Ant colony optimization fusion multi-source remote sensing data forest canopy height estimation method and device

Abstract

The application discloses an ant colony optimization fusion multi-source remote sensing data forest canopy height estimation method and device, and belongs to the technical field of forest resource remote sensing monitoring. The method firstly constructs a correction model of ICESat-2 laser radar data through an ant colony optimization algorithm, and calibrates the ICESat-2 data based on GEDI data; subsequently, the corrected ICESat-2 data and the GEDI data are fused, combined with multi-source auxiliary features such as Sentinel-2 optical images, terrain and climate, and an ant colony optimization algorithm is used to dynamically optimize super parameters, so that a canopy height prediction model integrated with XGBoost, RF and linear regression is constructed; finally, a spatially continuous forest canopy height map with a resolution of 30 meters is generated, and the accuracy is verified by using the GEDI data. Through intelligent optimization and multi-source data fusion, the application significantly improves the estimation consistency and reliability of the forest canopy height in a complex environment.

Description

Technical Field

[0001] This invention belongs to the field of forest resource remote sensing monitoring technology, specifically relating to a method and device for estimating forest canopy height by ant colony optimization and fusion of multi-source remote sensing data. Background Technology

[0002] Forest canopy height is a key parameter for estimating forest aboveground biomass, assessing carbon sequestration capacity, and monitoring dynamic changes in forest structure and function. While traditional ground-based measurement methods offer high accuracy, they are time-consuming and labor-intensive, making large-scale, continuous dynamic monitoring difficult. The emergence of spaceborne lidar technology has provided a revolutionary tool for observing the three-dimensional structure of forests at global and regional scales. The Global Ecosystem Dynamics Survey (GEDI) payload can acquire high-precision waveform data of forest vertical structure, but its observations are distributed in discrete stripes with discontinuous spatial coverage. The photon-counting lidar on the Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) provides denser global sampling, but its retrieved canopy height products have significant errors under complex terrain or closed canopy conditions, resulting in relatively insufficient data reliability. Currently, comprehensively utilizing the high precision of GEDI and the wide coverage of ICESat-2 to generate seamless, high-precision canopy height products represents a cutting-edge direction for remote sensing forestry applications.

[0003] Existing technologies mostly employ single or fixed machine learning models (such as RF and gradient boosting decision trees) to directly extrapolate the spatial height of GEDI or ICESat-2 data. However, these methods have significant limitations: First, model performance heavily relies on human experience for hyperparameter tuning, a cumbersome process that makes it difficult to obtain the globally optimal solution; second, the model structure is fixed, lacking the ability to adaptively optimize for the characteristics of multi-source heterogeneous data and the optimal modeling strategy under different regional and forest type conditions; third, they fail to systematically address the systematic biases and accuracy differences between ICESat-2 and GEDI data, and direct fusion introduces errors, limiting further improvement in the accuracy of the final product. Therefore, there is an urgent need for an intelligent estimation framework that can automatically optimize model structure and parameters, effectively fuse and correct multi-source lidar data, and integrate auxiliary information such as multispectral, topographic, and climatic data to achieve high-precision, highly spatially continuous forest canopy height mapping. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method and apparatus for estimating forest canopy height by fusing multi-source remote sensing data using ant colony optimization (ACO). The method employs the ant colony optimization algorithm (ACO) to integrate multi-source data such as ICESat-2, GEDI, Sentinel-2, topographic data, and climate data to construct a LiDAR data correction model and a canopy height prediction model, thereby generating continuous canopy height maps.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A forest canopy height estimation method based on ant colony optimization and fusion of multi-source remote sensing data, the method comprising:

[0007] Step 1: Multi-source remote sensing data preprocessing and feature extraction;

[0008] Step 2, Spaceborne LiDAR Data Fusion and Correction: Using GEDI canopy height data as a benchmark, an ant colony optimization algorithm is used to construct a correction model for ICESat-2 data, generating a corrected ICESat-2 canopy height dataset.

[0009] Step 3: Construct the optimal forest canopy height extrapolation model: The ICESat-2 canopy height dataset and the preprocessed GEDI canopy height data are fused together as training samples. The features extracted from the multi-source remote sensing data are combined, and the ant colony optimization algorithm is applied again to dynamically construct the optimal integrated prediction model.

[0010] Step 4: Forest canopy height prediction: Use the optimal integrated prediction model to generate a continuous canopy height map of the study area.

[0011] Furthermore, step 1 includes: acquiring and processing GEDI, ICESat-2, Sentinel-2 optical images, topographic and climate data, extracting features related to forest canopy height, and filtering features extracted from multi-source remote sensing data based on recursive feature elimination and cross-validation algorithms to retain predictive features used to build the model.

[0012] Furthermore, step 2 includes: using GEDI data as the response variable and key variables selected from ICESat-2 data as explanatory variables; constructing an integrated calibration model of RF, XGBoost and LightGBM regressors that are recursively stacked in sequence using the ant colony optimization algorithm; and calibrating the ICESat-2 data by optimizing the hyperparameters and stacking structure of each sub-model.

[0013] Furthermore, in the ant colony optimization algorithm, artificial ants construct solutions in the parameter space through a probability transfer mechanism, and each solution corresponds to a set of model hyperparameter combinations; the algorithm evaluates the quality of the solution based on the model prediction error, and guides the search process to converge toward a globally better solution through a pheromone update mechanism.

[0014] Furthermore, in step 3, the optimal ensemble prediction model is an ensemble model composed of XGBoost, RF and linear regression models recursively stacked in sequence.

[0015] Furthermore, the optimal ensemble prediction model structure and the hyperparameters of each sub-model are dynamically determined by applying the ant colony optimization algorithm again. The hyperparameters include the learning rate and tree structure parameters of XGBoost, the number of trees and maximum depth of RF, and the regularization parameters of linear regression.

[0016] Furthermore, in step 4, independent GEDI sample data are used as the true values, and the accuracy of the generated continuous canopy height map is verified by calculating the decision coefficient, root mean square error, and mean absolute error index.

[0017] On the other hand, the present invention provides a forest canopy height estimation device that optimizes the fusion of multi-source remote sensing data using ant colony optimization, comprising:

[0018] The preprocessing module is used for preprocessing and feature extraction of multi-source remote sensing data;

[0019] The calibration module is used to construct a calibration model for ICESat-2 data based on GEDI canopy height data and employs an ant colony optimization algorithm to generate a calibrated ICESat-2 canopy height dataset.

[0020] The prediction module is used to fuse the corrected ICESat-2 canopy height dataset with the preprocessed GEDI canopy height data as training samples, and combine the features extracted from multi-source remote sensing data to dynamically construct the optimal integrated prediction model by applying the ant colony optimization algorithm again.

[0021] The generation module is used to generate a continuous canopy height map of the study area by applying the optimal integrated prediction model.

[0022] Thirdly, the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned ant colony optimization fusion method for estimating forest canopy height from multi-source remote sensing data.

[0023] Fourthly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned ant colony optimization fusion method for estimating forest canopy height from multi-source remote sensing data.

[0024] The beneficial effects of this invention are as follows:

[0025] Significantly improves estimation accuracy and consistency: The ICESat-2 data correction model constructed using the ant colony optimization algorithm effectively corrects the systematic bias between it and high-precision GEDI data, increasing the consistency determination coefficient R between the two. 2The accuracy improved significantly from 0.45 to 0.91, with the root mean square error reduced by about 60%, laying a high-precision data foundation for subsequent fusion and prediction.

[0026] Achieving intelligent model construction and optimization: The ant colony optimization algorithm is used to dynamically search and determine the optimal combination of hyperparameters and ensemble structure of the prediction model, overcoming the limitations of traditional machine learning methods that rely on manual parameter tuning and fixed models. This enables the model to adapt to data features, thereby achieving stronger generalization ability and higher prediction reliability.

[0027] Generate high-resolution seamless products: By fusing corrected multi-source lidar data (ICESat-2 and GEDI) with Sentinel-2 optical imagery, terrain, climate and other multi-dimensional auxiliary features, the integrated prediction model can generate a spatially continuous forest canopy height map with a resolution of 30 meters, effectively making up for the lack of spatial coverage of a single sensor.

[0028] The method is robust and widely applicable: The integrated optimization framework adopted in this invention does not depend on specific sensor types or geographical regions. Its core lies in the adaptive integration and correction of multi-source data through intelligent algorithms. Therefore, it has strong robustness and can be extended to the estimation of canopy height in different geographical environments and forest types, providing an efficient and reliable technical solution for large-scale dynamic monitoring of forest resources. Attached Figure Description

[0029] Figure 1 This is a schematic diagram illustrating the principle of the forest canopy height estimation method based on ant colony optimization fusion of multi-source remote sensing data according to the present invention.

[0030] Figure 2 The image shows a comparison of the accuracy of ICESat-2 data before and after correction, where (a) corresponds to before correction and (b) corresponds to after correction.

[0031] Figure 3 This is a map showing the predicted forest canopy height generated based on the method of this invention.

[0032] Figure 4 The figure shows a comparison of experimental results based on the present invention and existing technologies. (a) shows the prediction accuracy of the RF model when using the original ICESat-2 data as training samples, (b) shows the prediction accuracy of the RF model when using corrected ICESat-2 data combined with GEDI samples as training samples, and (c) shows the prediction accuracy of the method of the present invention. Detailed Implementation

[0033] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0034] like Figure 1As shown, this invention aims to provide a high-precision, highly spatially continuous forest canopy height estimation method to address the problems in existing technologies, such as difficulties in fusing multi-source lidar data, the dependence of extrapolation models on fixed structures, and performance limitations imposed by hyperparameter selection. To achieve the above objective, this invention provides an ant colony optimization method for fusing multi-source remote sensing data to estimate forest canopy height, the method comprising:

[0035] Step 1: Multi-source remote sensing data preprocessing and feature extraction: Acquire and process multi-source remote sensing data including GEDI (L2A), ICESat-2 (ATL08), Sentinel-2 optical imagery, ERA5 climate data, and SRTM topographic data to extract key feature variables characterizing forest canopy height. A recursive feature elimination algorithm with cross-validation is used to filter the extracted features to identify and retain key variables for subsequent modeling.

[0036] The purpose of feature selection is to reduce overfitting, computational costs, and model complexity by removing irrelevant or redundant features. The implementation of recursive feature elimination consists of two parts: recursive feature elimination and cross-validation.

[0037] Specifically, recursive feature elimination uses a base model to rank all features according to their importance. Then, it sequentially selects different numbers of features to construct feature subsets, and the base model calculates the average score for each number of feature subsets. The optimal number of features is determined by the number of features corresponding to the highest average score, and finally, that number of features is selected in descending order of importance.

[0038] Step 2: Fusion and Correction of Spaceborne LiDAR Data: Using high-precision GEDI canopy height data (such as RH95) as a benchmark, an ant colony optimization algorithm is used to construct a correction model for systematic biases in ICESat-2 canopy height. This model aims to eliminate systematic errors and noise in the original ICESat-2 data, making it fusionable with the GEDI data in terms of accuracy and consistency, thereby generating a more consistent and less error-prone ICESat-2 canopy height dataset.

[0039] Previous studies have shown that GEDI's measurement accuracy is significantly higher than that of ICESat-2. Therefore, this invention uses the forest canopy height index (RH95) from GEDI L2A data as the response variable, and uses the RFECV method to screen key variables from ICESat-2 (ATL08) data as explanatory factors. Subsequently, the ACO algorithm is applied to construct an ICESat-2 data calibration model, which consists of three sub-models: RF, XGBoost, and LightGBM regressors. Its hyperparameters are obtained through iterative optimization using ACO.

[0040] Specifically, the ACO algorithm optimizes the hyperparameter configurations and model combination structures of each sub-model to achieve the best fitting effect for the calibration model. During the ACO optimization process, the hyperparameter combinations of each sub-model are considered candidate solutions in the ant colony search space. Each artificial ant corresponds to a specific model configuration scheme; that is, each hyperparameter combination is considered a solution constructed by an "ant." These hyperparameter combinations include the number of trees and maximum depth parameter of the RF regression model, the learning rate and subtree structure parameters of the XGBoost model, and the number of leaf nodes and regularization parameters of the LightGBM model. During the search process, the ants gradually construct complete parameter combination solutions through a probability transition mechanism and train the corresponding ensemble calibration model based on these parameter combinations.

[0041] This ensemble calibration model employs a recursive stacking approach, using the predictions of three base models—RF, XGBoost, and LightGBM regressors—as input features for the next sub-model in sequence. First, the RF model makes predictions on the data, and its result serves as input to the XGBoost model. Then, the XGBoost model makes predictions based on the RF output, and its result again serves as input to the LightGBM model. In this way, each sub-model further reduces the prediction error based on the output of the previous sub-model, thereby improving the model's prediction accuracy and enhancing the stability of the results. The ACO algorithm plays a crucial role in this process, optimizing the hyperparameter configuration and stacking structure of each sub-model to ensure that the model achieves optimal performance on different datasets.

[0042] ACO (Aggregate Optimization) is a swarm intelligence optimization algorithm based on the foraging behavior mechanism of ants. Its basic principle is that ants communicate indirectly through pheromone trails, guiding collective decision-making. ACO balances exploration and utilization through pheromone updating and evaporation mechanisms. In this invention, ACO is used as a dynamic hyperparameter optimization framework. During the search process, different parameter values ​​are probabilistically selected based on pheromone intensity and heuristic function, gradually constructing a complete set of ICESat-2 data calibration model parameter solutions.

[0043] In ACO, a group of artificial ants constructs a solution step by step by moving through the graph. The probability that the k-th ant moves from parameter node i to parameter node j at iteration t is defined as... :

[0044] ,

[0045] in, It represents the pheromone intensity on the parameter edge (i, j) (i.e., from parameter node i to parameter node j) at iteration step t. This indicates the heuristic information regarding the effect of this parameter value on model performance; α and β control the relative importance of pheromones and heuristic information. It is the set of parameter nodes that the k-th ant can choose in its current state.

[0046] After all ants have constructed their solutions, an ICESat-2 canopy height correction model is trained based on the corresponding parameter solutions. Using the GEDI RH95 canopy height as a reference, the model's prediction error index is calculated through cross-validation. This error index serves as an evaluation function for the ant solutions, guiding the pheromone update process. The pheromone intensity is updated according to the following formula:

[0047] ,

[0048] in, This represents the pheromone evaporation rate, where m is the number of ants. The pheromone deposited by the kth ant is usually defined as:

[0049]

[0050] Where Q represents the pheromone intensity constant, This represents the prediction error or loss function value corresponding to the model configuration. By iteratively constructing solutions and updating pheromones, the algorithm can enhance the search near high-quality solutions while maintaining diversity through pheromone evaporation.

[0051] Through the above iterative process, ACO gradually guides the search to focus on the parameter combination that can significantly reduce the difference in canopy height between ICESat-2 and GEDI, thereby obtaining the optimal correction model.

[0052] Step 3: Construction of the optimal forest canopy height extrapolation model: The ICESat-2 data is corrected using the calibration model constructed in Step 2, and then spatially fused with the preprocessed GEDI data to generate an enhanced lidar canopy height sample set. This sample set is used as the training response variable, combined with key features selected from multiple sources such as Sentinel-2, climate, and topography as explanatory variables, and ACO is applied again for dynamic modeling.

[0053] This step involves iterative search and optimization using ACO (Active Coding), with the optimization objective being to minimize the model's prediction error on the validation set. The prediction model combines three basic algorithms: XGBoost, RF (Rapid Forwarding), and linear regression, employing a recursive stacking method. This recursive stacking approach allows the model to progressively improve prediction accuracy at each layer.

[0054] ACO ensures optimal performance of the prediction model by optimizing the hyperparameter configuration of each sub-model. Specifically, ACO optimizes multiple hyperparameters of the three base models: XGBoost, RF, and linear regression.

[0055] For the XGBoost model, ACO optimizes parameters such as the learning rate, number of trees, maximum depth, and subsample ratio. The learning rate controls the step size for each update, the number of trees determines the model's complexity, the maximum depth controls the learning capacity of each tree, and the subsample ratio affects the amount of training data per tree. Adjusting these parameters directly impacts the model's training speed, fitting ability, and ability to prevent overfitting. By fine-tuning these hyperparameters, ACO enables XGBoost to better capture the complex relationships in the training data, thereby significantly improving the prediction accuracy of forest canopy height.

[0056] For RF models, ACO optimizes important parameters such as the number of trees and maximum depth. The number of trees determines the model's complexity and stability, while the maximum depth controls the learning ability of each tree. By adjusting these hyperparameters, ACO ensures that the RF model can fully learn the details of the input features while avoiding overfitting or underfitting, thereby improving the model's generalization ability.

[0057] In linear regression models, ACO optimizes regularization parameters, particularly L1 and L2 regularization parameters, which play a crucial role in handling high-dimensional data. Regularization controls model complexity and prevents the model from overfitting to noise. By optimizing regularization parameters, ACO ensures that the linear regression model can adapt to variations in multi-source data and maintains high prediction accuracy while avoiding overfitting.

[0058] The iterative optimization process of ACO goes beyond optimizing a single hyperparameter; it simultaneously tunes multiple hyperparameters to ensure optimal model performance across multiple dimensions. These optimized hyperparameter configurations enable each sub-model to efficiently process features from multiple data sources (such as ICESat-2, GEDI, Sentinel-2, etc.) and effectively extrapolate forest canopy height. Ultimately, the ACO-optimized model output provides more accurate and reliable results for predicting forest canopy height.

[0059] It is important to note that the model order and selection used in the calibration and prediction steps differ in this invention. Specifically, the calibration process uses three models: RF, XGBoost, and LightGBM, while the prediction step uses XGBoost, RF, and linear regression. This difference is based on the different requirements of the tasks. When calibrating the data, RF is chosen as the first model due to its complex features and numerous nonlinear relationships, as it performs exceptionally well in handling complex, nonlinear data. XGBoost and LightGBM are then used as the second and third models, respectively, to further optimize the calibration results through a stacking approach, ultimately yielding highly accurate calibrated data. In the prediction step, linear regression is chosen as the last model to generate forest canopy height using the calibrated data. Although simpler than nonlinear models, it maintains high generalization ability and effectively improves prediction speed by combining the outputs of other regression models. This model order and selection is an optimization based on the different characteristics of the calibration and prediction tasks, ensuring that each stage leverages the maximum advantage of the model, thereby improving overall performance.

[0060] Step 4: Forest Canopy Height Prediction Mapping and Accuracy Verification: Using the optimal integrated prediction model constructed in Step 3, calculations are performed on each 30-meter pixel in the entire study area to generate a spatially continuous and seamless forest canopy height distribution map. Finally, using independent preprocessed GEDI sample data as a ground truth reference, the prediction results are rigorously verified and evaluated by calculating indicators such as the coefficient of determination, root mean square error, and mean absolute error.

[0061] Among them, the present invention uses three indicators - the determination coefficient ( Accuracy is evaluated using root mean square error (RMSE) and mean absolute error (MAE). RMSE and MAE are used to evaluate the goodness of fit of the model, while RMSE and MAE are used to evaluate the error between the predicted and actual values.

[0062] (1)

[0063] (2)

[0064] (3)

[0065] in, and These represent the actual value and the predicted value of the i-th sample, respectively. denoted as the sample mean, and n represents the total number of samples.

[0066] After applying the ACO model to calibrate the canopy height extracted from ICESat-2 ATL08, a scatter plot was generated to compare the height differences between ICESat-2 ATL08 and GEDI RH95 data before and after calibration, and the results were calculated. Evaluation metrics such as RMSE and MAE were used. The results show that the ACO optimization model significantly improved the fitting accuracy between ICESat-2 and GEDI data and enhanced data consistency.

[0067] like Figure 2 As shown in (a), the canopy height derived from ICESat-2 data before calibration is randomly distributed relative to GEDI observations, with a coefficient of determination of... The accuracy is only 0.45, with a root mean square error of 3.77 meters and a mean square error (MAE) as high as 3.47 meters, indicating a low degree of agreement between the two. In contrast, after calibration, as... Figure 2 As shown in (b), the data points are closely clustered on the 1:1 reference line. The mean square error (RMSE) was significantly improved to 0.91, the root mean square error (RMSE) decreased to 1.56 m, and the mean square error (MAE) decreased to 1.08 m. These results indicate that the corrected ICESat-2 data have significantly higher consistency and reliability with the GEDI-derived canopy height estimates.

[0068] like Figure 3 As shown, by utilizing ACO optimized model parameters and corrected ICESat-2 and GEDI data, combined with multi-source optical remote sensing data, and integrating multiple machine learning algorithms, a forest canopy height map with a spatial resolution of 30 meters was generated for the current study area. This map visually reflects the spatial distribution pattern and continuous variation characteristics of forest canopy height within the study area. Different colors or shades represent different canopy height values, clearly identifying differences in canopy height caused by high-density forests, low-lying stands, forest gaps, and topographic relief. Therefore, this invention achieves high-precision spatial extrapolation from discrete point observations to continuous surfaces, providing a high-resolution basic data product for regional forest structure and biomass assessment.

[0069] like Figure 4 As shown, the RF model optimized by random search is compared with the proposed forest canopy height prediction model. Three comparative experiments were conducted, and significant differences in prediction accuracy were observed under the three experimental settings. (a) When using the original ICESat-2 data as training samples, the prediction accuracy of the RF model is relatively low and the consistency is poor (R0). 2 =0.45, RMSE=4.15m, MAE=3.15m); (b) When training with calibrated ICESat-2 data combined with GEDI samples and applying the RF model, the prediction performance is slightly improved (R 2=0.47, RMSE=4.08m, MAE=3.12m); (c) Based on the corrected ICESat-2 data, the model and parameters optimized by ACO significantly improved the reliability of canopy height prediction ( =0.60, RMSE=3.37m, MAE=2.56m). In summary, the canopy height estimation method proposed in this study, by optimizing and integrating GEDI and ICESat-2 data through ACO, can achieve better results than traditional methods for estimating forest canopy height.

[0070] On the other hand, the present invention provides an ant colony optimization fusion multi-source remote sensing data forest canopy height estimation device, which includes various modules capable of implementing the various steps of the aforementioned method, specifically including:

[0071] The preprocessing module is used for preprocessing and feature extraction of multi-source remote sensing data;

[0072] The calibration module is used to construct a calibration model for ICESat-2 data based on GEDI canopy height data and employs an ant colony optimization algorithm to generate a calibrated ICESat-2 canopy height dataset.

[0073] The prediction module is used to fuse the corrected ICESat-2 canopy height dataset with the preprocessed GEDI canopy height data as training samples, and combine the features extracted from multi-source remote sensing data to dynamically construct the optimal integrated prediction model by applying the ant colony optimization algorithm again.

[0074] The generation module is used to generate a continuous canopy height map of the study area by applying the optimal integrated prediction model.

[0075] Thirdly, the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned ant colony optimization fusion method for estimating forest canopy height from multi-source remote sensing data.

[0076] Fourthly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned ant colony optimization fusion method for estimating forest canopy height from multi-source remote sensing data.

[0077] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A forest canopy height estimation method based on ant colony optimization fusion of multi-source remote sensing data, characterized in that, The method includes: Step 1: Multi-source remote sensing data preprocessing and feature extraction; Step 2, Spaceborne LiDAR Data Fusion and Correction: Using GEDI canopy height data as a benchmark, an ant colony optimization algorithm is used to construct a correction model for ICESat-2 data, generating a corrected ICESat-2 canopy height dataset. Step 3: Construct the optimal forest canopy height extrapolation model: The ICESat-2 canopy height dataset and the preprocessed GEDI canopy height data are fused together as training samples. The features extracted from the multi-source remote sensing data are combined, and the ant colony optimization algorithm is applied again to dynamically construct the optimal integrated prediction model. Step 4: Forest canopy height prediction: Use the optimal integrated prediction model to generate a continuous canopy height map of the study area.

2. The forest canopy height estimation method based on ant colony optimization fusion of multi-source remote sensing data according to claim 1, characterized in that, Step 1 includes: acquiring and processing GEDI, ICESat-2, Sentinel-2 optical images, topographic and climate data, extracting features related to forest canopy height, and filtering features extracted from multi-source remote sensing data based on recursive feature elimination and cross-validation algorithms to retain predictive features used to build the model.

3. The forest canopy height estimation method based on ant colony optimization fusion of multi-source remote sensing data according to claim 1, characterized in that, Step 2 includes: Using GEDI canopy height data as the response variable and key variables selected from ICESat-2 data as explanatory variables, an integrated calibration model of RF, XGBoost and LightGBM regressors recursively stacked in sequence was constructed using the ant colony optimization algorithm. The ICESat-2 data was calibrated by optimizing the hyperparameters of each sub-model and the stacking structure.

4. The forest canopy height estimation method based on ant colony optimization fusion of multi-source remote sensing data according to claim 3, characterized in that, In the ant colony optimization algorithm, artificial ants construct solutions in the parameter space through a probability transfer mechanism, and each solution corresponds to a set of model hyperparameter combinations. The algorithm evaluates the quality of the solution based on the model prediction error and guides the search process to converge toward a better global solution through a pheromone update mechanism.

5. The forest canopy height estimation method based on ant colony optimization fusion of multi-source remote sensing data according to claim 1, characterized in that, In step 3, the optimal ensemble prediction model is an ensemble model composed of XGBoost, RF and linear regression models recursively stacked in sequence.

6. The forest canopy height estimation method based on ant colony optimization fusion of multi-source remote sensing data according to claim 5, characterized in that, The optimal ensemble prediction model structure and the hyperparameters of each sub-model are dynamically determined by applying the ant colony optimization algorithm again. The hyperparameters include the learning rate and tree structure parameters of XGBoost, the number of trees and maximum depth of RF, and the regularization parameters of linear regression.

7. The forest canopy height estimation method based on ant colony optimization fusion of multi-source remote sensing data according to claim 1, characterized in that, In step 4, independent GEDI sample data are used as the true value, and the accuracy of the generated continuous canopy height map is verified by calculating the determination coefficient, root mean square error, and mean absolute error index.

8. A forest canopy height estimation device that optimizes and fuses multi-source remote sensing data using ant colony optimization, characterized in that... include: The preprocessing module is used for preprocessing and feature extraction of multi-source remote sensing data; The calibration module is used to construct a calibration model for ICESat-2 data based on GEDI canopy height data and employs an ant colony optimization algorithm to generate a calibrated ICESat-2 canopy height dataset. The prediction module is used to fuse the corrected ICESat-2 canopy height dataset with the preprocessed GEDI canopy height data as training samples, and combine the features extracted from multi-source remote sensing data to dynamically construct the optimal integrated prediction model by applying the ant colony optimization algorithm again. The generation module is used to generate a continuous canopy height map of the study area by applying the optimal integrated prediction model.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the ant colony optimization fusion multi-source remote sensing data forest canopy height estimation method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, enable the processor to implement the ant colony optimization fusion method for estimating forest canopy height based on any one of claims 1-7.

Images