Plant community structure inversion and ecological restoration strategy generation system and method
By using low-altitude remote sensing from drones and machine learning technology to invert plant community structure and generate precise ecological restoration strategies, the problems of time-consuming, labor-intensive, and insufficient resolution of traditional methods are solved, achieving efficient and low-cost ecological restoration.
Patent Information
- Application Number
- CN202511651002.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-12
- Publication Date
- 2026-02-10
AI Technical Summary
Traditional plant community surveys are time-consuming, labor-intensive, and inefficient. Satellite remote sensing has insufficient resolution and lacks accurate data to support ecological restoration, resulting in restoration plans that are not targeted, costly, and inefficient.
High-resolution vegetation images were acquired using UAV low-altitude remote sensing technology, and plant community structure was inverted by combining machine learning algorithms. Targeted ecological restoration strategies were then generated through expert systems and optimization algorithms.
To improve the efficiency and accuracy of ecological restoration, reduce restoration costs, and promote the rapid recovery and sustainable development of ecosystems.
Smart Images
Figure CN121505441A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ecological monitoring and restoration technology, and in particular to a system and method for inverting plant community structure and generating ecological restoration strategies. Background Technology
[0002] Traditional plant community surveys primarily rely on manual field investigations, a method that is time-consuming, labor-intensive, and inefficient, and difficult to implement in areas with complex terrain and vast areas. The data obtained are usually sampled, failing to comprehensively and continuously reflect the spatial heterogeneity of the community structure across the entire region. This results in a lack of precise data support for subsequent ecological restoration plans, relying heavily on experience and qualitative judgments, leading to strong subjectivity, low restoration efficiency, and high costs.
[0003] In recent years, satellite remote sensing and manned aerial remote sensing have been applied to large-scale ecological and environmental monitoring. However, their spatial resolution is often insufficient to identify species-level information, and they are limited by factors such as cloud cover and revisit cycles, making it difficult to meet the needs of refined ecological restoration. When formulating ecological restoration strategies, traditional methods often lack specificity due to a lack of precise understanding of the current state of the ecosystem. Commonly used general restoration schemes do not fully consider the characteristics of plant communities in different regions and the specific extent of ecosystem damage, resulting in poor restoration effects, ineffective utilization of invested human, material, and financial resources, and high restoration costs with low efficiency.
[0004] Low-altitude drone remote sensing technology offers advantages such as maneuverability, low cost, and high resolution, providing a new means to acquire detailed vegetation information. Meanwhile, machine learning technology demonstrates powerful capabilities in image recognition and pattern recognition. However, a systematic solution that deeply integrates drone remote sensing, machine learning, and ecological restoration decision-making is currently lacking to achieve full automation and intelligence from data acquisition to restoration strategy generation.
[0005] Therefore, proposing a system and method for inverting plant community structure and generating ecological restoration strategies based on UAV low-altitude remote sensing and machine learning to solve the problems existing in the current technology is an urgent problem to be solved by those skilled in the art. Summary of the Invention
[0006] In view of this, the present invention provides a system and method for inverting plant community structure and generating ecological restoration strategies. It utilizes low-altitude remote sensing technology from unmanned aerial vehicles (UAVs) to rapidly and efficiently acquire high-resolution vegetation images, and combines this with advanced machine learning algorithms for in-depth image analysis, achieving accurate inversion of plant community structure. Based on this, and in conjunction with ecological restoration goals, it uses expert systems and optimization algorithms to generate highly targeted ecological restoration strategies, thereby improving the efficiency and accuracy of ecological restoration, reducing restoration costs, and promoting the effective restoration and sustainable development of ecosystems.
[0007] To achieve the above objectives, the present invention adopts the following technical solution: A system for inverting plant community structure and generating ecological restoration strategies includes, in sequence, a UAV low-altitude remote sensing module, an image processing and feature extraction module, a machine learning inversion module, and an ecological restoration strategy generation module; wherein... The low-altitude remote sensing module for drones is used to fly over target ecological areas and acquire remote sensing images of vegetation; The image processing and feature extraction module is used to preprocess and extract features from the vegetation remote sensing image to obtain the morphological, texture and spectral features of the vegetation. The machine learning inversion module has a built-in pre-trained plant community structure inversion model, which is used to invert the species composition and spatial structure information of the plant community based on the extracted features. The ecological restoration strategy generation module has a built-in expert knowledge base and optimization algorithm. It is used to generate ecological restoration strategy schemes that include the types, quantities and spatial locations of replanted plants based on the inverted species composition and spatial structure information and the preset ecological restoration goals.
[0008] Optionally, the sensors carried by the UAV low-altitude remote sensing module in the aforementioned system include visible light cameras, multispectral cameras, and / or hyperspectral cameras.
[0009] In the aforementioned system, optionally, the preprocessing operations performed by the image processing and feature extraction module include: radiometric calibration, atmospheric correction, and image stitching.
[0010] In the above system, optionally, the plant community structure inversion model in the machine learning inversion module is one or more combinations of Convolutional Neural Network (CNN), Support Vector Machine (SVM), or Random Forest (RF).
[0011] The above-mentioned system, optionally, employs a convolutional neural network architecture of encoder-multi-task decoder for plant community structure inversion model, consisting of a shared feature extractor-encoder and multiple specific task heads-decoders.
[0012] The spatial structure information of the above system optionally includes: plant height, canopy width, density, and distribution pattern.
[0013] In the aforementioned system, optionally, the expert knowledge base in the ecological restoration strategy generation module stores rules on plant symbiotic relationships, soil and climate adaptability; the optimization algorithm is a multi-objective optimization algorithm, used to optimize plant community stability and restoration costs while meeting ecological restoration goals.
[0014] The system described above optionally includes a results output and visualization module, which is used to visualize the inverted plant community structure information and the generated ecological restoration strategy schemes in the form of charts and / or map overlays.
[0015] This invention also provides a method for inverting plant community structure and generating ecological restoration strategies, based on the plant community structure inversion and ecological restoration strategy generation system described in any one of the above claims, comprising the following steps: Vegetation images of the target ecological area are acquired through low-altitude remote sensing using drones; the acquired vegetation images are then preprocessed and feature extracted. Using a pre-trained machine learning model, the species composition and spatial structure information of plant communities are retrieved based on the extracted features; Based on the plant community structure information obtained from the inversion, an expert knowledge base and optimization algorithm are invoked to generate a targeted ecological restoration strategy. The ecological restoration strategy specifies the types, quantities, and specific locations of plants that need to be replanted.
[0016] The present invention also provides a training device for a plant community structure inversion model, comprising: The data acquisition unit is used to acquire a training dataset, which includes multiple sets of UAV remote sensing image patches and corresponding ground truth labels marked by field surveys. The ground truth labels include at least species classification labels, vegetation coverage labels, and plant height labels. The model building unit is used to build the encoder-multi-task decoder convolutional neural network architecture. The training unit is used to train the convolutional neural network by taking the UAV remote sensing image patch as input, the corresponding ground truth label as the supervision target, and adopting a multi-task learning method. The model output unit is used to save the model parameters and generate the trained plant community structure inversion model after training is completed.
[0017] As can be seen from the above technical solution, compared with the prior art, the present invention provides a system and method for inverting plant community structure and generating ecological restoration strategies based on UAV low-altitude remote sensing and machine learning, which has the following beneficial effects: Improving the efficiency of ecological restoration: Low-altitude remote sensing technology using drones can quickly acquire large-area vegetation image data. Compared with traditional manual monitoring, it significantly shortens the data collection time, enabling a comprehensive understanding of the ecosystem status of the target area in a short period of time. This provides a guarantee for the timely formulation and implementation of ecological restoration strategies, thereby effectively improving the overall efficiency of ecological restoration. Improving the accuracy of ecological restoration: Through precise analysis of high-resolution images using machine learning algorithms, plant species can be accurately identified, the spatial structure of plant communities can be analyzed in detail, and key information such as vegetation cover and biomass can be obtained. Based on this precise data, the ecological restoration strategies generated by combining expert systems and optimization algorithms fully consider the actual ecological conditions of the target area, are highly targeted, can better meet the needs of ecosystem restoration, improve vegetation survival rate, and promote the rapid recovery of the ecosystem. Reducing the cost of ecological restoration: Precise ecological restoration strategies avoid the blind spots and waste of resources that may result from traditional general restoration solutions; by rationally determining the types, quantities, and locations of replanted plants, as well as providing scientific maintenance and management recommendations, unnecessary human, material, and financial resources can be effectively reduced, thereby lowering the cost of ecological restoration and improving resource utilization efficiency to maximize the benefits of ecological restoration. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0019] Figure 1 This is a structural block diagram of a plant community structure inversion and ecological restoration strategy generation system disclosed in this invention; Figure 2 This is a flowchart of a method for generating plant community structure inversion and ecological restoration strategies disclosed in this invention; Figure 3 This is a schematic diagram of the plant community structure inversion model disclosed in this invention. Detailed Implementation
[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0021] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0022] See Figure 1 As shown, this invention discloses a system for inverting plant community structure and generating ecological restoration strategies, comprising a UAV low-altitude remote sensing module, an image processing and feature extraction module, a machine learning inversion module, and an ecological restoration strategy generation module connected in sequence; wherein, The UAV low-altitude remote sensing module is used to fly over target ecological areas and acquire high-resolution vegetation remote sensing images. The image processing and feature extraction module is used to preprocess and extract features from vegetation remote sensing images to obtain the morphological, texture and spectral features of vegetation. The machine learning inversion module has a built-in pre-trained plant community structure inversion model, which is used to invert the species composition and spatial structure information of the plant community based on the extracted features. The ecological restoration strategy generation module has a built-in expert knowledge base and optimization algorithm. It is used to generate ecological restoration strategy schemes that include the types, quantities and spatial locations of replanted plants based on the inverted species composition and spatial structure information and the preset ecological restoration goals.
[0023] Furthermore, the sensors carried by the UAV's low-altitude remote sensing module include visible light cameras, multispectral cameras, and / or hyperspectral cameras. Specifically, an electric multi-rotor UAV can be selected as the flight platform, with an endurance of no less than 30 minutes, a flight speed that can be flexibly adjusted between 5-15 meters per second, a maximum flight altitude of 500 meters, and the ability to withstand winds below level 4, ensuring stable flight in complex natural environments. The UAV is equipped with a high-precision GPS positioning module and an inertial navigation system, enabling it to comprehensively cover the target area by photographing according to a preset flight path. The UAV is responsible for automated flight within the target area, carrying high-resolution visible light, multispectral, or hyperspectral sensors to acquire detailed image data of the surface vegetation. This data is then transmitted in real time to the ground control center via a data transmission system, providing raw data support for subsequent image processing and analysis.
[0024] Furthermore, the image processing and feature extraction module performs preprocessing operations including radiometric calibration, atmospheric correction, and image stitching. Specifically, the original image is radiometrically calibrated, atmospherically corrected, and orthorectified, and then stitched together to generate an orthophoto map (DOM) and a digital surface model (DSM) of the region. Subsequently, morphological features (such as texture and shape), spatial features (such as distribution patterns), and spectral features (such as vegetation indices) of vegetation are extracted from the image.
[0025] Furthermore, the plant community structure inversion model in the machine learning inversion module is one or more combinations of Convolutional Neural Network (CNN), Support Vector Machine (SVM), or Random Forest (RF). The machine learning inversion module includes a pre-trained plant community structure inversion model (such as a deep learning CNN). The plant community structure inversion model takes extracted image features as input and outputs the inversion results of the plant community structure, including the identification and classification of major species, and spatial structure parameters such as plant height, canopy width, and density.
[0026] Furthermore, the plant community structure inversion model employs an encoder-multi-task decoder convolutional neural network architecture, consisting of a shared feature extractor (encoder) and multiple task-specific heads (decoders), see [link to relevant documentation]. Figure 3 As shown.
[0027] Plant community structure inversion model structure: (1) Input data format: fixed-size image patches (e.g., 256x256 pixels) cropped from UAV orthophotos (DOM) and digital surface models (DSM).
[0028] Data Channel: Channels 1-3: RGB visible light band (provides color and texture information); Channel 4: Near-infrared band (NIR, which strongly reflects vegetation health and biomass); Channel 5: Digital Elevation Model (DEM) or topographic indices (such as slope and aspect) derived from DSM provide topographic information; Optional channels: Calculated vegetation indices (such as NDVI) can be used as additional input channels;
[0029] Normalization: All input channels are normalized at the pixel level.
[0030] 2) Shared encoder The shared encoder, specifically a deep convolutional neural network, preferably using a ResNet-50 or VGG-16 architecture, with the top fully connected classification layer removed. This encoder takes the aforementioned multi-channel image patch X as input and, through a series of convolutions, batch normalization, ReLU activation, and pooling operations, outputs a high-dimensional, abstract feature map F. Its function is to extract general, high-level features (such as edges, shapes, textures, and spatial context) useful for various tasks from the original multi-band imagery.
[0031] The forward propagation of the shared encoder consists of basic convolutional blocks and residual blocks (when using ResNet). The calculation of the basic convolutional blocks follows the formula: ;in, For batch normalization layers, the convolutional output is standardized to accelerate training and improve stability; To modify the activation function of the linear unit; This is the convolution operation for the l-th layer; This is the output activation value of the (l-1)th layer, and also the input of the lth layer; is the output activation value of the l-th layer, which is the output feature map after processing by the convolution operation, batch normalization (BatchNorm), and ReLU activation function in the l-th layer of the neural network.
[0032] The calculation of residual blocks introduces shortcut links, and its core calculation formula is: ;in, For quick joins, if the input and output dimensions are the same, then F is always equal to the mapping: = If the dimensions are different, then F is a 1×1 convolution used to adjust the dimensions; This is the output activation value of the (l+1)th layer.
[0033] , , , , ; in, This represents the first convolutional layer of layer l. The output (without batch normalization). Indicates to Apply the first batch normalization of layer l The result after processing; Indicates to The result after applying the ReLU activation function is used as the input to the second convolutional layer; This represents the second convolutional layer of layer l+1. The output of , whose input is ,Right now After ReLU activation, it passes through a second convolutional layer. The resulting original convolution output; Indicates to Apply the second batch normalization of layer l+1 The output of this structure is effective for training deep networks.
[0034] Furthermore, through the maximum pooling layer Spatial downsampling of the feature map This indicates the position of the output feature map after the max pooling operation at layer l. The activation value at that location; In the input feature map Above, the pooling window corresponding to the output position (i, j); This indicates that the unpooled feature map of layer (l-1) is in The activation value at that location, where Indicates pooled window The local spatial coordinates; (i, j) represent the spatial coordinates of the output feature map; This represents the channel (depth) dimension of the feature map.
[0035] Multi-task decoder The decoder consists of multiple parallel task heads, each responsible for inverting a specific community structure parameter.
[0036] Task Head 1: Species Classification Head Architecture: After the feature map output by the shared encoder, a 1x1 convolutional layer is connected for channel compression, followed by a transposed convolutional layer for upsampling to restore the resolution to that of the input image.
[0037] Output layer: A convolutional layer with a softmax activation function, the number of output channels is equal to the number of target species categories N.
[0038] Final output: A species classification map. Each pixel location outputs an N-dimensional vector representing the probability that the pixel belongs to each of N preset species. The category with the highest probability is taken as the predicted species for that pixel.
[0039] Task Heads 2 & 3: Crown Width / Density and Plant Height Inversion Heads (Regression Tasks) Architecture: Similar to the classification head, but the structure can be simpler. It typically contains several upsampling layers and convolutional layers.
[0040] Output layer: A single-channel convolutional layer using a sigmoid or linear activation function.
[0041] Final output: Canopy Coverage / Density Map: Each pixel value is between [0, 1], representing the vegetation cover or canopy density at that location.
[0042] Plant height map: Each pixel value is a relative or calibrated absolute height value. It is retrieved by establishing a regression relationship with field measurement data.
[0043] The training and optimization of the plant community structure inversion model are detailed below: (1) Loss function Since it's multi-task learning, the overall loss function is a weighted sum of the losses from each task. The overall loss function L of the multi-task learning model... total Loss of species classification L species Coverage regression loss L coverage and plant height regression loss L height The weighted summation is calculated using the following formula: Total loss L total =λ1*L species + λ2*L coverage + λ3*L height Among them, L species A pixel-level weighted cross-entropy loss is used to address the class imbalance problem; L coverage A smoothed L1 loss is used to enhance regression robustness; L height The root mean square error logarithmic loss is used to make the model focus more on the relative error of plant height. λ1, λ2, λ3: hyperparameters used to balance the importance of different tasks and prevent any one task from dominating the entire training process. The specific definitions of each loss function are as follows: A. Loss of species classification L species Species classification is a typical multi-class classification problem, and the pixel-level weighted cross-entropy loss function is preferred. This function can effectively handle the class imbalance problem commonly found in vegetation images (such as background pixels far exceeding the pixels of some rare species).
[0044] Species taxonomy loss L species Formula form: In the formula, N is the total number of pixels in a batch, and C is the total number of species categories (including the "background" or "non-vegetation" category). The true species label for the i-th pixel (an integer ranging from 1 to C). This is the input data for the i-th pixel. These are the trainable parameters of the neural network model.
[0045] Predict the probability that the i-th pixel belongs to class j for the model.
[0046] For the indicator function, when the true label of the i-th pixel... The value is 1 when it equals j, and 0 otherwise.
[0047] The weight assigned to class j. This weight is typically inversely proportional to the frequency of class j in the training data, thus giving rarer species a higher penalty for misclassification.
[0048] B. Coverage Regression Loss L coverage Vegetation cover is a continuous value ranging from [0, 1], thus this is a regression problem. A smooth L1 loss function is preferred, as it is less sensitive to outliers than mean squared error (MSE) and smoother at zero than mean absolute error (MAE), which is beneficial for training stability.
[0049] Coverage regression loss L coverage Formula form: , , In the formula, N is the total number of pixels in a batch. The actual vegetation cover of the i-th pixel (value between 0 and 1). Let be the vegetation cover predicted by the model for the i-th pixel. β is a hyperparameter that controls the inflection point where the loss function transforms from a quadratic to a linear form. It is typically set to 1.0. For smoothing L1 loss function; C. Plant height regression loss L heigh Plant height is also a continuous value, but it may have no upper limit. To improve the consistency of the model's regression accuracy for both short and tall plants, the root mean square error logarithmic loss is preferred. This loss takes the logarithm of the predicted and true values before calculating the error, making the model focus more on relative error rather than absolute error.
[0050] Plant height regression loss L heigh Formula form: In the formula, N is the total number of pixels in a batch. is the true plant height value of the i-th pixel. The plant height value predicted by the model for the i-th pixel. A very small smoothing constant is used to prevent undefined results during logarithmic calculations (such as when the plant height is 0). This loss is equivalent to... The penalty is the difference between the predicted value and the actual value.
[0051] (2) Training data (1) Source: In multiple typical ecological areas, UAV remote sensing flights and detailed field community surveys were carried out simultaneously.
[0052] (2) Annotation: Species truth map: Pixel-level species labeling of typical areas in the image is performed through manual visual interpretation or quadrat surveys combined with GPS positioning.
[0053] True coverage / crown width map: can be obtained by measurement within quadrats or calculated from high-precision annotations.
[0054] True plant height map: obtained by spatial interpolation of real height data acquired through drone LiDAR or ground-based measured sample point data.
[0055] Post-processing and integration of inversion results (1) After the model makes predictions for each image patch, it stitches all the prediction results back into a complete map of the entire region. Subsequently, using image processing techniques such as connected component analysis, individual plant individuals or patches are identified from the species classification map and canopy map, and then the following statistics are obtained: Species composition: Calculate the pixel area ratio of each species as its abundance in the community.
[0056] (2) Spatial structure: Average height / crown width: Calculate the average height and crown width for each identified individual or patch.
[0057] Density: The number of individuals per unit area.
[0058] Distribution pattern: Analyze the spatial distribution of individuals through spatial point patterns to determine whether it is clustered, random, or uniform.
[0059] Furthermore, spatial structure information includes: plant height, canopy width, density, and distribution pattern.
[0060] Furthermore, the expert knowledge base in the ecological restoration strategy generation module stores plant symbiotic relationships, soil and climate adaptability rules; the optimization algorithm is a multi-objective optimization algorithm, which is used to optimize plant community stability and restoration costs while meeting the ecological restoration goals.
[0061] Specifically, the ecological restoration strategy generation module integrates an ecological expert knowledge base, which stores rules regarding plant physiological and ecological characteristics, interspecific interactions (such as competition and symbiosis), and suitable environments. Combining the current community structure data obtained from the previous module with user-defined restoration goals (such as improving biodiversity and preventing soil erosion), a multi-objective optimization algorithm is used to calculate and ultimately output the optimal ecological restoration strategy. This strategy specifically includes the recommended plant species for replanting or thinning, their precise quantities, and their specific spatial locations.
[0062] Furthermore, the results output and visualization module is used to visualize the inverted plant community structure information and the generated ecological restoration strategies in the form of charts and / or map overlays.
[0063] and Figure 1 Corresponding to the system shown, this invention also discloses a method for inverting plant community structure and generating ecological restoration strategies. For specific steps, please refer to [link to relevant documentation]. Figure 2 As shown: High-resolution vegetation images of the target ecological area are acquired through low-altitude remote sensing using drones; the acquired vegetation images are then preprocessed and feature extracted. Using a pre-trained machine learning model, the species composition and spatial structure information of plant communities are retrieved based on the extracted features; Based on the plant community structure information obtained from the inversion, an expert knowledge base and optimization algorithm are invoked to generate a targeted ecological restoration strategy. The ecological restoration strategy specifies the types, quantities, and specific locations of plants that need to be replanted.
[0064] Another embodiment of the present invention provides a training device for a plant community structure inversion model, comprising: The data acquisition unit is used to acquire a training dataset, which includes multiple sets of UAV remote sensing image patches and corresponding ground truth labels marked by field surveys. The ground truth labels include at least species classification labels, vegetation coverage labels, and plant height labels. The model building unit is used to build the encoder-multi-task decoder convolutional neural network architecture.
[0065] The training unit is used to train the convolutional neural network by taking the UAV remote sensing image patch as input, the corresponding ground truth label as the supervision target, and adopting a multi-task learning method. The model output unit is used to save the model parameters and generate the trained plant community structure inversion model after training is completed.
[0066] For the system or system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and relevant details can be found in the description of the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.
[0067] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A system for inverting plant community structure and generating ecological restoration strategies, characterized in that, It includes, in sequence, a UAV low-altitude remote sensing module, an image processing and feature extraction module, a machine learning inversion module, and an ecological restoration strategy generation module; among which, The low-altitude remote sensing module for drones is used to fly over target ecological areas and acquire remote sensing images of vegetation; The image processing and feature extraction module is used to preprocess and extract features from vegetation remote sensing images to obtain the morphological, texture and spectral features of vegetation. The machine learning inversion module has a built-in pre-trained plant community structure inversion model, which is used to invert the species composition and spatial structure information of the plant community based on the extracted features. The ecological restoration strategy generation module has a built-in expert knowledge base and optimization algorithm. It is used to generate ecological restoration strategy schemes that include the types, quantities and spatial locations of replanted plants based on the inverted species composition and spatial structure information and the preset ecological restoration goals.
2. The plant community structure inversion and ecological restoration strategy generation system according to claim 1, characterized in that, The sensors carried by the UAV low-altitude remote sensing module include visible light cameras, multispectral cameras, and / or hyperspectral cameras.
3. The plant community structure inversion and ecological restoration strategy generation system according to claim 1, characterized in that, The preprocessing operations performed by the image processing and feature extraction module include radiometric calibration, atmospheric correction, and image stitching.
4. The plant community structure inversion and ecological restoration strategy generation system according to claim 1, characterized in that, The plant community structure inversion model in the machine learning inversion module is one or more combinations of Convolutional Neural Network (CNN), Support Vector Machine (SVM), or Random Forest (RF).
5. The plant community structure inversion and ecological restoration strategy generation system according to claim 4, characterized in that, The plant community structure inversion model employs a convolutional neural network architecture of encoder-multi-task decoder, consisting of a shared feature extractor-encoder and multiple specific task heads-decoders.
6. The plant community structure inversion and ecological restoration strategy generation system according to claim 1, characterized in that, Spatial structure information includes: plant height, crown width, density, and distribution pattern.
7. The plant community structure inversion and ecological restoration strategy generation system according to claim 1, characterized in that, The expert knowledge base in the ecological restoration strategy generation module stores plant symbiotic relationships, soil and climate adaptability rules; the optimization algorithm is a multi-objective optimization algorithm, which is used to optimize plant community stability and restoration cost while meeting the ecological restoration goals.
8. The plant community structure inversion and ecological restoration strategy generation system according to claim 1, characterized in that, Also includes: The results output and visualization module is used to visualize the inverted plant community structure information and the generated ecological restoration strategies in the form of charts and / or map overlays.
9. A method for inverting plant community structure and generating ecological restoration strategies, characterized in that, A plant community structure inversion and ecological restoration strategy generation system based on any one of claims 1-8 includes the following steps: Vegetation images of the target ecological area are acquired through low-altitude remote sensing using drones; the acquired vegetation images are then preprocessed and feature extracted. Using a pre-trained machine learning model, the species composition and spatial structure information of plant communities are retrieved based on the extracted features; Based on the plant community structure information obtained from the inversion, an expert knowledge base and optimization algorithm are invoked to generate a targeted ecological restoration strategy. The ecological restoration strategy specifies the types, quantities, and specific locations of plants that need to be replanted.
10. A training device for a plant community structure inversion model, characterized in that, include: The data acquisition unit is used to acquire a training dataset, which includes multiple sets of UAV remote sensing image patches and corresponding ground truth labels marked by field surveys. The ground truth labels include at least species classification labels, vegetation coverage labels, and plant height labels. A model building unit is used to build the encoder-multi-task decoder convolutional neural network architecture as described in claim 5; The training unit is used to train the convolutional neural network by taking the UAV remote sensing image patch as input, the corresponding ground truth label as the supervision target, and adopting a multi-task learning method. The model output unit is used to save the model parameters and generate the trained plant community structure inversion model after training is completed.