifically, this includes: acquiring spectral image data and environmental data by using multispectral imaging sensors, soil sensors, and meteorological sensors mounted at ecological monitoring points; multispectral imaging sensors are used to acquire topographic, soil, and canopy images of vegetation under different wavelengths, facilitating subsequent extraction of Normalized Difference Vegetation Index (NDVI) features for restoration status assessment; soil sensors, including soil salinity sensors, pH sensors, and moisture sensors, are used to acquire geological parameters in environmental data, including soil salinity, soil moisture content, and soil pH; meteorological sensors, including atmospheric electric field monitors, temperature and humidity sensors, and radiation sensors, are used to acquire meteorological parameters in real-time environmental data, including rainfall, atmospheric conductivity, ambient temperature, ambient humidity, and solar radiation intensity. All of the above sensors are shown in the figure and are adaptively installed at the ecological monitoring points. The significance of the above analysis lies in the fact that by constructing a multi-regional ecological perception system, including ecological baseline areas, multiple areas to be restored, and multiple ecological monitoring points, and by using monitoring terminals equipped with multispectral imaging sensors, soil sensors, and meteorological sensors, it is possible to capture the spectral reflectance characteristics of vegetation and habitat microenvironment data in an all-weather, non-contact manner, thereby achieving three-dimensional perception of the ecological site. Furthermore, the equipment has edge computing capabilities, which can reduce data transmission bandwidth costs.

[0020] Build a rules engine to generate the first succession curve, including: Image data within the ecological baseline area is acquired, and image recognition algorithms are used to extract the normalized vegetation index and reflectance values ​​at N spectral band positions, where N is greater than 0. Specifically, this includes: performing radiometric calibration and atmospheric correction on the image data to eliminate sensor errors and atmospheric scattering interference; introducing a pre-trained semantic segmentation model based on the U-Net architecture to perform pixel-level classification on the corrected image data, identifying and segmenting vegetation-covered areas, generating a binary vegetation mask, and performing matrix multiplication between the binary vegetation mask and the image data to filter out non-vegetation background noise; extracting the pixel matrix of the red and near-infrared bands within the mask area to obtain the normalized vegetation index; simultaneously, based on the pixel coordinates of the mask area, indexing and extracting the pixel grayscale values ​​of N preset specific spectral channels in the image data, such as blue, green, and red-edge bands, to construct the corresponding N-dimensional spectral reflectance values, achieving accurate and localized reflectance value acquisition. Based on the time series, normalized vegetation indices (NDIs) for multiple time series are generated, and linear regression fitting is performed to generate NDI trend lines. Variance calculation is performed on the NDI trend lines to obtain numerical volatility. By comparing the volatility with a standard volatility threshold, the succession steady-state level is determined: if the volatility is less than or equal to the standard volatility threshold, the current volatility is marked as 'a' and edited as a level 1 character. The combination of 'a' and the level 1 character generates a high steady-state level, indicating that the ecosystem in this region is extremely stable and has strong resistance to disturbances; if the volatility is greater than the standard volatility threshold, the current volatility is... The rate is marked as 'b' and edited as a second-level character. The combination of 'b' and the second-level character generates a low steady-state level, indicating that the ecology in this area is extremely fragile and highly susceptible to environmental influences. The standard fluctuation threshold is as follows: retrieve historical normalized vegetation index data of the ecological benchmark area, extract the corresponding historical volatility according to the time series, calculate the mean and standard deviation of the historical volatility, and set the standard fluctuation threshold as the sum of the mean and twice the standard deviation. It should be noted that the value of the multiple is just an example and should be set according to the actual situation, which will not be elaborated here. For the N spectral band positions, perform iterative cross-sparsening, sort the spectral bands in order of center wavelength from shortest to longest, extract the odd-numbered positions at 1, 3, 5...m to form a first class of bands, and extract the even-numbered positions at 2, 4, 6...m+1 to form a second class of bands; where m is an odd number. For the reconstruction processing of Class I and Class II bands, the reflection value of one type of band is used to simulate and reconstruct the reflection value of the other type of band, and the absolute value of the difference between the reconstructed simulated value and the reflection value is calculated as the reconstruction loss. Bands that meet the corresponding successional steady-state level are selected, specifically including: selecting bands that meet the preset reconstruction accuracy threshold of the corresponding successional steady-state level; automatically matching the corresponding reconstruction accuracy threshold based on different successional steady-state levels: for high steady-state levels, due to ecological stability and clear ecological succession change patterns, a lower reconstruction accuracy threshold is set to allow for a larger degree of sparsity processing, and the corresponding bands are marked as relevant bands; for low successional levels, due to ecological sensitivity and large fluctuations in ecological succession change patterns, a higher reconstruction accuracy threshold is set to ensure that every tiny degradation signal is captured, and the corresponding bands are marked as core bands. Based on N spectral band positions, for any relevant band, a bidirectional index is performed. The core band closest to its relevant band position along the wavelength decreasing direction is determined as the first adjacent core band; the core band closest to its relevant band position along the wavelength increasing direction is determined as the second adjacent core band. For example, because of the aforementioned odd-even alternating sparse processing method, for any relevant band at an even position, its corresponding first and second adjacent core bands are located one position less than and one position greater than its position index, respectively. It should be noted that if a relevant band is located at the boundary of the N spectral band positions, resulting in no adjacent core bands on one side, the two closest core bands on the same side are automatically selected as the corresponding adjacent core bands to ensure the continuity of subsequent reflection gradient difference calculations. The reflection values ​​corresponding to the relevant band, the first adjacent core band, and the second adjacent core band are obtained. The arithmetic mean of the reflection values ​​corresponding to the first adjacent core band and the second adjacent core band is obtained to obtain the mean reflection value. Each reflection value of the relevant band is compared with the mean reflection value of its adjacent core bands, and the absolute value of the corresponding difference is calculated and marked as the reflection gradient difference. The ratio of the reflection gradient difference to the sum of the reflection gradient differences of all eliminated bands is calculated, and the ratio is defined as the succession coefficient. It should be noted that the larger the reflection gradient difference, the more unique succession information that the band carries that cannot be replaced by neighboring bands, and the higher the corresponding succession coefficient. The smaller the reflection gradient difference, the lower the corresponding succession coefficient. A blank ecological succession trajectory flow is preset, and the reflection value of the core band is weighted and calibrated based on the succession coefficient. The inverse setting of the core band is multiplied by the corresponding succession coefficient and used as the ecological succession trajectory point. The first succession curve is obtained by fitting. It should be noted that, to a certain extent, the succession information of the relevant bands is projected back into the sampled value of the core band, so as to achieve high-fidelity restoration of the ecological succession trajectory covering the feature information of the entire band while reducing the data processing dimension.

[0021] Load environmental data to perform parameter compensation on the first succession curve, generating a second succession curve, including: Meteorological and geological parameters were retrieved from multiple areas to be repaired. The meteorological parameters included atmospheric conductivity, ambient temperature, ambient humidity, and solar radiation intensity. The geological parameters included soil salinity, soil moisture content, and soil pH. Establish the first correlation between meteorological parameters and the first successional curve, establish the second correlation between geological parameters and the first successional curve, and construct a compensation factor generation model based on the first and second correlations. Compensation factors for each region to be repaired are generated based on the compensation factor generation model, including: Determine the fluctuation deviations of meteorological parameters: Compare atmospheric conductivity, ambient temperature, ambient humidity, and solar radiation intensity with their corresponding preset benchmark parameters, including: calculating the absolute value of the difference between atmospheric conductivity and preset benchmark conductivity to obtain the first deviation; calculating the absolute value of the difference between ambient temperature and preset benchmark ambient temperature to obtain the second deviation; calculating the absolute value of the difference between ambient humidity and preset benchmark ambient humidity to obtain the third deviation; calculating the absolute value of the difference between solar radiation intensity and preset benchmark solar radiation intensity to obtain the fourth deviation; assign a first weight, a second weight, a third weight, and a fourth weight to the first deviation, the second deviation, the third deviation, and the fourth deviation, respectively, and the sum of the first weight, the second weight, the third weight, and the fourth weight is 1; it should be noted that normalization processing is required during the calculation process to eliminate the influence of dimensions; First weight: Atmospheric conductivity directly reflects regional air pollutants, such as dust particles and precipitation ion concentration. A larger first deviation value indicates abnormal atmospheric conductivity; for example, excessively high conductivity indicates pollutant accumulation, easily leading to clogged stomata in vegetation leaves and soil acidification; excessively low conductivity indicates insufficient atmospheric nutrient ions, which is detrimental to soil microbial activation. During image recognition, the monitored greenness level of vegetation leaves will be directly reduced, thus automatically increasing the first weight. Conversely, a smaller first deviation value indicates normal atmospheric conductivity, with minimal interference to the ecological environment, and the first weight will decrease accordingly. The calculation is performed by looking up a preset nonlinear mapping table, such as based on... Image recognition identifies the greenness of plant leaves and assigns corresponding weights for adaptive weighting. A second weight is assigned based on environmental temperature, which directly determines plant growth. Image recognition can capture indicators of abnormal environmental temperature, such as plant height. Larger deviations can lead to high temperatures and stunted growth. To address this, a tiered weighting system is used, such as a preset second deviation threshold. For example, when the second deviation is below ±2℃, the second weight automatically decreases, indicating limited impact of temperature fluctuations on the restoration process and no significant abnormalities in plant growth monitored by image recognition. Conversely, when the second deviation is above ±5℃, the second weight increases accordingly, highlighting extreme temperature fluctuations. The destructive effect of extreme temperatures on ecological restoration; it should be noted that ±2℃ and ±5℃ are only example indicators, and specific parameters should be set according to actual conditions; third weight: environmental humidity directly determines the soil water retention capacity of vegetation. Image recognition can capture the manifestations caused by abnormal humidity, such as soil cracking and wilting of vegetation leaves; prolonged high or low humidity will exacerbate the restoration risk. Excessive humidity can easily cause root rot and soil compaction, while excessively low humidity will lead to seedling death and reduced vegetation coverage; the calculation is performed using a tiered weighting system, such as: a preset third deviation threshold. If the third deviation is greater than or equal to the preset third deviation threshold, the third weight is adjusted accordingly. High value; if the third deviation is less than the preset third deviation threshold, the third weight takes a lower value; fourth weight: light radiation intensity determines the photosynthetic efficiency of vegetation. Image recognition can identify the characteristics caused by abnormal light, such as: yellowing of vegetation leaves and uneven coverage; if the fourth deviation is larger, the impact is more significant: excessive light can easily scorch vegetation leaves and inhibit the growth of shade-loving pioneer plants; insufficient light will lead to insufficient photosynthetic efficiency of vegetation, weak growth, and difficulty in achieving the required vegetation coverage; during calculation, the corresponding weight is obtained by looking up the preset piecewise nonlinear mapping table; the larger the fourth deviation, the higher the fourth weight is assigned, and the smaller the fourth deviation, the lower the fourth weight is assigned. The first deviation, second deviation, third deviation, and fourth deviation are multiplied by their corresponding first to fourth weights to obtain the weighted values ​​of each indicator. The weighted values ​​are then summed, and the summation result is marked as the fluctuation deviation. Determine the ratio deviation of geological parameters: Calculate multiple ratios of geological parameters to preset benchmark intervals, including: calculating the ratio of soil salinity to the median of the preset benchmark salinity interval to obtain the first ratio; calculating the ratio of soil moisture to the median of the preset benchmark moisture interval to obtain the second ratio; calculating the ratio of soil pH to the median of the preset benchmark pH interval to obtain the third ratio; assign a fifth weight to the first ratio, a sixth weight to the second ratio, and a seventh weight to the third ratio, with the sum of the fifth, sixth, and seventh weights being 1; multiply the first, second, and third ratios by their corresponding fifth to sixth weights to obtain the weighted values ​​of each indicator, sum the weighted values, and mark the sum as the ratio deviation; it should be noted that normalization processing is required during the calculation process to eliminate the influence of dimensions; The fifth, sixth, and seventh weights were determined using the coefficient of variation method. This method assigns weights to each indicator based on the degree of variation between the current and target values. If the numerical difference of an indicator is large, it can clearly distinguish each evaluated object, indicating that the indicator has rich discriminative information and should therefore be given a larger weight. Conversely, if the numerical difference of each evaluated object on a certain indicator is small, then the indicator's ability to distinguish each evaluated object is weak, and therefore it should be given a smaller weight. This method directly utilizes the information contained in each indicator to calculate the weight of the indicator, thus possessing objectivity. Based on the time series, a three-dimensional change map between fluctuation deviation and ratio deviation is established. The fluctuation deviation is plotted as the X-axis, the ratio deviation as the Y-axis, and the time series as the Z-axis. A three-dimensional change map is generated through fitting, and then overlaid and compared with a standard three-dimensional change map to obtain corresponding horizontal and vertical screenshots. The overlapping and non-overlapping total areas are then compared to generate the total overlapping area and the total non-overlapping area. It should be noted that the non-overlapping total area represents the severity of environmental disturbance to the ecosystem. A larger non-overlapping total area indicates a higher value of the corresponding compensation factor, indicating a more severe degree of external disturbance or degradation to the current ecosystem. The ratio of the non-overlapping total area to the sum of the overlapping and non-overlapping total areas is marked as the compensation factor. The first succession curve f(t) received from all areas to be restored is retrieved, where t is time and f(t) is the corresponding reflectance value. A predetermined succession period is set, and compensation adjustments are performed based on the compensation factor, including proportional adjustments and trajectory adjustments: Execution of proportional correction: The unit value 1 is subtracted from the compensation factor (1 minus the compensation factor), and the difference is defined as the correction coefficient. A product compression operation is then performed on the vertical axis of the first succession curve. This involves extracting the reflection value corresponding to each time point of the first succession curve, multiplying it by the correction coefficient, and replacing the original reflection value with the corrected feature value obtained from the multiplication. This proportionally compresses the entire first succession curve downwards in the numerical dimension, achieving proportional scaling of the trajectory. Execution trajectory correction: Multiply the duration of the succession time period by the compensation factor, define the product as the phase lag, and perform an additive offset operation on the time axis coordinate of the first succession curve. That is, add the phase lag to each original sampling time point on the first succession curve, thereby shifting the entire first succession curve backward on the time axis to achieve positive trajectory displacement. The curves after proportional and trajectory corrections are then integrated to obtain the final second succession curve.

[0022] The significance of the above analysis lies in the following: by combining the rule engine of the ecological benchmark area, the corresponding succession coefficient is determined by the ratio of volatility to standard volatility threshold, and a first succession curve is constructed; by loading environmental data to perform parameter compensation on the first succession curve, a second succession curve adapted to multiple areas to be restored is generated, thereby realizing the overall management of ecological benchmarks, restoration, and environmental elements.

[0023] S2: Based on the second succession curve, extract trend features, combine multispectral image data, set up a multi-condition encoder, determine the initial repair state, and establish multiple succession change partitions in each area to be repaired according to the time series; A multi-condition encoder includes a timing encoder, a visual encoder, and a decoding unit; The amplitude and succession rate of the second succession curve are extracted using a time-series encoder as trend features. Semantic and texture features are extracted by performing convolution operations on multispectral image data using a visual encoder. The feature vectors corresponding to trend features, semantic features, and texture features, as well as the attention weights that have a mapping relationship among the three, are mapped to a unified space. Feature fusion is performed on the trend features, semantic features, and texture features to obtain a joint feature tensor. The joint feature tensor is calculated by the decoding unit, including linear transformation and activation operation, and the initial repair state is output. The value of the initial repair state is represented by a probability value. Multiple successional change partitions are established for each region to be repaired based on the time series, including: Based on the initial restoration state, the collection cycle timestamps and sampling frequency ranges of the ecological monitoring points are obtained, and the minimum sampling interval is determined based on the sampling frequency range; at the same time, the standard time granularity is extracted based on the second succession curve. Based on the initial repair state, a dynamic mapping relationship is established between the minimum sampling interval and the standard time granularity to calculate the dynamic partition step size. Specifically, this includes: performing a difference operation on the standard time granularity and the minimum physical sampling interval to obtain the sampling elasticity interval; performing a product operation on the value of the initial repair state and the sampling elasticity interval to obtain the dynamic time increment; and performing a summation operation on the minimum physical sampling interval and the dynamic time increment to obtain the dynamic partition step size. Using the timestamp corresponding to the initial repair state as the starting anchor point, the acquisition cycle is discretized according to the dynamic partition step size to obtain multiple succession change partitions.

[0024] The significance of the above analysis lies in the following: by setting up a multi-condition encoder, the semantic features of the multispectral image are coupled with the trend features of the second succession curve to output a numerical initial restoration state, thus solving the problem of incomplete representation of single-modal data; at the same time, based on the quality of the initial restoration state, the acquisition cycle is dynamically adjusted and multiple succession change partitions are divided. This adaptive partitioning mechanism effectively balances monitoring accuracy and computing resources, reduces the missed detection rate or wasted computing power caused by improper sampling intervals, and significantly improves the timeliness of ecological restoration assessment.

[0025] S3: Based on the succession change partition, and synchronized to the preset ecological association topology network output spatiotemporal index table, a regression model is introduced to analyze the succession process of the second succession curve and predict the next succession change partition where ecological degradation occurs. Based on successional change partitioning, and synchronized to a pre-defined ecological association topology network, a spatiotemporal index table is output, including: A pre-defined ecological association topology network is retrieved, and topology nodes are extracted, including root nodes, intermediate cascade nodes, and monitoring terminal data acquisition nodes. Simultaneously, each successional change partition is mapped to the temporal attributes of the topology nodes. All topological nodes in the synchronized ecological association topological network are traversed to obtain the node attributes and connection relationships of each topological node. Based on the connection relationships between nodes, the succession change partitions of all nodes that are connected to the topological nodes in the succession change partition are determined, forming a set of associated succession change partitions. The set of node attributes, the number of topological levels, and the set of associated succession change partitions are integrated into a spatiotemporal index table.

[0026] A regression model is introduced to analyze the evolution process of the second succession curve, including: Within a preset time period, the first succession curve and the second succession curve of the area to be repaired are obtained; the deviation trajectory of the second succession curve relative to the first succession curve is calculated: the first succession curve and the second succession curve are aligned in time, and at the same time point, the vertical axis index of the second succession curve and the first succession curve is subtracted point by point, and the difference of all time points is statistically analyzed to form the deviation trajectory. The positive and negative slopes of the deviation trajectories are identified and grouped into positive and negative succession sets, respectively. The corresponding mean and standard deviation are extracted, and multiple succession monitoring vectors are constructed through different combinations, including: [μ1+μ2, σ1+σ2], [μ1+μ2, σ1-σ2], [μ1-μ2, σ1+σ2], and [μ1-μ2, σ1-σ2]; where μ1 and μ2 are the mean values ​​of the positive and negative succession sets, respectively, and σ1 and σ2 are the standard deviations of the positive and negative succession sets, respectively. The regression model is used to establish the correlation expression between each succession monitoring vector and the compensation factor. By taking the positive succession set, the negative succession set and multiple succession monitoring vectors as independent variables of the regression model and the compensation factor as the dependent variable, the two are aligned according to the time series to generate the sample dataset required for iterative training of the regression model. In the regression model, the multinomial regression equation is used as the basic model. Multiple candidate model structures of different orders are also pre-defined, including first-order, second-order and higher-order patterns, to form the regression hypothesis space. The purpose is to consider the linear relationship of the independent variables and to introduce higher-order power terms of the independent variables, such as quadratic terms, cubic terms, and interactive product terms between different features. The least squares method is used as the parameter estimation strategy, specifically including: calculating the sum of squared residuals between the predicted value output by each candidate model and the actual compensation factor; adjusting the weight coefficients and intercept terms in the polynomial regression equation through an iterative optimization algorithm until the sum of squared residuals reaches its minimum value, thereby obtaining the regression coefficient set with the best fit; substituting the optimized regression coefficient set into the corresponding polynomial regression equation to generate a mathematical expression describing the quantitative mapping relationship between the succession monitoring vector and the compensation factor, i.e., the correlation expression, including: the first correlation expression between [μ1+μ2, σ1+σ2] and the compensation factor, the second correlation expression between [μ1+μ2, σ1-σ2] and the compensation factor, the third correlation expression between [μ1-μ2, σ1+σ2] and the compensation factor, and the fourth correlation expression between [μ1-μ2, σ1-σ2] and the compensation factor; it should be noted that during the regression model calculation process, the succession monitoring vector and the compensation factor involved need to be Z-score standardized to eliminate the influence of dimensions; Extract the corresponding coefficient of determination R based on multiple correlation expressions. 2 The coefficient of determination (R²) is a numerical characteristic used to reflect the relationship between the compensation factor and multiple successional monitoring vectors. It is a statistical indicator used to reflect the reliability of the regression model in explaining changes in the dependent variable. A larger R² value indicates a greater goodness of fit, a higher degree of explanation by the independent variable for the dependent variable, and a higher percentage of the total change caused by the independent variable. 2The correlation expressions are sorted in descending order to determine the last one in the sort. Simultaneously, a minimum threshold for the coefficients is set. If the coefficients are less than the minimum threshold, it indicates that the compensation factor regulation has failed, ecological restoration is out of control, and severe degradation has occurred. Conversely, it indicates that the compensation factor regulation is effective. Feature importance analysis is performed on the correlation expressions of the area to be restored to determine the contribution of each feature to the compensation factor. Each contribution corresponds to a specific combination of succession monitoring vectors. It should be noted that the feature importance analysis process is a conventional technique and will not be elaborated upon here. Simultaneously, in cases of severe degradation, image data and topological location of the current area to be repaired are extracted, matched with information in the corresponding spatiotemporal index table, and a unique spatial index and corresponding temporal index of the current successional change partition are determined, forming a complete spatiotemporal identifier for the currently severely degraded area to be repaired; adjacent partitions with associated succession to the current spatial index are retrieved as candidate spatial partitions; candidate time nodes are retrieved one degradation diffusion cycle following the current time index; wherein, in the degradation diffusion cycle, R 2 If the value is less than the minimum threshold, the intersection of the candidate spatial partition and the candidate time node is the intersection point. The successional change partition corresponding to the intersection point is locked. Combined with the spatiotemporal index table, the successional change partition where the intersection point is located is retrieved and marked as the next successional change partition to experience ecological degradation.

[0027] The significance of the above analysis lies in the fact that by combining compensation factors and using regression models to establish multiple correlation expressions between each succession monitoring vector and the compensation factors, situations of severe ecological degradation can be screened out. This approach ensures that the regression model is trained and predicted based on real ecological succession and environmental response patterns, thereby significantly improving the accuracy and robustness of predicting the next ecological degradation succession zone.

[0028] Example 2: This invention provides a regional ecological restoration status assessment system based on image recognition; Figure 2 This is a schematic diagram of a regionalized ecological restoration status assessment system according to an exemplary embodiment; please refer to... Figure 2 The system includes: a region determination module, a state assessment module, and a degradation prediction module, and the region determination module, the state assessment module, and the degradation prediction module are connected in communication. The region determination module identifies the ecological baseline area, multiple areas to be restored, and multiple ecological monitoring points. Specifically, a rule engine is built within the ecological baseline area to generate the first succession curve. The first succession curve is then distributed to multiple areas to be restored to perform cascade calibration. Environmental data is loaded to perform parameter compensation on the first succession curve, thereby generating the second succession curve. State assessment module: Based on the second succession curve, trend features are extracted, multispectral image data is combined, a multi-condition encoder is set to determine the initial repair state, and multiple succession change partitions are established in each area to be repaired according to the time series. Degradation prediction module: Based on succession change partitions and synchronized to the preset ecological association topology network output spatiotemporal index table, a regression model is introduced to analyze the succession process of the second succession curve and predict the next succession change partition where ecological degradation will occur.

[0029] In the application, the multiple formulas mentioned are all calculated by removing dimensions and taking their numerical values. The formula is a formula obtained by software simulation based on a large amount of data to obtain the most recent real situation. The formula is set by a person skilled in the art according to the actual situation.

[0030] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0031] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0032] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A regionalized ecological restoration status assessment method based on image recognition, characterized in that, The method includes: identifying ecological baseline areas, multiple areas to be restored, and multiple ecological monitoring points; Specifically, a rule engine is built within the ecological baseline area to generate the first succession curve. The first succession curve is then distributed to multiple areas to be restored to perform cascade calibration. Environmental data is loaded to perform parameter compensation on the first succession curve, thereby generating the second succession curve. Based on the second succession curve, trend features are extracted, and multispectral image data is combined to set up a multi-condition encoder to determine the initial repair state. Multiple succession change partitions are established in each area to be repaired according to the time series. Based on the succession change partitions and synchronized to the pre-defined ecological association topology network output spatiotemporal index table, a regression model is introduced to analyze the succession process of the second succession curve and predict the next succession change partitions where ecological degradation will occur.

2. The regional ecological restoration status assessment method based on image recognition according to claim 1, characterized in that: The ecological baseline area, multiple areas to be restored, and multiple ecological monitoring points were identified, including: Upon receiving a monitoring command, the system acquires the geographical location, spectral code, and sampling frequency of the monitoring terminal, and generates a unique hash value. The monitoring terminal is equipped with a multispectral imaging sensor, a soil sensor, and a meteorological sensor. Meanwhile, node attributes are determined at the topological hierarchy in the preset ecological association topology network; among them, the root node is the ecological baseline area, the intermediate cascade nodes are the areas to be restored, and the monitoring terminal acquisition nodes are the ecological monitoring points, which are used to collect spectral image data and environmental data; among them, the spectral image data includes image data of topography, soil, and vegetation cover, and the environmental data includes meteorological parameters and geological parameters.

3. The regional ecological restoration status assessment method based on image recognition according to claim 2, characterized in that: Build a rules engine to generate the first succession curve, including: Image data within the ecological baseline area were acquired, and image recognition algorithms were used to extract the normalized vegetation index and reflectance values ​​at N spectral band positions. Volatility is calculated based on the normalized vegetation index, and the volatility is compared with the standard volatility threshold to determine the succession steady state level. For the N spectral band positions, perform iterative cross-sparsening to divide the spectral bands into Class I bands and Class II bands; where Class I bands represent the odd-numbered positions of the spectral bands arranged in order of their center wavelengths, and Class II bands represent the even-numbered positions of the spectral bands arranged in order of their center wavelengths. Reconstruction processing is performed on Class I and Class II bands to screen spectral bands that meet the corresponding succession steady-state level, including core bands and related bands; the reflection gradient difference between related bands and corresponding core bands is calculated to determine the succession coefficients of each spectral band position; A blank ecological succession trajectory flow is preset, and the reflection value of the core band is weighted and calibrated based on the succession coefficient to obtain the first succession curve.

4. The regional ecological restoration status assessment method based on image recognition according to claim 3, characterized in that: By comparing volatility with the standard volatility threshold, the succession steady-state level is determined, including: If the volatility is less than or equal to the standard volatility threshold, mark the current volatility as 'a' and edit it as a first-level character. Combine 'a' and the first-level character to generate a high steady-state level. If the volatility is greater than the standard volatility threshold, the current volatility is marked as 'b' and edited as a secondary character. The 'b' and the secondary character are then combined to generate a low steady-state level.

5. The regional ecological restoration status assessment method based on image recognition according to claim 3, characterized in that: Load environmental data to perform parameter compensation on the first succession curve, including: Meteorological and geological parameters were retrieved from multiple areas to be repaired. The meteorological parameters included atmospheric conductivity, ambient temperature and humidity, and solar radiation intensity, while the geological parameters included soil salinity, soil moisture content, and soil pH. Establish the first correlation between meteorological parameters and the first successional curve, establish the second correlation between geological parameters and the first successional curve, and construct a compensation factor generation model based on the first and second correlations. The compensation factors for each area to be repaired are generated based on the compensation factor generation model, including: determining the fluctuation deviation of meteorological parameters and determining the ratio deviation of geological parameters; Based on the time series, a three-dimensional change graph between fluctuation deviation and ratio deviation is established and overlaid with a standard three-dimensional change graph to obtain the corresponding horizontal and vertical screenshots. The overlapping total area and non-overlapping total area are then compared to generate the graph. The ratio of the non-overlapping total area to the sum of the overlapping and non-overlapping total areas is marked as a compensation factor. The first succession curve is then compensated and corrected to obtain the second succession curve.

6. The regional ecological restoration status assessment method based on image recognition according to claim 5, characterized in that: Configure a multi-condition encoder to determine the initial repair state, including: A multi-condition encoder includes a timing encoder, a visual encoder, and a decoding unit; The amplitude and succession rate of the second succession curve are extracted using a time-series encoder as trend features. Semantic and texture features are extracted by performing convolution operations on multispectral image data using a visual encoder. Feature fusion is performed on trend features, semantic features, and texture features to obtain a joint feature tensor. The joint feature tensor is then solved by the decoding unit to output the initial repair state. The value of the initial repair state is represented by a probability value.

7. The regional ecological restoration status assessment method based on image recognition according to claim 6, characterized in that: Multiple successional change partitions are established for each region to be repaired based on the time series, including: Based on the initial restoration state, obtain the collection cycle timestamps and sampling frequency ranges of the ecological monitoring points, and determine the minimum sampling interval based on the sampling frequency range; Standard time granularity is extracted based on the second succession curve; The dynamic partitioning step size is determined based on the minimum sampling interval and the standard time granularity. The timestamp corresponding to the initial repair state is used as the starting anchor point. The acquisition cycle is discretized according to the dynamic partitioning step size to obtain multiple successive change partitions.

8. The regional ecological restoration status assessment method based on image recognition according to claim 7, characterized in that: Map each successional change partition to a time attribute of a topological node.

9. The regional ecological restoration status assessment method based on image recognition according to claim 8, characterized in that: The analysis of the succession process of the second succession curve includes: Calculate the deviation trajectory of the second succession curve relative to the first succession curve, identify the positive and negative slopes of the deviation trajectory, and extract the mean and standard deviation to construct multiple succession monitoring vectors; The current compensation factor is retrieved, and a regression model is used to establish multiple correlation expressions between each succession monitoring vector and the compensation factor. The coefficients of determination of the correlation expressions are extracted, and the cases of severe degradation are screened by sorting in descending order. Combined with the spatiotemporal index table, the next succession change partition to experience severe ecological degradation is predicted. The spatiotemporal index table includes a set of node attributes, a number of topological levels, and a set of associated succession change partitions.

10. A regionalized ecological restoration status assessment system based on image recognition, characterized in that, The system includes: a region determination module: determining the ecological baseline area, multiple areas to be restored, and multiple ecological monitoring points; Specifically, a rule engine is built within the ecological baseline area to generate the first succession curve. The first succession curve is then distributed to multiple areas to be restored to perform cascade calibration. Environmental data is loaded to perform parameter compensation on the first succession curve, thereby generating the second succession curve. State assessment module: Based on the second succession curve, trend features are extracted, multispectral image data is combined, a multi-condition encoder is set to determine the initial repair state, and multiple succession change partitions are established in each area to be repaired according to the time series. Degradation prediction module: Based on succession change partitions and synchronized to the preset ecological association topology network output spatiotemporal index table, a regression model is introduced to analyze the succession process of the second succession curve and predict the next succession change partition where ecological degradation will occur.

Images