Tree species classification method based on laser point cloud data and remote sensing image and related equipment

Abstract

The invention provides a tree species classification method based on laser point cloud data and remote sensing images and related equipment. The method comprises the steps that the laser point cloud data of a target area and the remote sensing images of a long-time sequence are acquired; extracting structural features of the tree based on the laser point cloud data, and extracting phenological features based on the remote sensing image; and inputting the structural features and the phenological features into a pre-trained neural network model for tree species classification to obtain a classification result. The tree species classification method based on the laser point cloud data and the remote sensing image and related equipment provided by the invention are simple and convenient, trees with similar morphological structures can be effectively distinguished, and the classification precision is high.

Description

Technical Field

[0001] This application relates to the field of tree species classification technology, and in particular to a tree species classification method and related equipment based on laser point cloud data and remote sensing images. Background Technology

[0002] Existing tree species classification methods mainly rely on manual field surveys, which suffer from low efficiency, high cost, and strong subjectivity. LiDAR technology can acquire laser point cloud data of trees, providing a new technical means for automatic tree species identification. By actively emitting laser pulses and receiving echoes, LiDAR technology can penetrate the vegetation canopy to obtain the three-dimensional spatial structure information of trees, offering advantages such as high precision, all-weather operation, and non-contact processing. It can be used for forest resource surveys, biodiversity monitoring, carbon sequestration, and precision forestry management, solving the problems of low efficiency, high cost, and strong subjectivity associated with traditional manual surveys, and achieving automated, high-precision tree species identification. However, some trees have similar morphological structures, making it difficult to effectively distinguish them based solely on point cloud features. Therefore, a more accurate tree species classification method is urgently needed. Summary of the Invention

[0003] In view of this, the purpose of this application is to propose a tree species classification method and related equipment based on laser point cloud data and remote sensing imagery to solve the above-mentioned technical problems.

[0004] A first aspect of this application provides a tree species classification method based on laser point cloud data and remote sensing imagery, comprising: acquiring laser point cloud data and long-term remote sensing imagery of a target area; extracting structural features of trees based on the laser point cloud data and extracting phenological features based on the remote sensing imagery; inputting the structural features and the phenological features into a pre-trained neural network model for tree species classification to obtain classification results.

[0005] Furthermore, the remote sensing imagery covers at least one complete growth cycle of the target area.

[0006] Furthermore, the structural features include the tree's height, diameter at breast height (DBH), crown diameter, crown area, crown volume, branch height, trunk volume, and trunk curvature.

[0007] Furthermore, the phenological characteristics include the start of the growing season, the end of the growing season, and the length of the growing season.

[0008] Furthermore, before inputting the structural features and phenological features into a pre-trained neural network model for tree species classification, the process includes: standardizing the structural features and phenological features.

[0009] Furthermore, the neural network model includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer includes 11 neurons, the first hidden layer includes 100 neurons, the second hidden layer includes 50 neurons, and the number of neurons in the output layer is equal to the number of tree species.

[0010] Furthermore, the activation function used in the neural network model is a linear rectified function, the neural network model uses an adaptive moment estimation optimizer for parameter updates, the patience coefficient of the early stopping mechanism of the neural network model is 20 rounds, and the maximum number of iterations is 500 rounds.

[0011] Furthermore, the classification results include a tree species classification confusion matrix heatmap and a tree species classification report, the tree species classification report including precision, recall and F1 score.

[0012] A second aspect of this application provides a tree species classification device based on laser point cloud data and remote sensing imagery, comprising: an acquisition module configured to acquire laser point cloud data and long-term remote sensing imagery of a target area; a feature extraction module configured to extract structural features of trees based on the laser point cloud data and extract phenological features based on the remote sensing imagery; and a classification module configured to input the structural features and the phenological features into a pre-trained neural network model for classification, thereby obtaining a tree species classification result.

[0013] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the tree species classification method based on laser point cloud data and remote sensing imagery as described in the first aspect above.

[0014] A fourth aspect of this application provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the tree species classification method based on laser point cloud data and remote sensing imagery as described in the first aspect above.

[0015] A fifth aspect of this application provides a computer program product, including computer program instructions that, when executed on a computer, cause the computer to perform the tree species classification method based on laser point cloud data and remote sensing imagery as described in the first aspect above.

[0016] As described above, this application provides a tree species classification method and related equipment based on laser point cloud data and remote sensing imagery. The method includes: acquiring laser point cloud data and long-term remote sensing imagery of a target area; extracting structural features of trees based on the laser point cloud data and extracting phenological features based on the remote sensing imagery; inputting the structural features and phenological features into a pre-trained neural network model for tree species classification to obtain the classification result. Extracting structural features from laser point cloud data and phenological features from remote sensing imagery, and then fusing these features into the model for tree species classification, has been tested and shown to significantly improve classification accuracy, effectively distinguishing between evergreen and deciduous trees, and also different types of deciduous trees. This tree species classification method and related equipment based on laser point cloud data and remote sensing imagery is simple and convenient, effectively distinguishing trees with similar morphological structures, and achieving high classification accuracy. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a tree species classification method based on laser point cloud data and remote sensing imagery, as described in an embodiment of this application.

[0019] Figure 2 This is a heatmap of the tree species classification confusion matrix for Comparative Example 1 of this application.

[0020] Figure 3 This is a heatmap of the tree species classification confusion matrix for Comparative Example 2 of this application.

[0021] Figure 4 This is a heatmap of the tree species classification confusion matrix for Comparative Example 3 of this application.

[0022] Figure 5 This is a heatmap of the tree species classification confusion matrix for Comparative Example 4 of this application.

[0023] Figure 6 This is a heatmap of the tree species classification confusion matrix for Comparative Example 5 of this application.

[0024] Figure 7 This is a heatmap of the tree species classification confusion matrix for Comparative Example 6 of this application.

[0025] Figure 8 This is a heatmap of the tree species classification confusion matrix for Comparative Example 7 of this application.

[0026] Figure 9This is a heatmap of the tree species classification confusion matrix for Comparative Example 8 of this application.

[0027] Figure 10 This is a heatmap of the tree species classification confusion matrix for Comparative Example 9 of this application.

[0028] Figure 11 This is a heatmap of the tree species classification confusion matrix for Comparative Example 10 of this application.

[0029] Figure 12 This is a heatmap of the tree species classification confusion matrix for Comparative Example 11 of this application.

[0030] Figure 13 This is a heatmap of the tree species classification confusion matrix for Comparative Example 12 of this application.

[0031] Figure 14 This is a heatmap of the tree species classification confusion matrix for Comparative Example 13 of this application.

[0032] Figure 15 This is a heatmap of the tree species classification confusion matrix for Comparative Example 14 of this application.

[0033] Figure 16 This is a heatmap of the tree species classification confusion matrix for Embodiment 1 of this application.

[0034] Figure 17 This is a heatmap of the tree species classification confusion matrix for Embodiment 2 of this application.

[0035] Figure 18 This is a heatmap of the tree species classification confusion matrix for Embodiment 3 of this application.

[0036] Figure 19 This is a heatmap of the tree species classification confusion matrix for Embodiment 4 of this application.

[0037] Figure 20 This is a heatmap of the tree species classification confusion matrix for Embodiment 5 of this application.

[0038] Figure 21 This is a heatmap of the tree species classification confusion matrix for Embodiment 6 of this application.

[0039] Figure 22 This is a schematic diagram of a tree species classification device based on laser point cloud data and remote sensing imagery, as described in an embodiment of this application.

[0040] Figure 23 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0042] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0043] Some technologies offer a LiDAR point cloud tree species classification method based on single morphological features, extracting the canopy height model (CHM) features from individual tree point clouds and inputting them into an SVM classifier to distinguish between conifers and broad-leaved trees. However, this method has a single feature dimension, utilizing only static geometric features and failing to consider the physiological rhythm differences between tree species (such as budding and leaf fall periods). It struggles to distinguish between morphologically similar but phenologically different species (such as evergreen Masson pine and deciduous fir). Furthermore, it lacks core stand indicators, omitting the basic forestry survey indicator diameter at breast height (DBH), leading to a disconnect between classification results and forest volume estimation, thus limiting its practical value. Additionally, the model's generalization ability is insufficient; traditional machine learning algorithms have limited ability to nonlinearly map high-dimensional features, resulting in a significant decrease in accuracy under complex stand conditions.

[0044] Some technologies offer an end-to-end tree species identification method based on deep learning point cloud segmentation. This method voxelizes individual tree point clouds and inputs them into a PointNet++ network to automatically extract deep features and complete classification. However, it suffers from poor physical interpretability. The "black box" nature of neural networks leads to unclear classification criteria, making it difficult to trace the causes of errors and failing to meet the interpretability requirements of forestry surveys. Furthermore, it has a high data dependency, requiring massive amounts of labeled point cloud data for training, while sample acquisition in the forestry field is extremely costly, resulting in a sharp drop in accuracy under small sample conditions. Additionally, it fails to integrate multi-source information, utilizing only point cloud geometric coordinates and neglecting to incorporate time-series phenological information (such as the start / end of the growing season), thus underutilizing feature information.

[0045] Currently, most laser point cloud tree species classification methods rely solely on structural features of trees. Some methods combining remote sensing utilize spectral features from remote sensing images. This application, however, introduces phenological features extracted from remote sensing. Addressing the shortcomings of related technologies, such as single feature dimensions, lack of core stand indicators, failure to utilize phenological information, and poor model interpretability, this application provides a tree species classification method based on 11-dimensional comprehensive features and a multi-layered sensing mechanism. This method can achieve high-precision, interpretable, and forestry-applicable tree species identification.

[0046] The following describes specific embodiments in conjunction with... Figures 1 to 23 The technical solution of this application will be described in detail below.

[0047] Some embodiments of this application provide a tree species classification method based on laser point cloud data and remote sensing imagery, such as... Figure 1 As shown, it includes the following steps:

[0048] S1. Acquire laser point cloud data and long-term remote sensing images of the target area.

[0049] For target areas such as woodlands, handheld LiDAR scanners can be used to acquire LiDAR point cloud data of the woodland. If the handheld LiDAR scanner does not have a built-in positioning system, control points need to be set up around the woodland. Control points should be sufficiently stable and not easily moved or damaged. Concrete stakes, metal nails, or other durable materials can be used to mark the control points to ensure long-term stability. Use GPS to record detailed location information for each control point, including its latitude and longitude, and a sketch of its distribution, for later point cloud calculation reference. Ensure there are no strong electromagnetic interference sources around the control points, such as high-voltage lines or radio stations. Also, record the tree species at each control point.

[0050] Long-term remote sensing images of forests can be obtained through satellite image centers. Long-term images refer to remote sensing images that cover at least one complete growth cycle, that is, the process of forests from early spring when leaves sprout to when the branches and leaves are lush and all the leaves have fallen.

[0051] S2. Extract the structural features of trees based on the laser point cloud data, and extract the climate features based on the remote sensing images.

[0052] The point cloud data from the LiDAR scanner can be processed using its built-in software, and geometric registration can be performed using the aforementioned control points. The obtained point cloud with actual geospatial coordinates is preprocessed, including resampling and denoising. The CSF (Cloth Simulation Filter) algorithm is used for ground point classification, separating the point cloud into ground points and non-ground points. Point cloud height is normalized based on the ground points. Then, individual tree segmentation is performed based on the tree trunk. From the individual tree point cloud data, location parameters, such as spatial coordinates, can be obtained. Structural features of the trees can be used to extract morphological parameters, such as diameter at breast height (DBH), tree height (TH), and height below the branch (HFB); crown parameters, such as crown diameter (CD), north-south crown diameter (NSCD), east-west crown diameter (EWCD), crown area (CA), and crown volume (CV); and trunk parameters, such as trunk volume (TV) and trunk curvature (TC). Among them, tree height can reflect the vertical growth of trees, diameter at breast height (DBH) is the core indicator reflecting tree growth, crown diameter can characterize the horizontal expansion of the crown, north-south crown diameter and east-west crown diameter can describe the directional differences in crown morphology, crown area and crown volume can quantify the crown's ability to occupy space, height below the branch reflects the height of the trunk without branches, trunk volume can characterize trunk biomass, and trunk curvature can characterize trunk morphology.

[0053] For the remote sensing imagery used in this application, the embodiments and comparative examples employ Sentinel-2 L2A imagery from December 2024 to November 2025 (excluding images from December 2025 due to snowfall), which is an atmospherically corrected surface inversion image. NDVI (Normalized Difference Vegetation Index) values ​​are first calculated for the imagery. A long-term NDVI series image is constructed. The time-series NDVI curve is smoothed using a dual-logic function. Phenological features are extracted using a dynamic thresholding method, including SOS (Start of Season), EOS (End of Season), DOS (Duration of Season), and NDVI max (Annual NDVI Max). The Start of Season is the date of vegetation greening based on the time-series NDVI change; the End of Season is the date of leaf fall / growth stagnation based on the time-series NDVI decay; and the growth season length characterizes the length of the tree's annual growth period.

[0054] S3. Input the structural features and phenological features into a pre-trained neural network model to classify tree species and obtain the classification results.

[0055] By combining static morphological and structural features with dynamic phenological features, this method captures morphological and physiological-ecological differences among tree species, significantly improving the ability to distinguish similar species, such as evergreen and deciduous trees, and early-budding and late-budding trees. Tests have shown that combining these two features into a neural network model results in higher classification accuracy compared to classifying by inputting structural or phenological features alone.

[0056] The classification results include a tree species confusion matrix heatmap and a tree species classification report. The tree species confusion matrix heatmap provides a visual representation of the classification results. The tree species classification report includes evaluation metrics such as precision, recall, and F1 score, which provide insights into the classification performance. The report may also include model parameters, error details, etc., without any specific limitations.

[0057] This tree species classification method, based on laser point cloud data and remote sensing imagery, is simple and convenient, effectively distinguishing trees with similar morphological structures and achieving high classification accuracy. Combining structural and phenological characteristics for tree species classification solves the problem of some trees with similar appearances that are difficult to distinguish using only spatial geometric features. Introducing phenological characteristics not only distinguishes between evergreen and deciduous trees but also differentiates between various types of deciduous trees.

[0058] In some embodiments, the structural features include tree height, diameter at breast height (DBH), crown diameter, crown area, crown volume, branch height, trunk volume, and trunk curvature. The phenological features include the start of the growing season, the end of the growing season, and the length of the growing season.

[0059] As mentioned earlier, trees have a wide variety of structural features. This application conducted extensive testing and found that the highest classification accuracy was obtained by using the eight-dimensional structural features (TH+DBH+CD+CA+CV+HFB+TV+TC) as input model for individual structural features.

[0060] The phenological cycle of trees can be described as a one-dimensional timeline consisting of three key nodes: Phenological Cycle = [Start of Growing Season (SOS)] → [During Growing Season] → [End of Growing Season (EOS),] where the length of the growing season (DOS) = EOS - SOS. SOS (Start of Season) characterizes the tree's budding / greening, representing its photoperiodic sensitivity and accumulated temperature requirements. EOS (End of Season) characterizes the tree's leaf fall / dormancy, representing its cold resistance and nutrient recovery strategy. DOS (Duration of Season) characterizes the tree's growth strategy; a long DOS indicates a fast-growing tree, while a short DOS indicates a conservative tree.

[0061] Table 1. Tree Species Classification Table

[0062] Based on these three phenological characteristics, trees can be classified as shown in Table 1, where DOY represents the cumulative day of year. It can be seen that using only the SOS or EOS feature cannot distinguish between tree species with "early budding-early leaf fall" (short DOS) and those with "late budding-late leaf fall" (also short DOS); all three phenological characteristics must be used simultaneously. This application also tested the phenological characteristics. For individual phenological characteristics, using the SOS+EOS+DOS three-dimensional phenological feature input model for classification yielded the highest classification accuracy. Furthermore, combining the aforementioned eight-dimensional structural features with these three-dimensional phenological features as input model for classification achieves even higher classification accuracy.

[0063] In some embodiments, the neural network model includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer includes 11 neurons, the first hidden layer includes 100 neurons, the second hidden layer includes 50 neurons, and the number of neurons in the output layer is equal to the number of tree species.

[0064] By using a multilayer perceptron (MLP) neural network model, higher test accuracy can be achieved, and the highest overall efficiency score is achieved when the first hidden layer contains 100 neurons and the second hidden layer contains 50 neurons.

[0065] In some embodiments, before inputting the structural features and phenological features into a pre-trained neural network model for tree species classification, the process includes standardizing the structural features and phenological features. The neural network model uses a linear rectified function as its activation function, employs an adaptive moment estimator optimizer for parameter updates, and has an early stopping mechanism with a patience coefficient of 20 rounds and a maximum iteration count of 500 rounds.

[0066] Label encoding is used to convert tree species categories into numerical labels. Standard deviation standardization (StandardScaler) is then applied to normalize each feature dimension, eliminating dimensional differences. The normalization formula is as follows: The overall classification performance is evaluated using accuracy. The model output can provide a list of misclassified samples, including the sample ID, true label, and predicted label.

[0067] Comparative Example 1 A dataset was built based on a forest area. Phenological characteristics were associated with individual trees according to their location parameters, thus constructing a dataset containing information on tree species, structural features, and phenological characteristics. The forest area includes six tree species (130 poplar trees, 130 black locust trees, 130 mulberry trees, 130 maple trees, 130 walnut trees, and 130 Chinese pine trees). The dataset was divided into training and testing sets in an 8:2 ratio. Stratified sampling was used to ensure a consistent proportion of samples in each category. Unique identifiers (IDs) were retained for each sample to enable accurate tracking of misclassified samples.

[0068] A neural network model was constructed, consisting of an input layer (11 neurons), a first hidden layer (100 neurons), a second hidden layer (50 neurons), and an output layer (6 neurons, corresponding to 6 tree species: poplar, black locust, mulberry, maple, walnut, and pine). The model uses the ReLU activation function. The Adam optimizer is used for parameter updates, with adaptive learning rate adjustment. Early stopping is implemented: a validation set ratio of 10%, a patience factor of 20 epochs, and a maximum of 500 iterations. Batch gradient descent (batch size=32) is used for parameter optimization.

[0069] Then, only the 3D structural features (TH+DBH+CD) were input into the model for tree species classification. The classification accuracy was 66.03%, and the specific classification report is shown in Table 2. The tree species classification confusion matrix heatmap is shown below. Figure 2 As shown.

[0070] Table 2 Classification Report Form

[0071] Comparative Example 2 Using the model from Comparative Example 1, only the 5-dimensional structural features (TH+DBH+CD+CA+CV) were input into the model for tree species classification, without considering directionality and trunk morphology. The classification accuracy was 73.72%, and the specific classification report is shown in Table 3. The tree species classification confusion matrix heatmap is shown below. Figure 3 As shown.

[0072] Table 3 Classification Report Form

[0073] Comparative Example 3 Using the model from Comparative Example 1, only the 6-dimensional structural features (TH+DBH+CD+CA+CV+HFB) were input into the model for tree species classification. The classification accuracy was 75.64%, and the specific classification report is shown in Table 4. The tree species classification confusion matrix heatmap is shown below. Figure 4 As shown.

[0074] Table 4 Classification Report Form

[0075] Comparative Example 4 Using the model from Comparative Example 1, only the 7-dimensional structural features (TH+DBH+CD+CA+CV+NSCD+EWCD) were input into the model for tree species classification. The classification accuracy was 68.59%, and the specific classification report is shown in Table 5. The tree species classification confusion matrix heatmap is shown below. Figure 5 As shown.

[0076] Table 5 Classification Report Form

[0077] Comparative Example 5 Using the model from Comparative Example 1, only the 7-dimensional structural features (TH+DBH+CD+CA+CV+HFB+TV) were input into the model for tree species classification. The classification accuracy was 76.92%, and the specific classification report is shown in Table 6. The tree species classification confusion matrix heatmap is shown below. Figure 6 As shown.

[0078] Table 6 Classification Report Form

[0079] Comparative Example 6 Using the model from Comparative Example 1, 8-dimensional structural features (TH+DBH+CD+CA+CV+HFB+TV+TC) were input into the model for tree species classification. The classification accuracy was 77.56%, and the specific classification report is shown in Table 7. The tree species classification confusion matrix heatmap is shown below. Figure 7 As shown.

[0080] Table 7 Classification Report Form

[0081] Comparative Example 7 Using the model from Comparative Example 1, 10-dimensional structural features (TH+DBH+CD+CA+CV+HFB+TV+TC+NSCD+EWCD) were input into the model for tree species classification. The classification accuracy was 76.28%, and the specific classification report is shown in Table 8. The tree species classification confusion matrix heatmap is shown below. Figure 8 As shown.

[0082] Table 8 Classification Report Form

[0083] Comparative Example 8 Using the model from Comparative Example 1, 9-dimensional structural features (TH+DBH+CD+CA+CV+HFB+TV+TC+HCR) were input into the model for tree species classification. The classification accuracy was 76.92%, and the specific classification report is shown in Table 9. The tree species classification confusion matrix heatmap is shown below. Figure 9 As shown.

[0084] Table 9 Classification Report Form

[0085] Comparative Examples 1 to 8 were used to test the classification accuracy of structural features of different dimensions. The results are summarized in Table 10. The classification accuracy was highest when 8-dimensional features (TH+DBH+CD+CA+CV+HFB+TV+TC) were selected as the structural features of single tree point clouds. Too many or too few structural features would reduce the classification accuracy.

[0086] Table 10 Classification Results

[0087] Comparative Example 9 Using the model from Comparative Example 1, 1D phenological features (SOS) were input into the model for tree species classification. The classification accuracy was 67.95%, and the specific classification report is shown in Table 11. The tree species classification confusion matrix heatmap is shown below. Figure 10 As shown.

[0088] Table 11 Classification Report Form

[0089] Comparative Example 10 Using the model from Comparative Example 1, one-dimensional phenological features (EOS) were input into the model for tree species classification. The classification accuracy was 64.74%, and the specific classification report is shown in Table 12. The tree species classification confusion matrix heatmap is shown below. Figure 11 As shown.

[0090] Table 12 Classification Report Form

[0091] Comparative Example 11 Using the model from Comparative Example 1, one-dimensional phenological features (DOS) were input into the model for tree species classification. The classification accuracy was 78.21%, and the specific classification report is shown in Table 13. The tree species classification confusion matrix heatmap is shown below. Figure 12 As shown.

[0092] Table 13 Classification Report Form

[0093] Comparative Example 12 Using the model from Comparative Example 1, 2D phenological features (SOS+EOS) were input into the model for tree species classification. The classification accuracy was 85.26%, and the specific classification report is shown in Table 14. The tree species classification confusion matrix heatmap is shown below. Figure 13 As shown.

[0094] Table 14 Classification Report Form

[0095] Comparative Example 13 Using the model from Comparative Example 1, 2D phenological features (EOS+DOS) were input into the model for tree species classification. The classification accuracy was 75.00%, and the specific classification report is shown in Table 15. The tree species classification confusion matrix heatmap is shown below. Figure 14 As shown.

[0096] Table 15 Classification Report Form

[0097] Comparative Example 14 Using the model from Comparative Example 1, 3D phenological features (SOS+EOS+DOS) were input into the model for tree species classification. The classification accuracy was 87.18%, and the specific classification report is shown in Table 16. The tree species classification confusion matrix heatmap is shown below. Figure 15 As shown.

[0098] Table 16 Classification Report Form

[0099] Example 1 Using the model from Comparative Example 1, 8-dimensional structural features (TH+DBH+CD+CA+CV+HFB+TV+TC) and 1-dimensional phenological features (SOS) were input into the model for tree species classification. The classification accuracy was 88.46%. The specific classification report is shown in Table 17, and the tree species classification confusion matrix heatmap is shown below. Figure 16 As shown.

[0100] Table 17 Classification Report Form

[0101] Example 2 Using the model from Comparative Example 1, 8-dimensional structural features (TH+DBH+CD+CA+CV+HFB+TV+TC) and 1-dimensional phenological features (EOS) were input into the model for tree species classification. The classification accuracy was 89.10%. The specific classification report is shown in Table 18, and the tree species classification confusion matrix heatmap is shown below. Figure 17 As shown.

[0102] Table 18 Classification Report Form

[0103] Example 3 Using the model from Comparative Example 1, 8-dimensional structural features (TH+DBH+CD+CA+CV+HFB+TV+TC) and 1-dimensional phenological features (DOS) were input into the model for tree species classification. The classification accuracy was 91.67%. The specific classification report is shown in Table 19, and the tree species classification confusion matrix heatmap is shown below. Figure 18 As shown.

[0104] Table 19 Classification Report Form

[0105] Example 4 Using the model from Comparative Example 1, 8-dimensional structural features (TH+DBH+CD+CA+CV+HFB+TV+TC) and 2-dimensional phenological features (SOS+EOS) were input into the model for tree species classification. The classification accuracy was 95.51%, and the specific classification report is shown in Table 20. The tree species classification confusion matrix heatmap is shown below. Figure 19 As shown.

[0106] Table 20 Classification Report Form

[0107] Example 5 Using the model from Comparative Example 1, 8-dimensional structural features (TH+DBH+CD+CA+CV+HFB+TV+TC) and 3-dimensional phenological features (SOS+EOS+DOS) were input into the model for tree species classification. The classification accuracy was 96.79%. The specific classification report is shown in Table 21, and the tree species classification confusion matrix heatmap is shown below. Figure 20 As shown.

[0108] Table 21 Classification Report Form

[0109] Example 6 Using the model from Comparative Example 1, 8-dimensional structural features (TH+DBH+CD+CA+CV+HFB+TV+TC) and 4-dimensional phenological features (SOS+EOS+DOS+NDVImax) were input into the model for tree species classification. The classification accuracy was 95.51%. The specific classification report is shown in Table 22, and the tree species classification confusion matrix heatmap is shown below. Figure 21 As shown.

[0110] Table 22 Classification Report Form

[0111] Comparative Examples 9 to 14 tested the classification accuracy of phenological features with different dimensions. The results are summarized in Table 23. The highest classification accuracy was achieved when 3-dimensional features (SOS+EOS+DOS) were selected as the phenological features. Examples 1 to 6 tested the classification accuracy of phenological features plus structural features with different dimensions. The results are summarized in Table 23. Comparing the data from the examples and comparative examples, it can be seen that the classification accuracy of combining phenological features and structural features into the input model is higher than that of inputting phenological features or structural features alone. Among them, the highest classification accuracy was achieved when 11-dimensional features (TH+DBH+CD+CA+CV+HFB+TV+TC+SOS+EOS+DOS) were selected as the input model. Too many or too few features will reduce the classification accuracy.

[0112] Table 23 Classification Results

[0113] Example 7 The model from Comparative Example 1 was adjusted to use a single-layer narrow hidden layer structure with 50 neurons. It was tested on a forest area, using the aforementioned 11-dimensional features as input features. The test results were: Fold 1 / 5... TestAcc=0.8974, Fold 2 / 5... TestAcc=0.9615, Fold 3 / 5... TestAcc=0.9167, Fold 4 / 5... TestAcc=0.9487, Fold 5 / 5... TestAcc=0.9103, TestAcc=0.9269±0.0242, Overfit=0.0147, Time=0.46s.

[0114] Example 8 The model of Comparative Example 1 was adjusted to use a single-layer wide hidden layer structure with 100 neurons. It was tested on a forest area, using the aforementioned 11-dimensional features as input features. The test results were: Fold 1 / 5... TestAcc=0.9487, Fold 2 / 5... TestAcc=0.9615, Fold 3 / 5... TestAcc=0.9103, Fold 4 / 5... TestAcc=0.8846, Fold 5 / 5... TestAcc=0.9551, TestAcc=0.9321±0.0297, Overfit=0.0061, Time=0.43s.

[0115] Example 9 The model in Comparative Example 1 was adjusted to use a two-layer hidden layer structure. The first hidden layer contains 100 neurons, and the second hidden layer contains 50 neurons. The model was tested on a forest land. The input features used the aforementioned 11-dimensional features. The test results were as follows: Fold1 / 5... TestAcc=0.9423, Fold2 / 5... TestAcc=0.9679, Fold3 / 5... TestAcc=0.9167, Fold4 / 5... TestAcc=0.9615, Fold5 / 5... TestAcc=0.9423, TestAcc=0.9462±0.0179, Overfit=0.0256, Time=0.60s.

[0116] Example 10 The model in Comparative Example 1 was adjusted to use a two-layer wide hidden layer structure. The first hidden layer contains 200 neurons, and the second hidden layer contains 100 neurons. The model was tested on a forest land. The input features used the aforementioned 11-dimensional features. The test results were as follows: Fold 1 / 5... TestAcc=0.9551, Fold 2 / 5... TestAcc=0.9679, Fold 3 / 5... TestAcc=0.9423, Fold 4 / 5... TestAcc=0.9295, Fold 5 / 5... TestAcc=0.9103, TestAcc=0.9410±0.0200, Overfit=0.0279, Time=2.06s.

[0117] Example 11 The model in Comparative Example 1 was adjusted to use a three-layer hidden layer structure. The first hidden layer contains 100 neurons, the second hidden layer contains 50 neurons, and the third hidden layer contains 25 neurons. A test was conducted on a forest area, using the aforementioned 11-dimensional features as input features. The test results were: Fold 1 / 5... TestAcc=0.9359, Fold 2 / 5... TestAcc=0.9551, Fold 3 / 5... TestAcc=0.9167, Fold 4 / 5... TestAcc=0.9679, Fold 5 / 5... TestAcc=0.9295, TestAcc=0.9410±0.0183, Overfit=0.0215, Time=0.53s.

[0118] Example 12 The model in Comparative Example 1 was adjusted to use a three-layer wide hidden layer structure. The first hidden layer contains 200 neurons, the second hidden layer contains 100 neurons, and the third hidden layer contains 505 neurons. A test was conducted on a forest area, using the aforementioned 11-dimensional features as input features. The test results were: Fold 1 / 5... TestAcc=0.9423, Fold 2 / 5... TestAcc=0.9551, Fold 3 / 5... TestAcc=0.9167, Fold 4 / 5... TestAcc=0.9615, Fold 5 / 5... TestAcc=0.9679, TestAcc=0.9487±0.0181, Overfit=0.0231, Time=1.92s.

[0119] Examples 7 to 12 compared the impact of different hidden layer structures on model classification, and the test results are summarized in Table 24.

[0120] Table 24 Model Evaluation Table

[0121] It can be seen that the model with a double hidden layer (100-50) structure achieves Pareto optimality in terms of accuracy-complexity trade-off, stability, diminishing marginal utility threshold, and small sample adaptability. As shown in Table 25, for the accuracy-complexity tradeoff, Example 9 achieved a test accuracy of 94.62%, a significant improvement over the single-layer structure (92-93%) (Cohen's d>0.6), while the number of parameters (6,556) was only 23.5% of that of the three-layer wide-depth structure (27,856). For optimal stability, Example 9 had the lowest 5-fold cross-validation standard deviation of 1.79%, proving the reproducibility of the results. For the diminishing marginal utility threshold, the three-layer structure only improved accuracy by 0.25%, but increased computational cost by 225%, which does not meet the requirements for practical engineering applications. For small sample adaptability, the sample-to-parameter ratio of this dataset was 119:1 (780 / 6,556), which is within the optimal generalization range of MLP (100-200:1), while the three-layer structure decreased to 28:1, significantly increasing the risk of overfitting.

[0122] Therefore, the double hidden layer (100-50) in Example 9 is the optimal network depth for classifying tree species based on structural and phenological features, and it is at the Pareto front in the three-dimensional trade-off of classification accuracy, model stability, and computational efficiency.

[0123] Table 25 Pareto Evaluation Table

[0124] Based on the same inventive concept, this application also provides a tree species classification device based on laser point cloud data and remote sensing imagery, such as... Figure 22 As shown, it includes: an acquisition module 21, configured to acquire laser point cloud data and long-term remote sensing images of the target area; a feature extraction module 22, configured to extract structural features of trees based on the laser point cloud data and extract phenological features based on the remote sensing images; and a classification module 23, configured to input the structural features and the phenological features into a pre-trained neural network model for classification to obtain the classification result of the tree species.

[0125] The apparatus in the above embodiments is used to implement the tree species classification method based on laser point cloud data and remote sensing imagery corresponding to any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0126] Based on the same inventive concept, corresponding to any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the tree species classification method based on laser point cloud data and remote sensing images as described in any of the above embodiments.

[0127] Figure 23 This embodiment illustrates a more specific hardware structure of an electronic device, which may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0128] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0129] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0130] The input / output interface 1030 is used to connect input / output modules to realize information input and output. The input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. The input devices can include keyboards, mice, touchscreens, microphones, various sensors, etc., and the output devices can include displays, speakers, vibrators, indicator lights, etc.

[0131] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (e.g., USB, Ethernet cable, etc.) or wireless means (e.g., mobile network, WIFI, Bluetooth, etc.).

[0132] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0133] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0134] The electronic devices described above are used to implement the tree species classification methods based on laser point cloud data and remote sensing images corresponding to any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0135] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the tree species classification method based on laser point cloud data and remote sensing images as described in any of the above embodiments.

[0136] The non-transitory computer-readable medium of this embodiment includes both permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0137] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the tree species classification method based on laser point cloud data and remote sensing images as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0138] Based on the same concept, corresponding to the methods of any of the above embodiments, this application also provides a computer program product, including computer program instructions. When the computer program instructions are run on a computer, the computer causes the computer to execute the tree species classification method based on laser point cloud data and remote sensing images as described in any of the above embodiments, which has the beneficial effects of the corresponding method embodiments, and will not be repeated here.

[0139] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.

[0140] Furthermore, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the apparatus may be shown in block diagram form. This is to prevent the embodiments of this application from being difficult to understand, and it also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully within the understanding of those skilled in the art). In setting forth specific details to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application may be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0141] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description.

[0142] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A tree species classification method based on laser point cloud data and remote sensing imagery, characterized in that, include: Acquire laser point cloud data and long-term remote sensing images of the target area; Structural features of trees are extracted based on the laser point cloud data, and climatic features are extracted based on the remote sensing images; The structural features and phenological features are input into a pre-trained neural network model for tree species classification to obtain the classification results.

2. The tree species classification method based on laser point cloud data and remote sensing imagery according to claim 1, characterized in that, The remote sensing imagery covers at least one complete growth cycle of the target area.

3. The tree species classification method based on laser point cloud data and remote sensing imagery according to claim 1, characterized in that, The structural features include tree height, diameter at breast height (DBH), crown diameter, crown area, crown volume, branch height, trunk volume, and trunk curvature.

4. The tree species classification method based on laser point cloud data and remote sensing imagery according to claim 1, characterized in that, The phenological characteristics include the start of the growing season, the end of the growing season, and the length of the growing season.

5. The tree species classification method based on laser point cloud data and remote sensing imagery according to claim 1, characterized in that, The step of inputting the structural features and phenological features into a pre-trained neural network model for tree species classification includes, prior to: The structural features and the phenological features are standardized.

6. The tree species classification method based on laser point cloud data and remote sensing imagery according to claim 1, characterized in that, The neural network model includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer includes 11 neurons, the first hidden layer includes 100 neurons, the second hidden layer includes 50 neurons, and the number of neurons in the output layer is equal to the number of tree species.

7. The tree species classification method based on laser point cloud data and remote sensing imagery according to claim 1, characterized in that, The neural network model uses a linear rectified function as the activation function, and an adaptive moment estimator optimizer is used to update the parameters. The early stopping mechanism of the neural network model has a patience coefficient of 20 rounds and a maximum number of iterations of 500 rounds.

8. The tree species classification method based on laser point cloud data and remote sensing imagery according to claim 1, characterized in that, The classification results include a tree species classification confusion matrix heatmap and a tree species classification report, which includes precision, recall, and F1 score.

9. A tree species classification device based on laser point cloud data and remote sensing imagery, characterized in that, include: The acquisition module is configured to acquire laser point cloud data and long-term remote sensing images of the target area; The feature extraction module is configured to extract structural features of trees based on the laser point cloud data and extract climatic features based on the remote sensing image. The classification module is configured to input the structural features and the phenological features into a pre-trained neural network model for classification, thereby obtaining the classification results of the tree species.

10. An electronic device, characterized in that, The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the tree species classification method based on laser point cloud data and remote sensing imagery as described in any one of claims 1-8.

Images