Vegetation classification method and system based on hyperspectral remote sensing and laser fusion

Through the fusion and processing of hyperspectral remote sensing and laser point cloud data, feature information is extracted and classification algorithm training is carried out, which solves the problem of single data and occlusion impact in tree species identification and vegetation classification, and achieves efficient and accurate vegetation classification results.

CN118941839APending Publication Date: 2024-11-12GUIZHOU POWER GRID CO LTD
View PDF 0 Cites 4 Cited by

Patent Information

Application Number
CN202410888720.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-04
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

Hyperspectral remote sensing and laser point cloud data fusion technology have problems such as single data, low accuracy, poor occlusion and poor identification accuracy in tree species identification and low data processing efficiency.

Method used

By obtaining hyperspectral remote sensing data and laser point cloud data for preprocessing and fusion registration, feature information is extracted, combined with spectral reflectivity curves and three-dimensional geometric features, training and classification is used using classification algorithms, and classification is improved through accuracy evaluation and parameter optimization Accuracy and reliability of results.

Benefits of technology

It realizes more accurate tree species identification and vegetation classification, improves data processing efficiency, and enhances the performance of the system and the accuracy and reliability of classification results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941839A_ABST
    Figure CN118941839A_ABST
Patent Text Reader

Abstract

The invention belongs to remote sensing science and technology, and the method comprises the steps: obtaining hyperspectral remote sensing data and laser point cloud data, carrying out the preprocessing, carrying out the fusion registration of the preprocessed data, and extracting features used for tree species recognition and vegetation classification from the fusion data; training and classifying the extracted features to identify tree species and classify vegetation, verifying the accuracy of a classification result through ground survey data and high-precision data, and evaluating the precision of the data; and parameters are optimized according to an evaluation result, a classification result is visually displayed, a classification map is generated, and a vegetation index map displays vegetation distribution and tree species types. According to the method, the hyperspectral remote sensing data and the laser point cloud data are fused, the advantages of the two data sources are fully utilized, the classification accuracy and reliability are improved, the data processing efficiency is improved through the optimization algorithm and parallel calculation, the tree species identification and vegetation classification speed is increased, and the working efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to remote sensing science and technology, in particular to a vegetation classification method and system based on hyperspectral remote sensing and laser fusion. Background Art

[0002] In the field of remote sensing, the fusion of hyperspectral remote sensing and laser point cloud data has been widely studied and applied. A common method is to register hyperspectral data with laser point cloud data, and then use classification algorithms to jointly analyze and process the data. This fusion can provide more comprehensive ground object information, thereby improving the accuracy of tree species identification and vegetation classification.

[0003] In practical applications, hyperspectral remote sensing and laser point cloud fusion technology has been applied in the fields of forest resource management, environmental monitoring and urban planning. For example, in forest resource management, this technology can be used for tree species classification, forest growth status assessment and fire monitoring.

[0004] However, there are also some potential defects and challenges in the fusion technology of hyperspectral remote sensing and laser point cloud. Here are some possible problems: steps such as data preprocessing, registration and feature extraction may require a lot of computing resources and expertise; the consistency and accuracy of hyperspectral remote sensing and laser point cloud data are crucial to the accuracy of the fusion results. Problems such as noise, registration errors and ground object occlusion during data acquisition may cause data inconsistency and inaccuracy, thus affecting the classification results; choosing a suitable classification algorithm and adjusting algorithm parameters are key issues in fusion technology. Different algorithms may be suitable for different scenarios and data types, but how to choose the best algorithm and adjust parameters is still a challenge; hyperspectral remote sensing and laser point cloud data usually have a large amount of data, which puts forward requirements for data storage and processing capabilities. At the same time, how to efficiently process large-scale data and obtain accurate classification results in a reasonable time is also a challenge. Summary of the invention

[0005] In view of the above-mentioned existing problems, the present invention is proposed. The present invention utilizes a method of fusing hyperspectral data with laser point cloud to solve the problems of single data, low precision, and occlusion affecting recognition accuracy in traditional tree species identification and vegetation classification, thereby achieving the effect of improving the accuracy and reliability of tree species identification and vegetation classification. By combining the spectral information of hyperspectral data and the spatial information of laser point cloud data, the present invention provides a comprehensive, multi-dimensional data acquisition and processing method that can more accurately extract and identify the characteristics of different tree species and vegetation types. At the same time, the present invention further improves the performance of the system and the accuracy of the classification results through precision evaluation and parameter optimization, thereby achieving efficient and reliable tree species identification and vegetation classification.

[0006] In order to solve the above technical problems, a vegetation classification method based on hyperspectral remote sensing and laser fusion is proposed, including:

[0007] Acquire hyperspectral remote sensing data and laser point cloud data, preprocess the collected data, fuse and register the preprocessed hyperspectral remote sensing and laser point cloud data, and extract features for tree species identification and vegetation classification from the fused data; train and classify the extracted features for tree species identification and vegetation classification, verify the accuracy of the classification results through ground survey data and high-precision data, and evaluate the data accuracy; optimize the parameters according to the evaluation results, visualize the classification results, and generate classification maps and vegetation index maps to display vegetation distribution and tree species types.

[0008] As a preferred scheme of the vegetation classification method based on hyperspectral remote sensing and laser fusion described in the present invention, wherein: the acquisition of hyperspectral remote sensing data and laser point cloud data includes hyperspectral remote sensing data, laser point cloud data and environmental factor data, the hyperspectral remote sensing data is acquired by an unmanned aerial vehicle, the laser point cloud data includes data obtained by scanning the ground by a laser radar, and the environmental factor data includes surrounding humidity, temperature and soil type.

[0009] The preprocessing of the collected data includes first screening the data through quality index evaluation to remove interference factors, performing atmospheric correction, radiation correction and geometric correction on the hyperspectral remote sensing data, atmospheric correction is used to eliminate atmospheric effects, radiation correction is used to convert digital values ​​into surface reflectivity, geometric correction is used to correct the spatial deformation of the image, and denoising, registration and ground extraction are performed on the laser point cloud data. When the SNR value of a pixel point is greater than a threshold, the data quality of the pixel point is considered to be high, and data after noise elimination and geometric distortion correction is obtained. When the SNR value of a pixel point is less than a threshold, the data quality is considered to be low and greatly affected by noise, and the current data is directly removed, wherein the SNR value is a signal-to-noise ratio, and the threshold is obtained according to a prescribed calculation method. The data that has passed the first screening is screened for a second time through feature importance evaluation, and the top 10% of features with the highest importance scores are selected as the final usable data.

[0010] As a preferred solution of the vegetation classification method based on hyperspectral remote sensing and laser fusion described in the present invention, the fusion registration includes data fusion through feature level fusion, decision level fusion, multi-scale fusion and collaborative fusion.

[0011] After fusion, registration is performed to combine the filtered usable data, the spectral information of the hyperspectral data and the spatial information of the laser point cloud data. The registration algorithm is used to keep the data consistent and to correspond the spatial positions of the hyperspectral data and the laser point cloud data:

[0012]

[0013] The weighted sum calculation of the characteristic function is expressed as:

[0014] β i (s,t)=cos(β1s+β2t)

[0015] Processing the matching of spatial information and noise suppression, the integral function calculation is expressed as:

[0016]

[0017] Where I represents the fusion registration function, represents hyperspectral data, represents the laser point cloud data, Ω represents the spatial domain of data fusion, s and t represent the spatial and temporal variables respectively, σ is the standard deviation of the Gaussian function, m represents the number of characteristic functions, and α i and β i Represent the weight coefficient and characteristic function respectively, β1 and β2 are the parameters of the characteristic function, τ1 and τ2 represent the integral function The domain of , γ represents the information filtering function for matching spatial information and suppressing noise.

[0018] A high I value indicates good registration quality and high data consistency, while a low I value indicates that there are problems with the registration and the registration algorithm needs to be adjusted. Different adjustments are made at different registration accuracies: when the I value is between 0.9 and 1.0, the registration quality is considered to be excellent and the registered data is consistent. No adjustment is required and tree species identification and vegetation classification can be performed directly.

[0019] When the I value is between 0.5 and 0.9, it is considered that the registration quality is flawed and the registered data is general. The parameters of β1, β2, s and t are adjusted. When the average value of the I value reaches 0.9 to 1.0 after three adjustments, the data registration is considered consistent and tree species identification and vegetation classification are allowed. When the average value of the I value does not reach 0.9 to 1.0 after three adjustments or any of the adjusted values ​​is lower than the unadjusted I value, the data consistency is considered insufficient and the registration quality is insufficient to support tree species identification and vegetation classification. The preprocessing is returned to eliminate noise and enhance features before fusion registration.

[0020] When the I value is lower than 0.5, it is considered that the registration quality is poor, the data is completely inconsistent, and cannot be used, so it is discarded.

[0021] As a preferred solution of the vegetation classification method based on hyperspectral remote sensing and laser fusion described in the present invention, the tree species identification and vegetation classification include extracting key features from the fused and registered data for tree species identification and vegetation classification, and extracting spectral features of vegetation using spectral reflectance curves. The laser point cloud data extracts three-dimensional geometric features of tree height, density, and shape, and extracts auxiliary data including terrain and soil type, and uses a classification algorithm to train and classify the extracted key features.

[0022] Based on the auxiliary data, we first determine the possible tree species and vegetation types that may exist at present, form a plant database, and then consider the overlap and occlusion of different plants to determine the tree species. We scan the ground with a lidar to obtain laser point cloud data, form a three-dimensional view, and lock the identified tree species. When there is a small area of ​​occlusion, based on the fused and registered data, we discard the incomplete spatial information in the three-dimensional geometric features, and delete the incomplete data of the spectral reflectance features on the spectral feature map. When there is a large area of ​​occlusion, we perform occlusion detection, perform compensation calculation based on the unobstructed data, and simulate the shape of the obstructed part. At the same time, the drone collects data from multiple directions and forms a shape based on the multi-directional data. Second, compare the accuracy of form one and form two. If the accuracy of the two forms is above 95%, it is considered that any one of form one and form two can be used for tree species identification. When the accuracy is lower than 95%, adjustments are made based on the combination of form one and form two to obtain the final form, and then tree species identification is performed. Different tree species are continuously locked for identification, and then vegetation classification is performed in the current area. Classification calculations are performed based on the trained key features and compared with the data contained in database one. If there is a consistency, the vegetation type of the current area can be confirmed. If there is no consistency, classification calculations are performed based on the tree species and key features of the current area to select the most suitable vegetation type.

[0023] As a preferred solution of the vegetation classification method based on hyperspectral remote sensing and laser fusion described in the present invention, the classification algorithm includes training and classifying the extracted key features using the classification algorithm, comparing the spectral reflectance curve with the filtered Combined with the shape parameter S, the spectral reflectance curve is combined with the geometric features using the integral function to obtain a comprehensive feature value, and the classification algorithm function f class Train the fused features and determine the model parameters:

[0024]

[0025] The trained model is tested on the data to obtain the classification result V class , determine the vegetation type corresponding to each data point:

[0026]

[0027] Where R(λ) represents the reflectivity at different wavelengths in the spectral reflectivity curve, H represents the tree height extracted from the laser point cloud data, D represents the tree density extracted from the laser point cloud data, S represents the tree shape parameters extracted from the laser point cloud data, and T i Represents the terrain parameters of the ith combination, Soil i represents the soil type parameter of the ith combination, λ1 and λ2 represent the starting and ending wavelengths of the spectral reflectance curve, k is the influencing factor of the shape parameter, n is the number of combinations of terrain and soil type, sig is the sigmoid function, ω j is the feature v j The weight of R, b is the bias term, m is the number of features, min(R) and max(R) are the minimum and maximum values ​​in the data set, respectively, and f class is the classification algorithm function.

[0028] As a preferred solution of the vegetation classification method based on hyperspectral remote sensing and laser fusion described in the present invention, the accuracy evaluation includes using an advanced spectral feature extraction function and using Fourier transform to analyze the frequency components of the spectral signal, which is expressed as:

[0029]

[0030] in, is the Fourier transform of the spectral reflectance curve S(δ), P is the frequency variable, e iP(A,B,C) It is the kernel function of the inverse Fourier transform, (A, B, C) represents the coordinates of a point in space, where A represents the horizontal position, B represents the vertical position, and C represents the third dimension.

[0031] The three-dimensional geometric characteristics of trees analyzed using wavelet transform are expressed as:

[0032]

[0033] Among them, d o,p,q It is the wavelet transform coefficient, which represents the local geometric features at scale o and position (p, q).

[0034] Integrate environmental features and use high-dimensional data integration to calculate auxiliary data on terrain and soil types obtained from hyperspectral images:

[0035]

[0036] in, is the environmental characteristic function, is a weight function used to emphasize important environmental features.

[0037] Use weighted feature fusion method for accuracy evaluation:

[0038]

[0039] Among them, F represents the accuracy comparison value between form 1 and form 2, n represents the number of plant species in the dataset, and R i (λ) represents the spectral reflectance of the i-th plant, R ref (λ) represents the reference spectral reflectance, f geo Represents the three-dimensional geometric characteristic function of the tree, f spec Represents the frequency component function of the spectral signal, f env Correction factor function representing terrain and soil type.

[0040] When F=1, it means that the morphological simulation of the obscured part is credible, and the accuracy of Form 1 and Form 2 is high. Any one of them can be used for tree species identification. When 0<F<1, it is necessary to make adjustments based on the combination of Form 1 and Form 2 to obtain the final form for tree species identification. When F=0, it means that the morphological simulation of the obscured part is not credible, and the morphological simulation is performed again.

[0041] As a preferred solution of the vegetation classification method based on hyperspectral remote sensing and laser fusion described in the present invention, wherein: the optimization of parameters according to the evaluation results includes optimizing parameters according to the accuracy evaluation of the classification results:

[0042]

[0043] By continuously optimizing parameters to update tree species identification and vegetation classification, the value of the loss function is reduced, the accuracy of form one and form two is improved, and the process of constantly correcting form one and form two is reduced. new and θ old Represent the updated parameters and the parameters before the update respectively, ε is the step size DE learning rate that controls the parameter update, N is the number of training samples, d i is the true label of the i-th sample, f j (e i ) is the jth feature extraction function used to extract the input data from e i Extract features from, Z(θ) is the regularization term to prevent overfitting, and ζ is the regularization strength coefficient.

[0044] Another object of the present invention is to provide a vegetation classification system based on hyperspectral remote sensing and laser fusion, which utilizes hyperspectral and laser point cloud fusion technology, combined with environmental factor data, and through modular methods such as data acquisition and preprocessing, data fusion and registration, feature extraction, accuracy evaluation and classification, parameter optimization, and visualization, solves the problems of insufficient accuracy, occlusion, poor data consistency, etc. in tree species identification and vegetation classification caused by a single data source, thereby achieving the effect of improving the accuracy, robustness, and reliability of tree species identification and vegetation classification. Through the present invention, not only can more abundant and accurate data support be provided, but also efficient data processing and model optimization can be achieved, thereby achieving high-precision identification and classification of different tree species and vegetation types in practical applications.

[0045] As a preferred solution of the tree species identification and vegetation classification system integrating hyperspectral and laser point cloud described in the present invention, it is characterized by including a data acquisition and preprocessing module, a data fusion and registration module, a feature extraction module, an accuracy assessment and classification module, a parameter optimization module and a visualization module.

[0046] The data acquisition and preprocessing module collects hyperspectral remote sensing data, laser point cloud data and environmental factor data, performs quality assessment, denoising and geometric correction preprocessing operations on the collected data, provides raw data for subsequent data preprocessing and fusion, improves data quality, and removes invalid and interfering data.

[0047] The data fusion and registration module combines the spectral information of the hyperspectral data and the spatial information of the laser point cloud data to perform data fusion and registration, thereby ensuring the consistency of the hyperspectral data and the laser point cloud data in spatial position and providing accurate basic data for feature extraction.

[0048] The feature extraction module extracts key features for tree species identification and vegetation classification from the fused data, and obtains spectral features and geometric features that can characterize different tree species and vegetation types.

[0049] The accuracy assessment and classification module uses a classification algorithm to train and classify the extracted features, establishes a classification model to effectively identify and classify different tree species and vegetation types, and compares the simulated morphologies and identifies tree species and classifies vegetation through accuracy assessment.

[0050] The parameter optimization module optimizes the system parameters according to the evaluation results, and reduces the classification error and improves the system performance by adjusting the parameters.

[0051] The visualization module visualizes the classification results, generates classification maps and vegetation index maps, and intuitively displays vegetation distribution and tree species types.

[0052] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the method described in the method for tree species identification and vegetation classification by fusion of hyperspectral and laser point cloud are implemented.

[0053] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the method described are used to implement tree species identification and vegetation classification by fusion of hyperspectral and laser point cloud.

[0054] The beneficial effects of the present invention are as follows: the present invention can more accurately identify and classify different tree species and vegetation types, by fusing hyperspectral remote sensing data and laser point cloud data, making full use of the advantages of the two data sources, improving the accuracy and reliability of classification, and by optimizing algorithms and parallel computing, improving the efficiency of data processing, accelerating the speed of tree species identification and vegetation classification, and improving work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0056] Figure 1 An overall flow chart of a vegetation classification method based on hyperspectral remote sensing and laser fusion provided in one embodiment of the present invention.

[0057] Figure 2 A hyperspectral image and a LIDARCHM image of a vegetation classification method based on hyperspectral remote sensing and laser fusion are provided in one embodiment of the present invention.

[0058] Figure 3 A comparison of classification results based on single tree crowns of a vegetation classification method based on hyperspectral remote sensing and laser fusion provided in one embodiment of the present invention.

[0059] Figure 4 The present invention provides various tree height distribution frequencies of a vegetation classification method based on hyperspectral remote sensing and laser fusion according to an embodiment of the present invention.

[0060] Figure 5 An embodiment of the present invention provides a vegetation classification method based on hyperspectral remote sensing and laser fusion, which is an AISA+CHM+SIPI fusion data classification thematic map based on SVM.

[0061] Figure 6A comparison of the number of vegetation within a tree crown in a vegetation classification method based on hyperspectral remote sensing and laser fusion provided in one embodiment of the present invention.

[0062] Figure 7 A system functional architecture diagram of a tree species identification and vegetation classification system that integrates hyperspectral and laser point cloud data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0064] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0065] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an embodiment that is mutually exclusive with other embodiments, either individually or selectively.

[0066] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0067] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0068] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0069] Example 1

[0070] Reference Figure 1 , which is the first embodiment of the present invention, provides a vegetation classification method based on hyperspectral remote sensing and laser fusion, including:

[0071] S1: Acquire hyperspectral remote sensing data and laser point cloud data, preprocess the collected data, fuse and register the preprocessed hyperspectral remote sensing and laser point cloud data, and extract features for tree species identification and vegetation classification from the fused data.

[0072] Furthermore, the acquisition of hyperspectral remote sensing data and laser point cloud data includes hyperspectral remote sensing data and laser point cloud data, the hyperspectral remote sensing data is acquired by an unmanned aerial vehicle, the laser point cloud data includes data obtained by scanning the ground with a laser radar, and the environmental factor data includes ambient humidity, temperature and soil type.

[0073] The preprocessing of the collected data includes first screening the data through quality index evaluation to remove interference factors, and performing atmospheric correction, radiation correction and geometric correction on the hyperspectral remote sensing data. Atmospheric correction is used to eliminate atmospheric effects. Their coordinated use can improve the quality and availability of remote sensing images and ensure that the image data accurately reflects the surface characteristics.

[0074] Radiation correction is used to convert digital values ​​into surface reflectivity, and geometric correction is used to correct the spatial deformation of the image. The laser point cloud data is denoised, registered, and ground extracted. When the SNR value of a pixel is greater than the threshold, the data quality of the pixel is considered to be high, and the data after noise elimination and geometric distortion correction is obtained. When the SNR value of a pixel is less than the threshold, the data quality is considered to be low and greatly affected by noise, and the current data is directly removed. The SNR value is the signal-to-noise ratio. For each pixel, the average reflectivity of the current band is calculated as the signal power, and the noise power is estimated by calculating the standard deviation of the pixel value:

[0075]

[0076] Preferably, the threshold is selected in such a way that, when the number of pixels is above the value a and below the value b, the average value of the value a and the value b is selected as the threshold.

[0077] The preprocessing of the collected data includes first screening the data through quality index evaluation to remove interference factors, performing atmospheric correction, radiation correction and geometric correction on hyperspectral remote sensing data, denoising, registration and ground extraction on laser point cloud data, when the SNR value of the pixel point is greater than the threshold, the data quality of the pixel point is considered to be high, and the data after noise elimination and geometric distortion correction is obtained, when the SNR value of the pixel point is less than the threshold, the data quality is considered to be low and greatly affected by noise, and the current data is directly removed, wherein, wherein, the SNR value is the signal-to-noise ratio, and the threshold is obtained according to the prescribed calculation method, and the data that passes the first screening is screened again through feature importance evaluation, and the top 10% of the features with the highest importance scores are selected as the final usable data. The acquired hyperspectral remote sensing data and laser point cloud data are preprocessed to eliminate noise, correct geometric distortion, etc. Atmospheric correction, radiation correction and geometric correction are performed on hyperspectral remote sensing data to ensure data consistency and accuracy. Denoising, registration and ground extraction are performed on laser point cloud data to obtain accurate ground object height information.

[0078] It should also be explained that data fusion is performed through feature-level fusion, decision-level fusion, multi-scale fusion, and collaborative fusion, as follows: Feature-level fusion The merging of features extracted from hyperspectral remote sensing and laser point cloud data into a unified feature set involves combining the spectral features of the hyperspectral data with the structural features of the laser point cloud data to create a richer and more comprehensive feature description.

[0079] Decision-level fusion uses hyperspectral data and laser point cloud data for classification and recognition respectively, and then fuses their classification results or decisions through voting weighted averaging. Deep learning fusion uses deep learning methods such as convolutional neural networks (CNN) or recurrent neural networks (RNN) to take hyperspectral data and laser point cloud data as network inputs to achieve data fusion and classification through end-to-end training.

[0080] Multi-scale fusion Considering the differences in spatial scales between hyperspectral data and laser point cloud data, a multi-scale fusion method can be developed to effectively integrate information at different scales.

[0081] Collaborative fusion involves the interaction and collaboration between hyperspectral data and laser point cloud data. For example, the terrain information provided by the laser point cloud data can be used to assist the classification of hyperspectral data, or the spectral information of the hyperspectral data can be used to assist the classification of the laser point cloud data.

[0082] These fusion methods can combine the respective advantages of hyperspectral data and laser point cloud data to improve the accuracy and robustness of tree species identification and vegetation classification. The choice of these methods depends on the specific application scenario data characteristics and research objectives.

[0083] Preferably, data fusion makes full use of hyperspectral remote sensing data to provide rich spectral information, which is conducive to distinguishing different vegetation species; laser point cloud data provides three-dimensional structural information and ground elevation information of vegetation, which is helpful for analyzing the spatial structure of vegetation.

[0084] It should also be noted that fusion registration includes combining the filtered usable data, the spectral information of the hyperspectral data and the spatial information of the laser point cloud data, keeping the data consistent through the registration algorithm, and matching the spatial positions of the hyperspectral data and the laser point cloud data:

[0085]

[0086] The weighted sum calculation of the characteristic function is expressed as:

[0087] β i (s,t)=cos(β1s+β2t)

[0088] Processing the matching of spatial information and noise suppression, the integral function calculation is expressed as:

[0089]

[0090] Where I represents the fusion registration function, represents hyperspectral data, represents the laser point cloud data, Ω represents the spatial domain of data fusion, s and t represent the spatial and temporal variables respectively, σ is the standard deviation of the Gaussian function, m represents the number of characteristic functions, and α i and β i Represent the weight coefficient and characteristic function respectively, β1 and β2 are the parameters of the characteristic function, τ1 and τ2 represent the integral function The domain of , γ represents the information filtering function for matching spatial information and suppressing noise.

[0091] A high I value indicates good registration quality and high data consistency, while a low I value indicates that there are problems with the registration and the registration algorithm needs to be adjusted. Different adjustments are made at different registration accuracies: when the I value is between 0.9 and 1.0, the registration quality is considered to be excellent and the registered data is consistent. No adjustment is required and tree species identification and vegetation classification can be performed directly.

[0092] When the I value is between 0.5 and 0.9, it is considered that the registration quality is flawed and the registered data is general. The parameters of β1, β2, s and t are adjusted. When the average value of the I value reaches 0.9 to 1.0 after three adjustments, the data registration is considered consistent and tree species identification and vegetation classification are allowed. When the average value of the I value does not reach 0.9 to 1.0 after three adjustments or any of the adjusted values ​​is lower than the unadjusted I value, the data consistency is considered insufficient and the registration quality is insufficient to support tree species identification and vegetation classification. The preprocessing is returned to eliminate noise and enhance features before fusion registration.

[0093] When the I value is lower than 0.5, it is considered that the registration quality is poor, the data is completely inconsistent, and cannot be used, so it is discarded.

[0094] S2: Train and classify the extracted features for tree species identification and vegetation classification, verify the accuracy of the classification results through ground survey data and high-precision data, and evaluate the accuracy of the data.

[0095] Furthermore, key features are extracted from the fused and registered data for tree species identification and vegetation classification, and spectral features of vegetation are extracted using spectral reflectance curves. The laser point cloud data is used to extract three-dimensional geometric features of tree height, density, and shape, and auxiliary data including terrain and soil type are extracted, and the extracted key features are trained and classified using a classification algorithm.

[0096] Based on the auxiliary data, we first determine the possible tree species and vegetation types that may exist at present, form a plant database, and then consider the overlap and occlusion of different plants to determine the tree species. We scan the ground with a lidar to obtain laser point cloud data, form a three-dimensional view, and lock the identified tree species. When there is a small area of ​​occlusion, based on the fused and registered data, we discard the incomplete spatial information in the three-dimensional geometric features, and delete the incomplete data of the spectral reflectance features on the spectral feature map. When there is a large area of ​​occlusion, we perform occlusion detection, perform compensation calculation based on the unobstructed data, and simulate the shape of the obstructed part. At the same time, the drone collects data from multiple directions and forms a shape based on the multi-directional data. Second, compare the accuracy of form one and form two. If the accuracy of the two forms is above 95%, it is considered that any one of form one and form two can be used for tree species identification. When the accuracy is lower than 95%, adjustments are made based on the combination of form one and form two to obtain the final form, and then tree species identification is performed. Different tree species are continuously locked for identification, and then vegetation classification is performed in the current area. Classification calculations are performed based on the trained key features and compared with the data contained in database one. If there is a consistency, the vegetation type of the current area can be confirmed. If there is no consistency, classification calculations are performed based on the tree species and key features of the current area to select the most suitable vegetation type.

[0097] It should also be noted that the classification algorithm is used to train and classify the extracted key features, and the spectral reflectance curve is compared with the filtered Combined with the shape parameter S, the spectral reflectance curve is combined with the geometric features using the integral function to obtain a comprehensive feature value, and the classification algorithm function f class Train the fused features and determine the model parameters:

[0098]

[0099] The trained model is tested on the data to obtain the classification result V class , determine the vegetation type corresponding to each data point:

[0100]

[0101] Where R(λ(represents the reflectivity at different wavelengths in the spectral reflectivity curve, H represents the tree height extracted from the laser point cloud data, D represents the tree density extracted from the laser point cloud data, S represents the tree shape parameters extracted from the laser point cloud data, and T i Represents the terrain parameters of the ith combination, Soil i represents the soil type parameter of the ith combination, λ1 and λ2 represent the starting and ending wavelengths of the spectral reflectance curve, k is the influencing factor of the shape parameter, n is the number of combinations of terrain and soil type, sig is the sigmoid function, ω j is the feature v j The weight of , b is the bias term, m is the number of features, min(R( and max(R( are the minimum and maximum values ​​in the data set, respectively, and f class is the classification algorithm function.

[0102] It should also be noted that the frequency components of the spectral signal analyzed using Fourier transform are expressed as follows through the advanced spectral feature extraction function:

[0103]

[0104] in, is the Fourier transform of the spectral reflectance curve S(δ(, P is the frequency variable, e iP(A,B,C) It is the kernel function of the inverse Fourier transform, (A, B, C) represents the coordinates of a point in space, where A represents the horizontal position, B represents the vertical position, and C represents the third dimension.

[0105] The three-dimensional geometric characteristics of trees analyzed using wavelet transform are expressed as:

[0106]

[0107] Among them, d o,p,q It is the wavelet transform coefficient, which represents the local geometric features at scale o and position (p, q).

[0108] Integrate environmental features and use high-dimensional data integration to calculate auxiliary data on terrain and soil types obtained from hyperspectral images:

[0109]

[0110] in, is the environmental characteristic function, is a weight function used to emphasize important environmental features.

[0111] Use weighted feature fusion method for accuracy evaluation:

[0112]

[0113] Among them, F represents the accuracy comparison value between form 1 and form 2, n represents the number of plant species in the dataset, and R i (λ) represents the spectral reflectance of the i-th plant, R ref (λ) represents the reference spectral reflectance, f geo Represents the three-dimensional geometric characteristic function of the tree, f spec Represents the frequency component function of the spectral signal, f env Correction factor function representing terrain and soil type;

[0114] When F=1, it means that the morphological simulation of the obscured part is credible, and the accuracy of Form 1 and Form 2 is high. Any one of them can be used for tree species identification. When 0<F<1, it is necessary to make adjustments based on the combination of Form 1 and Form 2 to obtain the final form for tree species identification. When F=0, it means that the morphological simulation of the obscured part is not credible, and the morphological simulation is performed again.

[0115] S3: Optimize the parameters according to the evaluation results, visualize the classification results, and generate classification maps and vegetation index maps to display vegetation distribution and tree species types.

[0116] Furthermore, the optimization of parameters includes optimizing parameters based on the accuracy evaluation of the classification results:

[0117]

[0118] By continuously optimizing parameters to update tree species identification and vegetation classification, the value of the loss function is reduced, the accuracy of form one and form two is improved, and the process of constantly correcting form one and form two is reduced. new and θ oldRepresent the updated parameters and the parameters before the update respectively, ε is the step size DE learning rate that controls the parameter update, N is the number of training samples, d i is the true label of the i-th sample, f j (e i ) is the jth feature extraction function used to extract the input data from e i Extract features from, Z(θ) is the regularization term to prevent overfitting, and ζ is the regularization strength coefficient.

[0119] It should be noted that the results of tree species identification and vegetation classification are displayed visually. Classification maps, vegetation index maps, etc. can be generated so that users can intuitively understand vegetation distribution, tree species types and other information.

[0120] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0121] Example 2

[0122] Reference Figure 2-Figure 6 , which is an embodiment of the present invention, provides a vegetation classification method based on hyperspectral remote sensing and laser fusion. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0123] This study aims to verify a new tree species identification and vegetation classification method that combines the fusion of hyperspectral remote sensing data and laser point cloud data. The study area is located in a nature reserve in my country, where there are many types of vegetation and complex terrain, providing an ideal testing environment for this experiment.

[0124] Before the experiment, we first acquired hyperspectral remote sensing data of the study area through drones, and used lidar technology to scan the ground in the same area to obtain laser point cloud data. In addition, we also collected relevant data on the surrounding environment, including humidity, temperature and soil type, in order to have a more comprehensive understanding of the environment in which vegetation grows.

[0125] In the data preprocessing stage, we strictly screened and corrected the hyperspectral remote sensing data and laser point cloud data. The interference factors in the data were removed through quality index evaluation. The hyperspectral remote sensing data was atmospherically corrected, radiometrically corrected, and geometrically corrected, while the laser point cloud data was denoised, registered, and ground extracted. We set an SNR threshold to judge the quality of the data and screened the data accordingly.

[0126] Subsequently, we adopted an innovative fusion registration method that combines the spectral information of hyperspectral data and the spatial information of laser point cloud data. Through the registration algorithm, we maintained the consistency of the data and achieved the spatial position correspondence between the hyperspectral data and the laser point cloud data. In this process, we used the weighted summation calculation of the characteristic function and the integral function calculation to handle the matching of spatial information and noise suppression.

[0127] In the tree species identification and vegetation classification stage, we extracted key features from the fused and registered data, including extracting the spectral features of vegetation using the spectral reflectance curve, and extracting the three-dimensional geometric features of tree height, density, and shape from the laser point cloud data. At the same time, we also considered auxiliary data such as terrain and soil type, and used classification algorithms to train and classify the extracted key features.

[0128] Finally, to evaluate the accuracy of the classification results, we used ground survey data and high-precision data for verification. Based on the verification results, we optimized the parameters and generated classification maps and vegetation index maps through visual display to clearly show the vegetation distribution and tree species types.

[0129] Table 1

[0130]

[0131] In all test subjects, we observed high SNR values ​​for both hyperspectral and laser point cloud data, indicating good data quality for subsequent analysis and classification. The fusion of these two data sources significantly improved the quality and integrity of the data compared to the traditional single data source approach.

[0132] The registration quality index is an important indicator for measuring the quality of fusion of hyperspectral data and laser point cloud data. As can be seen from the table, the registration quality index of all test objects exceeded 0.85, and area 1 even reached an excellent level of 0.95. This means that the effect of data fusion is very good, laying a solid foundation for accurate tree species identification and vegetation classification.

[0133] According to the table data, the tree species identification accuracy of all test objects is above 85%, and the vegetation classification accuracy is above 90%. These results show that the accuracy and reliability of tree species identification and vegetation classification are high, far exceeding the existing single technical methods.

[0134] Figure 2-Figure 7 These are some screenshots of typical tree species identification and vegetation classification technology based on hyperspectral remote sensing and laser point cloud fusion.

[0135] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0136] Example 3

[0137] The third embodiment of the present invention is different from the first two embodiments in that:

[0138] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0139] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0140] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0141] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0142] Example 4

[0143] Reference Figure 7 , which is the fourth embodiment of the present invention, provides a tree species identification and vegetation classification system that integrates hyperspectral and laser point cloud, including a data acquisition and preprocessing module, a data fusion and registration module, a feature extraction module, an accuracy assessment and classification module, a parameter optimization module and a visualization module.

[0144] The data acquisition and preprocessing module collects hyperspectral remote sensing data, laser point cloud data, and environmental factor data, and performs quality assessment, denoising, and geometric correction preprocessing operations on the collected data, providing raw data for subsequent data preprocessing and fusion, improving data quality, and removing invalid and interfering data.

[0145] The data fusion and registration module combines the spectral information of hyperspectral data and the spatial information of laser point cloud data to perform data fusion and registration, ensuring the consistency of the spatial position of hyperspectral data and laser point cloud data, and providing accurate basic data for feature extraction.

[0146] The feature extraction module extracts key features for tree species identification and vegetation classification from the fused data, and obtains spectral features and geometric features that can characterize different tree species and vegetation types.

[0147] The accuracy assessment and classification module uses classification algorithms to train and classify the extracted features, establishes classification models to effectively identify and classify different tree species and vegetation types, and compares the simulated morphologies and performs tree species identification and vegetation classification through accuracy assessment.

[0148] The parameter optimization module optimizes the system parameters according to the evaluation results, and reduces the classification error and improves the system performance by adjusting the parameters.

[0149] The visualization module visualizes the classification results, generates classification maps and vegetation index maps, and intuitively displays vegetation distribution and tree species types.

[0150] Preferably, the tree species identification and vegetation classification system that integrates hyperspectral and laser point cloud realizes efficient and accurate vegetation classification and tree species identification through the interactive work of six key modules. First, the data acquisition and preprocessing module collects hyperspectral remote sensing data, laser point cloud data, and environmental factor data, and performs preprocessing operations such as quality assessment, denoising, and geometric correction on the data to improve data quality and remove invalid and interfering data. The preprocessed data is then passed to the data fusion and registration module, which combines the spectral information of the hyperspectral data and the spatial information of the laser point cloud data to perform data fusion and registration to ensure the consistency of the hyperspectral data and the laser point cloud data in spatial position, and provide accurate basic data for feature extraction.

[0151] The feature extraction module extracts key features for tree species identification and vegetation classification from the fused data, and obtains spectral features and geometric features that can characterize different tree species and vegetation types. The extracted features are sent to the accuracy assessment and classification module, which uses the classification algorithm to train and classify the extracted features and establish a classification model to effectively identify and classify different tree species and vegetation types. At the same time, the simulated morphology is compared through accuracy assessment to verify the accuracy of the classification results.

[0152] The parameter optimization module optimizes the system parameters according to the evaluation results, and by adjusting the parameters, reduces the classification error and improves the system performance. Finally, the visualization module visualizes the classification results, generates classification maps and vegetation index maps, and intuitively displays the vegetation distribution and tree species types. The entire system forms a closed loop through the close interaction of these modules, ensuring that every step from data collection to result display is accurate and efficient, thereby achieving high-precision and high-reliability tree species identification and vegetation classification.

[0153] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. Vegetation classification method based on hyperspectral remote sensing and laser fusion, characterized by: include, Acquire hyperspectral remote sensing data and laser point cloud data, preprocess the collected data, fuse and register the preprocessed hyperspectral remote sensing and laser point cloud data, and extract features for tree species identification and vegetation classification from the fused data; Train and classify the extracted features for tree species identification and vegetation classification, verify the accuracy of the classification results through ground survey data and high-precision data, and evaluate the accuracy of the data; The parameters are optimized based on the evaluation results, the classification results are visualized, and classification maps and vegetation index maps are generated to display vegetation distribution and tree species types.

2. The vegetation classification method based on hyperspectral remote sensing and laser fusion as claimed in claim 1, characterized in that: The acquisition of hyperspectral remote sensing data and laser point cloud data includes hyperspectral remote sensing data, laser point cloud data and environmental factor data, wherein the hyperspectral remote sensing data is acquired by an unmanned aerial vehicle, the laser point cloud data includes data acquired by scanning the ground with a laser radar, and the environmental factor data includes ambient humidity, temperature and soil type; The preprocessing of the collected data includes first screening the data through quality index evaluation to remove interference factors, performing atmospheric correction, radiation correction and geometric correction on the hyperspectral remote sensing data, atmospheric correction is used to eliminate atmospheric effects, radiation correction is used to convert digital values ​​into surface reflectivity, geometric correction is used to correct the spatial deformation of the image, and denoising, registration and ground extraction are performed on the laser point cloud data. When the SNR value of a pixel point is greater than a threshold, the data quality of the pixel point is considered to be high, and data after noise elimination and geometric distortion correction is obtained. When the SNR value of a pixel point is less than a threshold, the data quality is considered to be low and greatly affected by noise, and the current data is directly removed, wherein the SNR value is a signal-to-noise ratio, and the threshold is obtained according to a prescribed calculation method. The data that has passed the first screening is screened for a second time through feature importance evaluation, and the top 10% of features with the highest importance scores are selected as the final usable data.

3. The vegetation classification method based on hyperspectral remote sensing and laser fusion as claimed in claim 2, characterized in that: The fusion registration includes data fusion through feature level fusion, decision level fusion, multi-scale fusion and collaborative fusion; After fusion, registration is performed to combine the filtered usable data, the spectral information of the hyperspectral data, and the spatial information of the laser point cloud data. The registration algorithm is used to keep the data consistent and to correspond the spatial positions of the hyperspectral data and the laser point cloud data: The weighted sum calculation of the characteristic function is expressed as: b i (s,t)=cos(β1s+β2t) Processing the matching of spatial information and noise suppression, the integral function calculation is expressed as: Among them, I represents the fusion registration function, represents hyperspectral data, represents the laser point cloud data, Ω represents the spatial domain of data fusion, s and t represent the spatial and temporal variables respectively, σ is the standard deviation of the Gaussian function, m represents the number of characteristic functions, and α i and β i Represent the weight coefficient and characteristic function respectively, β1 and β2 are the parameters of the characteristic function, τ1 and τ2 represent the integral function The domain of , γ represents the information filtering function for matching spatial information and noise suppression; A high I value indicates good registration quality and high data consistency, while a low I value indicates that there are problems with the registration and the registration algorithm needs to be adjusted. Different adjustments are made at different registration accuracy rates: when the I value is between 0.9 and 1.0, the registration quality is considered to be excellent and the registered data is consistent. No adjustment is required and tree species identification and vegetation classification can be performed directly. When the I value is between 0.5 and 0.9, it is considered that the registration quality is flawed and the registered data is general. The parameters of β1, β2, s and t are adjusted. When the average value of the I value reaches 0.9 to 1.0 after three adjustments, the data registration is considered consistent and tree species identification and vegetation classification are allowed. When the average value of the I value does not reach between 0.9 and 1.0 after three adjustments or any of the adjusted values ​​is lower than the unadjusted I value, the data consistency is considered insufficient and the registration quality is insufficient to support tree species identification and vegetation classification. The preprocessing is returned to eliminate noise and enhance features before fusion registration. When the I value is lower than 0.5, it is considered that the registration quality is poor, the data is completely inconsistent, and cannot be used, so it is discarded.

4. The vegetation classification method based on hyperspectral remote sensing and laser fusion as claimed in claim 3, characterized in that: The tree species identification and vegetation classification includes extracting key features from the fused and registered data for tree species identification and vegetation classification, extracting spectral features of vegetation using spectral reflectance curves, extracting three-dimensional geometric features of tree height, density, and shape from the laser point cloud data, and extracting auxiliary data including terrain and soil type, and using a classification algorithm to train and classify the extracted key features; Based on the auxiliary data, we first determine the possible tree species and vegetation types that may exist at present, form a plant database, and then consider the overlap and occlusion of different plants to determine the tree species. We scan the ground with a lidar to obtain laser point cloud data, form a three-dimensional view, and lock the identified tree species. When there is a small area of ​​occlusion, based on the fused and registered data, we discard the incomplete spatial information in the three-dimensional geometric features, and delete the incomplete data of the spectral reflectance features on the spectral feature map. When there is a large area of ​​occlusion, we perform occlusion detection, perform compensation calculation based on the unobstructed data, and simulate the shape of the obstructed part. At the same time, the drone collects data from multiple directions and forms a shape based on the multi-directional data. State 2, compare the accuracy of form one and form two. If the accuracy of the two forms is above 95%, it is considered that either form one or form two can be used for tree species identification. When the accuracy is lower than 95%, adjustments are made based on the combination of form one and form two to obtain the final form, and then tree species identification is performed. Different tree species are continuously locked for identification, and then vegetation classification is performed in the current area. Classification calculations are performed based on the trained key features and compared with the data contained in the database. If there is a consistency, the vegetation type of the current area can be confirmed. If there is no consistency, classification calculations are performed based on the tree species and key features of the current area to select the most suitable vegetation type.

5. The vegetation classification method based on hyperspectral remote sensing and laser fusion as claimed in claim 4, characterized in that: The classification algorithm includes training and classifying the extracted key features using the classification algorithm, comparing the spectral reflectance curve with the filtered Combined with the shape parameter S, the spectral reflectance curve is combined with the geometric features using the integral function to obtain a comprehensive feature value, and the classification algorithm function f class Train the fused features and determine the model parameters: The trained model is tested on the data to obtain the classification result V class , determine the vegetation type corresponding to each data point: Where R(λ) represents the reflectivity at different wavelengths in the spectral reflectivity curve, H represents the tree height extracted from the laser point cloud data, D represents the tree density extracted from the laser point cloud data, S represents the tree shape parameters extracted from the laser point cloud data, and T i Represents the terrain parameters of the ith combination, Soil i represents the soil type parameter of the ith combination, λ1 and λ2 represent the starting and ending wavelengths of the spectral reflectance curve, k is the influencing factor of the shape parameter, n is the number of combinations of terrain and soil type, sig is the sigmoid function, ω j is the feature v j The weight of R, b is the bias term, m is the number of features, min(R) and max(R) are the minimum and maximum values ​​in the data set, respectively, and f class is the classification algorithm function.

6. The vegetation classification method based on hyperspectral remote sensing and laser fusion as claimed in claim 5, characterized in that: The accuracy assessment includes analyzing the frequency components of the spectral signal using Fourier transform through an advanced spectral feature extraction function expressed as: in, is the Fourier transform of the spectral reflectance curve S(δ), P is the frequency variable, e iP(A,B,C) is the kernel function of the inverse Fourier transform, (A, B, C) represents the coordinates of a point in space, where A represents the horizontal position, B represents the vertical position, and C represents the third dimension; The three-dimensional geometric characteristics of trees analyzed using wavelet transform are expressed as: Among them, d o,p,q is the wavelet transform coefficient, representing the local geometric features at scale o and position (p, q); Integrate environmental features and use high-dimensional data integration to calculate auxiliary data on terrain and soil types obtained from hyperspectral images: in, is the environmental characteristic function, is a weight function used to emphasize important environmental features; Use weighted feature fusion method for accuracy evaluation: Among them, F represents the accuracy comparison value between form 1 and form 2, n represents the number of plant species in the dataset, and R i (λ) represents the spectral reflectance of the i-th plant, R ref (λ) represents the reference spectral reflectance, f geo Represents the three-dimensional geometric characteristic function of the tree, f spec Represents the frequency component function of the spectral signal, f env Correction factor function representing terrain and soil type; When F=1, it means that the morphological simulation of the obscured part is credible, and the accuracy of Form 1 and Form 2 is high. Any one of them can be used for tree species identification. When 0<F<1, it is necessary to make adjustments based on the combination of Form 1 and Form 2 to obtain the final form for tree species identification. When F=0, it means that the morphological simulation of the obscured part is not credible, and the morphological simulation is performed again.

7. The vegetation classification method based on hyperspectral remote sensing and laser fusion as claimed in claim 6, characterized in that: Optimizing the parameters according to the evaluation results includes optimizing the parameters according to the accuracy evaluation of the classification results: By continuously optimizing parameters to update tree species identification and vegetation classification, the value of the loss function is reduced, the accuracy of form one and form two is improved, and the process of constantly correcting form one and form two is reduced. new and θ old Represent the updated parameters and the parameters before the update respectively, ε is the step size DE learning rate that controls the parameter update, N is the number of training samples, d i is the true label of the i-th sample, f j (e i ) is the jth feature extraction function used to extract the input data from e i Extract features from, Z(θ) is the regularization term to prevent overfitting, and ζ is the regularization strength coefficient.

8. A system using the vegetation classification method based on hyperspectral remote sensing and laser fusion as claimed in any one of claims 1 to 7, characterized in that: It includes data acquisition and preprocessing module, data fusion and registration module, feature extraction module, accuracy evaluation and classification module, parameter optimization module and visualization module; The data acquisition and preprocessing module collects hyperspectral remote sensing data, laser point cloud data, and environmental factor data, performs quality assessment, denoising, and geometric correction preprocessing operations on the collected data, provides raw data for subsequent data preprocessing and fusion, improves data quality, and removes invalid and interfering data; The data fusion and registration module combines the spectral information of the hyperspectral data and the spatial information of the laser point cloud data to perform data fusion and registration, thereby ensuring the consistency of the hyperspectral data and the laser point cloud data in spatial position and providing accurate basic data for feature extraction; The feature extraction module extracts key features for tree species identification and vegetation classification from the fused data, and obtains spectral features and geometric features that can characterize different tree species and vegetation types; The accuracy assessment and classification module uses a classification algorithm to train and classify the extracted features, establishes a classification model to effectively identify and classify different tree species and vegetation types, and compares the simulated morphology and identifies tree species and classifies vegetation through accuracy assessment; The parameter optimization module optimizes the system parameters according to the evaluation results, and reduces the classification error and improves the system performance by adjusting the parameters; The visualization module visualizes the classification results, generates classification maps and vegetation index maps, and intuitively displays vegetation distribution and tree species types.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Citation Information

Cited By

  • Positioning mapping system, method and device for drill carriage, electronic equipment and medium

    CN120293122A

  • Forest patrol method and device based on unmanned aerial vehicle, electronic equipment and medium

    CN120631047A

  • Slope rock mass information acquisition method and system based on multi-source remote sensing data fusion

    CN120635718A

  • Tree species identification method and system based on multi-source remote sensing data fusion

    CN120873827A