Epimedium koreanum planting area identification method based on multi-source remote sensing image fusion

Abstract

The invention discloses a method for identifying an epimedium koreanum planting area based on multi-source remote sensing image fusion in the technical field of under-forest planting, and the method comprises the steps: obtaining multi-source remote sensing data of a target area, carrying out the preprocessing, and carrying out the recognition of an epimedium koreanum planting area based on the preprocessed multi-source remote sensing data; extracting multi-source feature data related to under-forest planting of the Korean epimedium, fusing the extracted multi-source features, inputting the fused multi-source features into a deep neural network classification model of a cross-modal attention mechanism, processing the fused multi-source features, and obtaining a deep neural network classification model of the cross-modal attention mechanism; according to the method, multi-source remote sensing data advantages of optics, radars, laser radars and the like are integrated, the limitation of a single data source in under-forest environment monitoring is overcome, and the method has the advantages of being high in accuracy, high in accuracy and high in reliability. And aiming at special growth requirements of the epimedium koreanum, key parameters such as canopy density and gradient are quantified, and accurate evaluation of ecological suitability is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of under-forest planting, and in particular to a method for identifying an epimedium planting area based on multi-source remote sensing image fusion. Background Art

[0002] Traditionally, the selection of understory planting areas relies primarily on manual field surveys, which are inefficient, costly, and subjective. Remote sensing imagery alone makes it difficult to accurately identify areas suitable for growing Epimedium koreana.

[0003] Currently, the application of remote sensing technology to identifying areas suitable for understory crop growth still faces numerous challenges: Relying on single remote sensing imagery yields poor identification results; spatial heterogeneity is significant under complex terrain conditions; and there is a lack of systematic methods for comprehensively assessing multiple growth environment parameters. Existing technologies, such as patent CN202410486135.7, define the ecological requirements for understory planting, providing a basis for target parameters in regional identification systems, but do not involve remote sensing technology. Patent CN118675059A proposes a method for identifying fruit tree planting areas based on multi-source remote sensing data, but this method fails to consider the unique understory environment and the ecological needs of medicinal plants. Patent CN119129388A addresses soil moisture estimation but does not address the issue of identifying areas suitable for understory crop cultivation. Therefore, a multi-source remote sensing fusion method specifically tailored for identifying Korean goat weed planting areas in understory crops is urgently needed. This method, based on optimized fused data, could provide a scientific basis for selecting planting sites for high-value medicinal plants like Korean goat weed. Summary of the Invention

[0004] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid blurring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0005] Therefore, the purpose of the present invention is to provide a method for identifying areas where Korean epimedium is planted based on multi-source remote sensing image fusion, so as to solve the problems raised in the above background technology.

[0006] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions: A method for identifying areas where Korean epimedium is planted based on multi-source remote sensing image fusion, the steps are as follows: S1. Obtain multi-source remote sensing data of the target area and perform preprocessing; S2. Extract multi-source feature data related to the understory planting of Epimedium koreanum based on the pre-processed multi-source remote sensing data; S3. Fusing the extracted multi-source features and inputting the fused multi-source features into a deep neural network classification model with a cross-modal attention mechanism to process the fused multi-source features and identify suitable planting areas. S4. Verify and optimize the identification results through field investigation; S5. Classify the suitable planting areas and output them visually.

[0007] As a preferred solution of the method for identifying Korean epimedium planting areas based on multi-source remote sensing image fusion described in the present invention, in step S1, the multi-source remote sensing data includes: high-resolution optical remote sensing images, synthetic aperture radar, lidar point cloud data and terrain data.

[0008] As a preferred solution of the method for identifying Korean epimedium planting areas based on multi-source remote sensing image fusion described in the present invention, in step S1, preprocessing of multi-source remote sensing data includes: radiation correction, atmospheric correction, geometric correction, image registration and noise removal.

[0009] As a preferred solution of the method for identifying Korean epimedium planting areas based on multi-source remote sensing image fusion described in the present invention, in step S2, the multi-source feature data includes canopy density, slope, soil properties and vegetation coverage.

[0010] As a preferred solution of the method for identifying Korean Epimedium planting areas based on multi-source remote sensing image fusion described in the present invention, the canopy density information is extracted and calculated as follows: Based on high-resolution optical remote sensing images, vegetation index or pixel decomposition methods are used to extract vegetation cover information. Combined with vegetation structure information obtained from radar remote sensing images, a canopy density inversion model is established to calculate the canopy density of the target area. Furthermore, a canopy void ratio analysis method based on a three-dimensional point cloud is used to calculate canopy density, combining LiDAR point cloud data and high-resolution optical images. The formula is: canopy density = 1 - (number of penetrating light pixels / total number of pixels). The suitable range of canopy density is set at 0.4-0.8, and Korean epimedium is graded according to different growth stages: the suitable canopy density for 1-2 year old plants is 0.6-0.7, and the suitable canopy density for plants over 3 years old is 0.4-0.5.

[0011] As a preferred solution of the method for identifying Korean epimedium planting areas based on multi-source remote sensing image fusion described in the present invention, the slope information is extracted and calculated as follows: Based on the DEM data, the neighborhood analysis method is used to calculate the slope of the target area, and the area with a slope between 5° and 25° is screened. The formula is: ; ; in, fx、fy are the slope change rates in the x and y directions respectively; It is the weighted sum of column indices j-1, j, and j+1 in row (i+1), reflecting the "overall trend" of the neighborhood in the next row; : It is the weighted sum of the column indices j-1, j, and j+1 in row (i-1), reflecting the "overall trend" of the neighborhood in the previous row; : It is the weighted sum of the row indices i-1, i, and i+1 in the (j+1) column, reflecting the "overall trend" of the neighborhood in the right column; : It is the weighted sum of the row indices i-1, i, and i+1 in the (j-1) column, reflecting the "overall trend" of the neighborhood in the left column; 8D is the "normalization coefficient" of discrete difference, and Slope represents the slope.

[0012] As a preferred solution of the method for identifying Korean Epimedium planting areas based on multi-source remote sensing image fusion described in the present invention, soil characteristics are extracted and calculated as follows: Sentinel-2 multispectral data was used to calculate the Soil Adjusted Vegetation Index and Normalized Difference Water Index, which indirectly assessed soil organic matter content and moisture. The formula is: ; ; Among them: NIR is the near infrared band reflectance, RED is the red light band reflectance, L is the soil adjustment coefficient, ρ Green is the reflectivity of the green light band, ρ NIR is the reflectance in the near-infrared band; The soil moisture content inverted by microwave remote sensing data is in the range of 20%-30%; The soil pH value was estimated based on the visible-near-infrared spectral characteristics, and the pH range of 5.5-7.0 was screened.

[0013] As a preferred solution of the method for identifying areas for planting Epimedium koreana based on multi-source remote sensing image fusion described in the present invention, the vegetation coverage is extracted and calculated as follows: The mixed pixel decomposition method was used to quantitatively assess vegetation cover and identify areas of broad-leaved forests dominated by oak and linden trees or sparse coniferous and broad-leaved mixed forests. The calculation was performed based on information fed back by the data source. The formula is as follows: ; Among them, R is the spectral reflectance of the pixel, Rveg is the spectral reflectance of the pure vegetation pixel, Rsoil is the spectral reflectance of the pure non-vegetation pixel, and FVC is the vegetation cover.

[0014] As a preferred embodiment of the method for identifying areas suitable for planting Epimedium koreana based on multi-source remote sensing image fusion according to the present invention, in step S3, the extracted multi-source features are fused and the fused multi-source features are input into a deep neural network classification model with a cross-modal attention mechanism. The fused multi-source features are processed and the steps for identifying suitable planting areas are as follows: An independent encoder is designed for each modality, and the features of different modalities are projected into a unified latent space through a multi-layer perceptron (MLP), maximizing the similarity of matching modal pairs while minimizing the similarity of non-matching pairs. Attention weights are generated to guide the fusion of key features, ensuring that the distribution of fused features in spectral space, polarization space, and three-dimensional space is consistent, achieving optimal fusion of multi-source remote sensing image data. The extracted multi-source features are fused to generate a fused feature image containing multiple information. The multi-scale segmentation algorithm is used to objectify the fused image so that adjacent pixels with similar feature spaces are segmented into one object.

[0015] As a preferred solution of the method for identifying areas where Epimedium koreana is planted based on multi-source remote sensing image fusion according to the present invention, in step S4, the specific steps of verifying and optimizing the identification results through field investigation are as follows: Randomly select 50-100 sampling points within the identified suitable planting areas, covering different slope aspects, altitudes and forest types; Measure the slope, aspect, and soil moisture at the sample points and compare them with remote sensing inversion data. Use drones or handheld spectrometers to measure the understory light intensity and canopy density to verify the accuracy of remote sensing inversion canopy density. The distribution density and growth status of Epimedium koreana at the sampling sites were collected to assess planting suitability. The Kappa coefficient, overall accuracy, and F1-score of the remote sensing identification results and field data were calculated. Use random forest or XGBoost algorithm to evaluate the importance of each feature and eliminate redundant features; If the multi-source image registration error is large, super-resolution reconstruction or deep learning registration algorithms are introduced to improve spatial consistency; Perform sensitivity analysis on the thresholds in the canopy density model, dynamically adjust them based on field data, and use support vector machines or neural networks to adjust hyperparameters through grid search or Bayesian optimization.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. Integrate the advantages of multi-source remote sensing data such as optics, radar, and lidar to overcome the limitations of a single data source in forest environment monitoring and improve recognition accuracy.

[0017] 2. In view of the special growth requirements of Korean epimedium, key parameters such as density and slope are quantified to achieve accurate assessment of ecological suitability and avoid the subjectivity of traditional methods.

[0018] 3. Rapid screening of large forest areas can be achieved through machine learning algorithms, which is dozens of times more efficient than manual surveys and reduces survey costs.

[0019] 4. Multi-temporal data analysis can monitor environmental changes in the planting area and provide decision support for field management (such as shade adjustment). BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and detailed embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be derived from these drawings without inventive effort. Among them: Figure 1 This is a flow chart of a method for identifying areas where Epimedium koreana is planted based on multi-source remote sensing image fusion according to the present invention; Figure 2 This is a schematic diagram of the multi-source remote sensing data fusion provided by the present invention. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0022] The present invention provides a method for identifying areas suitable for planting Korean epimedium under forests based on multi-source remote sensing image fusion, such as Figure 1 As shown, the steps are as follows: Step 1: Multi-source remote sensing image data acquisition and preprocessing: 1) Acquire multiple remote sensing data sources for the target area, including: high-resolution optical remote sensing imagery (Sentinel-2, GF-7, with spatial resolution better than 10m), synthetic aperture radar (SAR) data (Sentinel-1, C-band), laser radar (LiDAR) point cloud data (airborne LiDAR), and terrain data (DEM, with accuracy better than 5m); 2) Preprocessing of the acquired multi-source remote sensing image data includes radiometric correction (using ENVI software), atmospheric correction (using the FLAASH model), geometric correction (using SARscape software), image registration, and noise removal to eliminate errors caused by sensor differences, terrain undulations, and atmospheric conditions.

[0023] Step 2: Extraction of understory environmental parameters: Based on the pre-processed multi-source remote sensing image data, key environmental parameters related to the understory planting of Epimedium koreanum were extracted, including: 1) Extraction and calculation of canopy density: Based on high-resolution optical remote sensing imagery, vegetation index or pixel decomposition methods are used to extract vegetation cover information. Combined with vegetation structure information obtained from radar remote sensing imagery, a canopy density inversion model is established to calculate the canopy density of the target area. LiDAR point cloud data and high-resolution optical imagery can also be combined to calculate canopy density using a three-dimensional point cloud-based canopy void ratio analysis method. The formula is: canopy density = 1 - (number of penetrating light pixels / total number of pixels). The suitable range of canopy density is set at 0.4-0.8, and Korean epimedium is graded according to different growth stages: the suitable canopy density for 1-2 year old plants is 0.6-0.7, and the suitable canopy density for plants over 3 years old is 0.4-0.5.

[0024] 2) Slope information extraction and analysis: Based on DEM data, the neighborhood analysis method is used to calculate the slope of the target area. Areas with slopes between 5° and 25° are selected, with priority given to east and west sunny slopes. The formula is: ; ; in, fx、fy are the slope change rates in the x and y directions respectively; It is the weighted sum of column indices j-1, j, and j+1 in row (i+1), reflecting the "overall trend" of the neighborhood in the next row; : It is the weighted sum of the column indices j-1, j, and j+1 in row (i-1), reflecting the "overall trend" of the neighborhood in the previous row; : It is the weighted sum of the row indices i-1, i, and i+1 in the (j+1) column, reflecting the "overall trend" of the neighborhood in the right column; : It is the weighted sum of the row indices i-1, i, and i+1 in the (j-1) column, reflecting the "overall trend" of the neighborhood in the left column; 8D is the "normalization coefficient" of discrete difference, and Slope represents the slope.

[0025] 3) Soil property assessment: The Soil Adjusted Vegetation Index (SAVI) and Normalized Difference Water Index (NDWI) are calculated using Sentinel-2 multispectral data to indirectly assess soil organic matter content and moisture. The formula is: ; ; Among them, NIR is the near-infrared band reflectance, RED is the red light band reflectance, and L is the soil adjustment coefficient (the value range is 0~1, usually set according to soil brightness experience, such as the default value of 0.5). ρ Green is the reflectance of the green light band (usually corresponding to the wavelength of about 500–600 nm in remote sensing images), ρ NIR is the reflectance in the near-infrared band (usually corresponding to the wavelength band of approximately 700–1100 nm); The soil moisture content was inverted using microwave remote sensing (SAR) data, and the suitable range was 20%-30%; The soil pH value was estimated based on the visible-near-infrared spectral characteristics, and the pH range of 5.5-7.0 was screened.

[0026] 4) Vegetation Cover Analysis: The composite pixel decomposition (FVC) method was used to quantitatively assess vegetation cover and identify areas of broad-leaved forests dominated by oak and linden trees or sparse coniferous and broad-leaved mixed forests. The calculation was performed based on information fed back by the data source. The formula is as follows: ; Where R is the spectral reflectance of the pixel, R veg is the spectral reflectance of pure vegetation pixels, R soil is the spectral reflectance of pure non-vegetated (bare soil) pixels, and FVC is the vegetation cover (value range: 0~1).

[0027] By solving the above equation, we can get the expression of FVC: .

[0028] Step 3: Multi-source feature fusion and identification of suitable planting areas 1) Feature layer fusion: Using the Cross-Modal Contrastive Alignment Network fusion method, the pre-processed optical remote sensing images, radar remote sensing images, and LiDAR remote sensing images are optimized and fused, retaining the modality-specific information while ensuring that the features are forced to align in the shared space to obtain the optimal fused image. Figure 2As shown in the figure, the specific steps are: design an independent encoder for each modality; project the features of different modalities into a unified latent space through a multi-layer perceptron (MLP); maximize the similarity of matching modal pairs while minimizing the similarity of non-matching pairs; generate attention weights to guide the fusion of key features; make the distribution of fusion features consistent in spectral space, polarization space, and three-dimensional space to avoid the "feature tearing" problem of traditional fusion and achieve optimal fusion of multi-source remote sensing image data; fuse the extracted features (canopy density, slope, soil moisture, vegetation index, etc.) to generate a fused feature image containing multiple information.

[0029] 2) Object-oriented segmentation: A multi-scale segmentation algorithm (eCognition's multi-resolution segmentation) is used to objectify the fused image, so that adjacent pixels with similar feature spaces are segmented into one object, thereby reducing the "salt and pepper effect" and improving recognition accuracy.

[0030] 3) Machine Learning Classification: 31) A deep neural network classification model that incorporates a cross-modal attention mechanism is constructed to automatically learn the importance of features from different modalities and suppress redundant information. 32) Input features include: 33) object-level canopy density, slope, soil moisture, vegetation type, altitude, etc. 34) The output is a binary classification result, suitable for planting areas and unsuitable for planting areas. 35) Sub-classification is performed for different tree categories (such as supporting trees).

[0031] Step 4: Result Verification and Optimization 1) Field sampling verification Objective: To verify the reliability of remote sensing identification results through field investigation.

[0032] Methods: Sample plot layout: 50-100 sampling points were randomly selected within the identified suitable planting areas (e.g., slope 15°-30°, canopy density 0.4-0.7), covering different slope aspects, altitudes and stand types.

[0033] Environmental parameter collection: measured slope, aspect, soil moisture, etc. at sample points, and compared with remote sensing inversion data.

[0034] Vegetation parameter collection: Measure the light intensity and canopy density under the forest using drones or handheld spectrometers to verify the accuracy of remote sensing inversion of canopy density.

[0035] Biomass verification: Collect the distribution density and growth status of Epimedium koreana (or other indicator plants) in the sampling points to assess the suitability for planting.

[0036] Accuracy evaluation: Calculate the Kappa coefficient, overall accuracy (OA), and F1-score of the remote sensing recognition results and the field data, requiring OA ≥ 80%.

[0037] 2) Model optimization and parameter adjustment Objective: Improve the algorithm model based on the verification results to enhance adaptability and recognition accuracy.

[0038] Feature selection optimization: Random forest (RF) or XGBoost algorithm is used to evaluate the importance of each feature (such as NDVI, slope, and texture features) and eliminate redundant features.

[0039] Improvement of fusion algorithms: If large errors are found in multi-source image registration, super-resolution reconstruction or deep learning registration algorithms (such as SIFT+CNN) can be introduced to improve spatial consistency.

[0040] Conduct sensitivity analysis on the threshold values ​​(e.g. 0.4-0.7) in the canopy density model and dynamically adjust them based on field data.

[0041] Classifier parameter tuning: If using a support vector machine (SVM) or neural network, tune hyperparameters (such as kernel function and learning rate) through grid search or Bayesian optimization.

[0042] Step 5: Planting suitability classification and visualization The identification results are divided into three levels according to the matching degree of environmental parameters: highly suitable area, generally suitable area and unsuitable area; Highly suitable area: fully meets all the growth conditions of Korean epimedium (canopy density 0.5-0.7, slope 10-20°, soil pH 6.0-6.8, etc.); Generally suitable area: meets the main conditions but requires slight improvement (for example, the density is slightly high and thinning is required); Unsuitable area: key conditions are not met (canopy density <0.3 or >0.9, slope >35°).

[0043] In order to verify the effect of the above method of the present invention, the following example is provided for verification: Identification of the suitable area for growing Epimedium koreanum in Changbai Mountain area of ​​Jilin Province 1. Data acquisition: Sentinel-2 MSI imagery (10m resolution, 2024 growing season); Sentinel-1 SAR data (5×20m resolution, multi-temporal for 2024 years); Airborne LiDAR data (point density ≥ 4 points / m 2 ); 30m resolution DEM data; Historical planting records (50 sample points); 2. Preprocessing: Atmospheric correction (Sen2Cor) and cloud mask generation for Sentinel-2; Perform thermal noise removal, radiometric calibration, and terrain correction on Sentinel-1; LiDAR point cloud classification (ground / non-ground) and generation of canopy height model (CHM).

[0044] 3. Key parameter extraction: Canopy density: Calculated based on the penetration rate of LiDAR point cloud, the formula is: C = 1 - (Nground / Ntotal), Where Nground is the number of ground points, and Ntotal is the total number of points; Slope: calculated based on DEM using a 3×3 window, screening the 10°-25° area; Soil moisture: Fusion of Sentinel-1 backscatter coefficient and Sentinel-2 NDWI index; Vegetation type: Identify dominant oak and linden communities based on Sentinel-2 time series data.

[0045] 4. Classification model construction: Select the random forest algorithm and set 100 decision trees; Input features include: canopy density, slope, aspect, soil moisture, vegetation type, and elevation; 300 training samples (including historical planting sites and newly added sampling sites); The model performance was evaluated using 10-fold cross validation.

[0046] 5. Results and Analysis A highly suitable area of ​​1,200 hectares was identified, accounting for 9% of the study area; It is mainly distributed on northeast slopes at an altitude of 600-900m and a canopy density of 0.5-0.6; The verification accuracy reached 87.2% (Kappa=0.81), and compared with field surveys, the survey cost was saved by about 75%.

[0047] Although the present invention has been described above with reference to embodiments, various modifications may be made thereto and equivalent components may be substituted without departing from the scope of the present invention. In particular, as long as there are no structural conflicts, the various features of the embodiments disclosed herein may be combined with each other in any manner, and the omission of an exhaustive description of such combinations in this specification is solely for the sake of space and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for identifying areas where Epimedium koreana is planted based on multi-source remote sensing image fusion, characterized in that: Here are the steps: S1. Obtain multi-source remote sensing data of the target area and perform preprocessing; S2. Extract multi-source feature data related to the understory planting of Epimedium koreanum based on the pre-processed multi-source remote sensing data; S3. Fusing the extracted multi-source features and inputting the fused multi-source features into a deep neural network classification model with a cross-modal attention mechanism to process the fused multi-source features and identify suitable planting areas. S4. Verify and optimize the identification results through field investigation; S5. Classify the suitable planting areas and output them visually.

2. The method for identifying areas where Epimedium koreana is planted based on multi-source remote sensing image fusion according to claim 1, wherein: In step S1, the multi-source remote sensing data includes: high-resolution optical remote sensing images, synthetic aperture radar (SAR), lidar point cloud data, and terrain data.

3. The method for identifying areas where Epimedium koreana is planted based on multi-source remote sensing image fusion according to claim 1, wherein: In step S1, the preprocessing of multi-source remote sensing data includes: radiation correction, atmospheric correction, geometric correction, image registration and noise removal.

4. The method for identifying areas where Epimedium koreana is planted based on multi-source remote sensing image fusion according to claim 1, wherein: In step S2, the multi-source feature data includes canopy density, slope, soil properties and vegetation coverage.

5. The method for identifying areas where Epimedium koreana is planted based on multi-source remote sensing image fusion according to claim 4, wherein: The canopy density information is extracted and calculated as follows: Based on high-resolution optical remote sensing images, vegetation index or pixel decomposition methods are used to extract vegetation cover information. Combined with vegetation structure information obtained from radar remote sensing images, a canopy density inversion model is established to calculate the canopy density of the target area. Furthermore, a canopy void ratio analysis method based on a three-dimensional point cloud is used to calculate canopy density, combining LiDAR point cloud data and high-resolution optical images. The formula is: canopy density = 1 - (number of penetrating light pixels / total number of pixels). The suitable range of canopy density is set at 0.4-0.8, and Korean epimedium is graded according to different growth stages: the suitable canopy density for 1-2 year old plants is 0.6-0.7, and the suitable canopy density for plants over 3 years old is 0.4-0.

5.

6. The method for identifying areas where Epimedium koreana is planted based on multi-source remote sensing image fusion according to claim 4, wherein: Slope information extraction and calculation are as follows: Based on the DEM data, the neighborhood analysis method is used to calculate the slope of the target area, and the area with a slope between 5° and 25° is screened. The formula is: ； ； in, fx、fy are the slope change rates in the x and y directions respectively; It is the weighted sum of column indices j-1, j, and j+1 in row (i+1), reflecting the "overall trend" of the neighborhood in the next row. : It is the weighted sum of the column indices j-1, j, and j+1 in row (i-1), reflecting the "overall trend" of the neighborhood in the previous row; : It is the weighted sum of the row indices i-1, i, and i+1 in the (j+1) column, reflecting the "overall trend" of the neighborhood in the right column; : It is the weighted sum of the row indices i-1, i, and i+1 in the (j-1) column, reflecting the "overall trend" of the neighborhood in the left column; 8D is the "normalization coefficient" of discrete difference, and Slope represents the slope.

7. The method for identifying areas where Epimedium koreana is planted based on multi-source remote sensing image fusion according to claim 4, wherein: Soil properties are extracted and calculated as follows: Sentinel-2 multispectral data was used to calculate the Soil Adjusted Vegetation Index and Normalized Difference Water Index, which indirectly assessed soil organic matter content and moisture. The formula is: ； ； Among them: NIR is the near infrared band reflectance, RED is the red light band reflectance, L is the soil adjustment coefficient, ρ Green is the reflectivity of the green light band, ρ NIR is the reflectance in the near-infrared band; The soil moisture content inverted by microwave remote sensing data is in the suitable range of 20%-30%; The soil pH value was estimated based on the visible-near-infrared spectral characteristics, and the pH range of 5.5-7.0 was screened.

8. The method for identifying areas where Epimedium koreana is planted based on multi-source remote sensing image fusion according to claim 1, wherein: Vegetation coverage extraction and calculation are as follows: The mixed pixel decomposition method was used to quantitatively assess vegetation cover and identify areas of broad-leaved forests dominated by oak and linden trees or sparse coniferous and broad-leaved mixed forests. The calculation was performed based on information fed back by the data source. The formula is as follows: ； Among them, R is the spectral reflectance of the pixel, Rveg is the spectral reflectance of the pure vegetation pixel, Rsoil is the spectral reflectance of the pure non-vegetation pixel, and FVC is the vegetation cover.

9. The method for identifying areas where Epimedium koreana is planted based on multi-source remote sensing image fusion according to claim 1, wherein: In step S3, the extracted multi-source features are fused and input into a deep neural network classification model with a cross-modal attention mechanism. The steps for processing the fused multi-source features and identifying suitable planting areas are as follows: An independent encoder is designed for each modality, and the features of different modalities are projected into a unified latent space through a multi-layer perceptron, maximizing the similarity of matching modal pairs while minimizing the similarity of non-matching pairs. Attention weights are generated to guide the fusion of key features, ensuring that the distribution of fused features in spectral space, polarization space, and three-dimensional space is consistent, achieving optimal fusion of multi-source remote sensing image data. The extracted multi-source features are fused to generate a fused feature image containing multiple information. The multi-scale segmentation algorithm is used to objectify the fused image so that adjacent pixels with similar feature spaces are segmented into one object.

10. The method for identifying areas where Epimedium koreana is planted based on multi-source remote sensing image fusion according to claim 1, wherein: In step S4, the specific steps for verifying and optimizing the recognition results through field investigation are as follows: Randomly select 50-100 sampling points within the identified suitable planting areas, covering different slope aspects, altitudes and forest types; Measure the slope, aspect, and soil moisture at the sample points and compare them with remote sensing inversion data. Use drones or handheld spectrometers to measure the understory light intensity and canopy density to verify the accuracy of remote sensing inversion canopy density. The distribution density and growth status of Epimedium koreana at the sampling sites were collected to assess planting suitability. The Kappa coefficient, overall accuracy, and F1-score of the remote sensing identification results and field data were calculated. Use random forest or XGBoost algorithm to evaluate the importance of each feature and eliminate redundant features; If the multi-source image registration error is large, super-resolution reconstruction or deep learning registration algorithms are introduced to improve spatial consistency; Perform sensitivity analysis on the thresholds in the canopy density model, dynamically adjust them based on field data, and use support vector machines or neural networks to adjust hyperparameters through grid search or Bayesian optimization.

Images