A tree species identification method, device and equipment based on dense time sequence images

Abstract

The embodiment of the application provides a tree species identification method, device and equipment based on dense time sequence images, wherein the method comprises the following steps: acquiring dense long-time sequence multispectral remote sensing images and terrain data of a study area; image band fusion is performed according to the dense long-time sequence multispectral remote sensing images to construct 12 vegetation spectral indexes; land use classification is performed on features in the dense long-time sequence multispectral remote sensing images of the study area by using a linear spectral mixture model combined with terrain data, and the features are classified into building, water body and other features and forest land data sets, and pre-classification is performed based on the extracted forest land data sets; the RVI threshold method combined with the linear spectral mixture model is used to divide the forest land data sets into evergreen forest and deciduous forest; Savitzaky-Golay filtering transformation and spectral differential transformation are performed on the forest land data sets to obtain an NDVI transformed image set; and the classification effect of the tree species of the study area is discussed through a random forest model after the vegetation spectral index NDVI transformed image set is used as classification features and combined.

Description

Technical Field

[0001] This application relates to the field of forest resource survey technology, and in particular to a method, apparatus and equipment for tree species identification based on dense temporal images. Background Technology

[0002] Tree species information is a key parameter of interest to ecologists and forest managers. Accurately obtaining tree species distribution information in forest landscapes is crucial for ecological conservation, resource surveys, and environmental monitoring. Tree species composition is an indicator of biodiversity, and information on forest spatial distribution and composition is essential for biodiversity assessment and monitoring, as well as the maintenance and protection of forest ecosystems. Different tree species have different growth rates, timber densities, and biomass allocation patterns, which directly affect their carbon storage capacity. Accurate tree species identification allows for more precise estimation of regional forest carbon storage and the quantity of different types of forest resources.

[0003] Traditionally, forest resource inventory relies on statistical methods and field-based forest land inventory, which are costly, time-consuming, and spatially limited, making them difficult to implement in a short period. Currently, there is limited research on large-scale tree species classification. Remote sensing data, characterized by large-area and long-term monitoring, can capture forest type composition and forest structure information over large areas and inaccessible regions using multi-band and multi-mode sensors, compared to traditional fieldwork. It has been widely applied in forest vegetation monitoring and other fields. Many studies focus on the use of high-resolution or ultra-high-resolution images, but these datasets are expensive, have low temporal resolution, and are difficult to use for large-scale tree species identification. Furthermore, tree species identification in non-pure forest areas is more susceptible to the influence of farmland, buildings, water bodies, and other land features, making classification more difficult. Severe land feature mixing is a major reason for the current low accuracy in tree species identification.

[0004] Furthermore, the spectral characteristics of different forest tree species are similar, making tree species identification susceptible to the influence of forest stand structure and environmental background. Current research methods using hyperspectral images for tree species classification are prone to the phenomenon of "different species with the same spectrum." Even if there are phenological differences in the spectral curves of different dominant tree species, these phenological differences are not reflected in the temporal shift differences of the pixel spectral curves in long-term images, especially since the overall phenological curves of various dominant tree species in broad-leaved forests are similar, making them difficult to distinguish. How to expand the phenological differences among various tree species remains an urgent problem to be solved in tree species identification. Summary of the Invention

[0005] The embodiments of this application provide a tree species identification method, apparatus, and device based on dense temporal images. By incorporating temporal dynamic features into dense temporal images for large-scale tree species identification, the phenological differences between various tree species can be amplified, thereby improving the accuracy of forest tree species identification and achieving rapid and effective monitoring of tree species diversity.

[0006] Therefore, the embodiments of this application provide the following technical solutions:

[0007] In a first aspect, embodiments of this application provide a tree species identification method based on dense time-series imagery, comprising: acquiring dense long-term multispectral remote sensing images and topographic data of a study area; the dense long-term multispectral remote sensing images include all available remote sensing forest images with cloud cover less than 20%; constructing a vegetation spectral index by performing image band fusion based on the dense long-term multispectral remote sensing images; the vegetation spectral index is used to indicate the annual dynamic change characteristics of vegetation growth; and classifying land features in the dense long-term multispectral remote sensing images of the study area into land use categories such as buildings, water bodies, and others using a linear spectral mixture model combined with topographic data. The study collected land cover and forest datasets, and pre-classified the extracted forest datasets. The RVI thresholding method combined with a linear spectral mixture model was used to divide the forest datasets into evergreen forests and deciduous forests. Filtering and spectral differential transformations were performed on the forest datasets to obtain NDVI transformed image sets. These NDVI transformed image sets indicated the time-series-based remote sensing phenological characteristics of the plants, which reflected the growth and development status and cyclical changes of the plants in different seasons. The tree species in the study area were classified using a random forest model after combining the vegetation spectral index and the NDVI transformed image sets as classification features.

[0008] In this implementation, traditional forest tree species identification methods rely on ground surveys, which are costly, time-consuming, and have limited spatial coverage. Remote sensing data, on the other hand, enables efficient monitoring over large areas, especially in inaccessible regions, rapidly acquiring vast amounts of imagery. The use of dense temporal imagery allows for continuous monitoring of forest changes, suitable for large-scale and high-frequency tree species identification. Combining dense temporal imagery with vegetation indices (such as RVI and NDVI) better reflects plant growth changes and captures the dynamic characteristics of different tree species in different seasons. This dynamic characteristic amplifies phenological differences between tree species, improving classification accuracy. This scheme utilizes multiple data sources (multispectral remote sensing imagery, Sentinel-1 radar data, and topographic data, etc.) for tree species identification. Through feature fusion and model training, it comprehensively considers various environmental factors, resulting in more accurate tree species classification, particularly in complex forests and non-pure forest areas. Different tree species can exhibit very similar spectral characteristics in certain situations, especially among dominant species in broadleaf forests. By introducing temporal dynamic features and phenological changes, the problem of "different species sharing the same spectrum" can be effectively addressed, improving the accuracy of tree species classification. For example, red-edge vegetation indices such as the Modified Red Edge Simple Ratio Index (MSR) can help identify differences in tree species over time and reduce the impact of spectral mixing. This method can be effectively applied to large areas and is suitable for tree species classification in various forest types and non-pure forest areas (such as farmland, buildings, and water bodies). By combining different remote sensing data, it can overcome the identification difficulties caused by ground cover mixing and improve classification accuracy. Compared with traditional ground survey methods, this remote sensing technology can significantly save time and costs while providing real-time, continuous monitoring data, helping forest managers and ecologists efficiently obtain information on the distribution of tree species over large areas.

[0009] In one feasible implementation, the construction of vegetation spectral indices by image band fusion based on the dense long-term multispectral remote sensing images includes: preprocessing the dense long-term multispectral remote sensing images and topographic data to obtain forest growth images. The preprocessing includes image registration, resampling, time series and cloud cover filtering, radiometric correction, geometric correction, band synthesis, cloud and shadow masking, and image cropping.

[0010] In this implementation, the preprocessing steps ensure that the acquired remote sensing data is of high quality and accuracy, suitable for subsequent tree species identification and forest monitoring tasks. Specifically, preprocessing steps, especially image registration and resampling, ensure the temporal and spatial consistency of image data from different times and sources, providing a reliable foundation for subsequent analysis. Through steps such as cloud cover filtering, radiometric correction, and geometric correction, noise in the images is reduced, spectral differences between different tree species are enhanced, and the accuracy of tree species identification is improved. Cloud and shadow masking, along with other preprocessing steps, ensure that remote sensing data in complex environments (such as cloud cover, different terrains, etc.) still possesses good analytical capabilities, adapting to different types of forests and land cover classification. Cropping and other preprocessing operations reduce computational load and data redundancy, making subsequent analysis more efficient and accurate, especially in large-scale forest monitoring tasks, enabling the rapid provision of accurate analytical results. This preprocessing scheme ensures the quality of remote sensing data through a series of meticulous steps, from image registration to cropping, each step laying a solid foundation for subsequent forest growth monitoring and tree species identification. These preprocessing steps can effectively improve the accuracy and efficiency of forest monitoring, enabling the technical solution to be widely applied in large-scale, high-frequency forest resource surveys, ecological protection, and tree species identification.

[0011] In one feasible implementation, the resampling includes: resampling the terrain data to a spatial resolution of 10m using a linear sampling method; and resampling the dense long-term multispectral remote sensing image to 10m using bicubic interpolation for bands with a pixel resolution of 20m.

[0012] In this implementation, resampling of topographic data and remote sensing imagery ensures that information from different data sources maintains spatial resolution consistency, facilitating subsequent analysis and data fusion. For example, unifying the resolution of topographic data and remote sensing imagery to 10m effectively avoids spatial matching problems caused by resolution inconsistencies. Bicubic interpolation and linear interpolation are suitable for resampling different types of data; the former is suitable for processing detailed images, while the latter is suitable for simple and efficient processing of topographic data. Combining the two ensures both high accuracy of the resampled data and computational efficiency. For example, using bicubic interpolation, remote sensing imagery retains good edge sharpness and detail even after increasing spatial resolution, making it suitable for tasks requiring precise monitoring of forest changes. Linear interpolation, on the other hand, preserves the overall elevation change trend for topographic data resampling, ensuring no information loss due to excessive smoothing. The resampled high-resolution data better supports various subsequent analysis tasks such as forest growth and tree species identification. High-resolution topographic data helps to more accurately simulate the impact of topography on the forest ecosystem, while high-resolution remote sensing imagery provides more detail, aiding in more accurate forest resource assessment. Resampling methods are simple and efficient, especially linear interpolation and bicubic interpolation, which not only ensure data quality but also offer advantages in terms of time and computation for monitoring large-scale forest areas. This resampling technique, by combining linear and bicubic interpolation, can improve spatial resolution while maintaining the accuracy and detail of topographic and remote sensing image data. This method ensures data quality and improves the accuracy of subsequent analysis, making it particularly suitable for tasks such as forest monitoring and resource assessment.

[0013] As one feasible implementation, the method of using a linear spectral mixture model combined with topographic data to classify land use features in dense long-term multispectral remote sensing images of the study area includes:

[0014] A linear spectral blending model refers to the spectral reflectance of a pixel in a certain wavelength band. It is a linear combination of the reflectances of the components that make up the pixel, weighted by their area proportions, and is expressed as:

[0015]

[0016] Where: R iλ Let r be the spectral reflectance of the i-th pixel in the λ band; kλ f is the spectral reflectance of the k-th basic component in the u-th band; ki ξ is the abundance of the k-th endpoint element in the i-th pixel; iλ is the residual value, representing the multiple reflections and transmissions of light between pixel components, exhibiting a non-linear mixing effect; n is the number of terminal elements, and m represents the number of basic components constituting the pixel.

[0017] In this implementation, in remote sensing imagery, especially in forest, urban, and agricultural areas, a single pixel typically contains mixed spectra of multiple land cover types. Linear spectral mixture models, by considering the linear combination of spectral reflectance, can effectively handle this mixing effect, thereby accurately separating and extracting the contributions of different land cover components. Traditional remote sensing image classification methods (such as maximum likelihood and support vector machines) usually rely on single spectral features, while linear spectral mixture models can utilize multiple spectral bands of each pixel, combined with the proportions of different land cover components, to obtain more accurate classification results. By obtaining the abundance information of each component, similar land cover types can be better distinguished, improving classification accuracy. Linear spectral mixture models can be analyzed without strict prior classification information. By inverting the component abundance in pixels, the model can adaptively identify and classify land cover types, exhibiting strong adaptability and being particularly suitable for large-scale, multi-temporal datasets. For forest areas, linear spectral mixture models can effectively distinguish land cover types such as forest cover, bare soil, and water bodies, helping to monitor key indicators such as forest growth and forest cover change. Identifying the abundance of different components allows for a more detailed understanding of forest distribution, supporting more precise forest resource management. For dense, long-term multispectral remote sensing data, linear spectral mixing models can process images acquired at different time points, helping to analyze the spatiotemporal changes in land use. Changes in abundance values ​​can be used to monitor dynamic changes in land use. This model considers residual terms, effectively handling nonlinear effects caused by land cover mixing, reducing errors caused by spectral mixing, and improving reconstruction accuracy. The introduction of residual terms makes the model more adaptable to pixels against complex terrain backgrounds.

[0018] In one feasible implementation, the vegetation spectral index includes the ratio vegetation index (RVI); the pre-classification based on the extracted forest land dataset, dividing the forest land dataset into evergreen forest and deciduous forest, includes:

[0019] The RVI values ​​of sample points from winter and summer images of deciduous forests were randomly selected from field survey data and their ratios were processed. The minimum value of the relatively stable RVI ratio between the two images was used as the separation threshold.

[0020]

[0021] Among them, RVI S RVI serves as a reference sample for the separation threshold. summer RVI is the RVI value of summer image pixels. winter RVI values ​​for winter image pixels;

[0022] Obtain a reference sample RVI for deciduous forests s The mean M RVIand standard deviation D RVI ;

[0023] Based on the mean M RVI and standard deviation D RVI and multiple reference samples RVI s Plot the frequency histogram to obtain the reference sample RVI that follows a normal distribution. S Frequency and RVI S Normal distribution curve;

[0024] According to RVI S Frequency and RVI S The normal distribution curve is used to set the discrimination rule, and the deciduous forest and evergreen forest are determined by the discrimination rule.

[0025] In this embodiment, the technical solution effectively utilizes seasonal differences by leveraging the different spectral characteristics of deciduous forests in winter and summer. The spectral reflectance characteristics of vegetation differ significantly across seasons, especially the RVI values ​​of deciduous forests, which change dramatically between winter and summer. This variation allows for accurate differentiation of different types of forest land. The method automatically sets discrimination rules based on statistical analysis and normal distribution curves, making the classification process more objective, flexible, and adaptive. Using the mean and standard deviation to set discrimination rules can handle variations and noise in sample data, thereby improving classification accuracy. This technical solution classifies forest land based on the vegetation spectral index (RVI) using seasonal changes and sets automatic discrimination rules through statistical analysis, achieving efficient differentiation between evergreen and deciduous forests. Its advantages lie in fully utilizing seasonal differences, improving classification accuracy, and possessing a high degree of automation and adaptability, enabling rapid processing of large-scale remote sensing data.

[0026] In one feasible implementation, the vegetation spectral index includes time-series data of the Normalized Difference Vegetation Index (NDVI); the step of performing filtering and spectral differential transformation based on the forest dataset to obtain the NDVI transformed image set includes: based on the phenological information of the dominant tree species in the study area, using the Savitaky-Golay algorithm to perform polynomial filtering transformation on the NDVI time-series data to obtain smoothed and denoised NDVI time-series spectral curves of the dominant tree species; the phenological information includes seasonal characteristic information of the dominant tree species, which indicates the canopy changes during the bud opening period, peak leaf expansion period, peak growth period, full leaf color change period, and end of leaf fall for each tree species; and based on the first-order differential transformation and the second-order differential transformation, obtaining the NDVI time-series spectral curves of the dominant tree species after extracting different spectral parameters.

[0027] In this implementation, the Savitzky-Golay filtering algorithm effectively denoises and smooths time-series data, reducing the noise impact caused by external factors (such as climate change and observation errors), making the data more accurately reflect the actual growth status of vegetation. Introducing tree species phenological information into NDVI time-series data analysis provides a time reference based on the tree species' growth cycle. This information helps improve the accuracy of time-series data analysis, especially enabling precise analysis and discrimination within different growth stages of different tree species. Through first- and second-order differential transformations, vegetation growth changes can be analyzed in depth, including growth rate and acceleration of change. This not only helps reveal subtle changes during vegetation growth but also provides support for subsequent vegetation growth trend prediction and ecological monitoring. By extracting NDVI time-series curves of different tree species, the growth characteristics of different tree species in the same region can be effectively distinguished, especially in complex ecological environments, where the growth differences between tree species can be more clearly revealed through this scheme.

[0028] As one feasible implementation, the polynomial filtering transformation of NDVI time-series data using the Savitaky-Golay algorithm includes:

[0029] For NDVI time series data Y(t), (t=0,1…,Y N-1 Where N is a positive integer, for its 2k+1 neighboring data points (where k is a positive integer), fit an l-th order polynomial within the local region:

[0030] σ(t)=a0+a1t+a2t 2 +…+a l t l

[0031] In the formula, σ(t) is the fitted polynomial; a l (l = 0, 1, ..., l) are the coefficients of the polynomial;

[0032] The polynomial coefficients a are obtained by fitting local polynomial least squares method. l This minimizes the squared error between the fitted polynomial and the original data within the local neighborhood:

[0033]

[0034] In the formula, E is the squared error, and C j These are the weighting coefficients, Y(t+j) is the actual observed value at time point t+j, and σ(t+j) is the actual fitted value at time point t+j calculated using polynomial fitting.

[0035] Through the above polynomial fitting process, the smoothed data points Y can be obtained. (t) These smoothed data points Y(t) That is, the data after filtering and denoising, to obtain the NDVI time-series spectral curve of the dominant tree species after SG filtering transformation.

[0036] In this implementation, the smoothed data is more accurate for subsequent analysis and modeling. For example, when using NDVI data to construct vegetation growth models or predict ecological environment changes, the denoised data provides more reliable input and reduces errors caused by data noise. Specifically, the Savitky-Golay algorithm minimizes the squared error through polynomial fitting and least squares method, providing powerful denoising and smoothing capabilities for NDVI time-series data. This technique not only preserves the main trends of the vegetation growth cycle but also effectively removes high-frequency noise from the data, greatly improving the accuracy and analyzability of the data. Its flexibility, adaptability, and efficient computation give it significant advantages in remote sensing data processing, vegetation monitoring, and ecological environment research.

[0037] As one feasible implementation, obtaining the NDVI time-series spectral curves of the dominant tree species after extracting different spectral parameters based on first-order and second-order differential transforms includes:

[0038] The formulas for calculating the first-order and second-order differential transforms are as follows:

[0039]

[0040] In the formula, R(λ) i R(λ) is the spectral reflectance value at band i. i )′,R(λ i )″ represent the first and second order spectral values ​​between bands i and i+1, respectively, and Δλ is the step size between adjacent bands.

[0041] In this implementation, by performing first- and second-order differential transformations on NDVI time-series spectral data, different spectral variation characteristics can be extracted, which is of great significance for the identification and classification of dominant tree species. The first-order differential primarily reflects rapid changes between bands. For example, in the NDVI vegetation index curve, the first-order differential can reveal abrupt changes in vegetation growth, such as changes during rapid growth or decline phases. The second-order differential reflects the smoothness and curvature of the spectral curve. When analyzing vegetation growth, the second-order differential can effectively distinguish the changing trends of different vegetation types, especially during periods of slow or stable growth, and can better capture detailed changes, particularly during periods of relatively mild seasonal changes. Through first- and second-order differential transformations, the sensitivity to changes in vegetation growth can be improved, especially the ability to capture subtle changes in vegetation indices (such as NDVI), thereby identifying the growth status of tree species. For example, the first-order differential transformation can reveal rapid changes in vegetation during spring recovery or autumn decline, while the second-order differential transformation can capture more subtle changes, such as changes in the curvature of vegetation growth. Different tree species exhibit varying NDVI time-series spectral curves during their growth process. First- and second-order differential transforms can extract richer spectral features, such as the changing trends of vegetation reflectance and the fluctuations of the spectral curve. This information allows for further differentiation of the spectral curves of different tree species, supporting remote sensing identification and monitoring. Performing first- and second-order differential processing on spectral data yields more feature information, improving the accuracy of remote sensing data analysis. In studies such as vegetation growth cycle analysis, forest health monitoring, and soil moisture analysis, differential transforms help analysts more accurately identify change points and key features, leading to more precise results. Since the essence of differential transforms is to measure the rate of change, it can effectively reduce the influence of certain low-frequency noise in the original data, especially for parts with small reflectance changes. Differential transforms can amplify these small changes while suppressing meaningless low-frequency fluctuations, thus reducing noise interference.

[0042] Spectral data processing methods based on first- and second-order differential transforms can accurately reveal important information such as vegetation growth status, tree species characteristics, and environmental changes by extracting the rate of change between different spectral bands. This method has significant advantages in remote sensing data processing, effectively improving the accuracy of data analysis and enhancing the monitoring capabilities for vegetation growth and the ecological environment.

[0043] In one feasible implementation, the method further includes: verifying the accuracy of the tree species identification result, including:

[0044] The tree species identification results were evaluated from two aspects: overall accuracy and kappa coefficient.

[0045] The overall accuracy evaluation involves comparing the pixel-level tree species identification results of long-term remote sensing image data with ground truth values ​​to assess the accuracy of the classification results. The formula for calculating the overall accuracy is as follows:

[0046] Overall accuracy = (TP + TN) / (TP + TN + FP + FN)

[0047] TP represents the number of samples that are truly positive and correctly predicted as positive, TN represents the number of samples that are truly negative and correctly predicted as negative, FP represents the number of samples that are truly negative but incorrectly predicted as positive, and FN represents the number of samples that are truly positive but incorrectly predicted as negative.

[0048] The Kappa coefficient is a statistical indicator used to evaluate the consistency between classifiers or evaluators.

[0049] The formula for calculating the Kappa coefficient is as follows:

[0050]

[0051] Where Po is the sum of the number of correctly classified samples in each class divided by the total number of samples, which is the overall classification accuracy; assuming the number of true samples in each class are a1, a2, ..., a C The predicted number of samples for each class are b1, b2, ..., b C If the total number of samples is n, then:

[0052]

[0053] The kappa calculation result is -1 to 1, but usually the kappa falls between 0 and 1. It can be divided into five groups to represent different levels of consistency: 0.0 to 0.20 very low consistency (slight), 0.20 to 0.40 fair consistency (fair), 0.40 to 0.60 moderate consistency (moderate), 0.60 to 0.80 substantial consistency (substantial), and 0.80 to 1 almost perfect consistency (almost perfect).

[0054] In this implementation, overall accuracy provides an intuitive percentage, indicating the proportion of correctly classified samples out of all samples, while the Kappa coefficient considers the consistency between the classifier and the actual situation, statistically evaluating the classifier's performance and avoiding misleading results due to uneven data distribution. Using both overall accuracy and the Kappa coefficient to evaluate the accuracy of tree species identification results ensures the reliability and effectiveness of remote sensing image data classification. Through the comprehensive evaluation of overall accuracy and the Kappa coefficient, this technical solution provides a comprehensive and quantitative method for verifying the accuracy of tree species identification results. This method not only accurately evaluates the performance of the classifier but also quantifies the accuracy and consistency of the classification results, which is of great significance for tree species identification in multi-temporal remote sensing image analysis. Furthermore, it provides clear direction and basis for subsequent model optimization and data analysis, demonstrating strong applicability and practical value.

[0055] Secondly, this application also provides a tree species identification device based on dense time-series imagery, comprising: an acquisition module for acquiring dense long-term multispectral remote sensing images and topographic data of a study area; the dense long-term multispectral remote sensing images include all available remote sensing forest images with cloud cover less than 20%; a processing module for constructing a vegetation index by performing image band fusion based on the dense long-term multispectral remote sensing images; the vegetation spectral index is used to indicate the annual dynamic change characteristics of vegetation growth; and a linear spectral mixture model is used to classify land features in the dense long-term multispectral remote sensing images of the study area into land use categories such as buildings and water features. The study collected data on vegetation bodies and other land features, as well as forest land. Based on the extracted forest land dataset, pre-classification was performed. The forest land dataset was divided into evergreen forest and deciduous forest using the RVI thresholding method combined with a linear spectral mixture model. Savitzaky-Golay filtering and spectral differential transformation were then performed on the forest land dataset to obtain an NDVI transformed image set. The NDVI transformed image set indicated the time-series-based remote sensing phenological characteristics of the plants, which indicated the growth and development status and cyclical changes of the plants in different seasons. Based on the vegetation spectral index and the NDVI transformed image set, the tree species in the study area were classified using a random forest model.

[0056] Thirdly, embodiments of this application also provide a computing device, including at least one memory for storing a program; at least one processor for executing the program stored in the memory; wherein the memory is coupled to the processor, and when the program stored in the memory is executed, the processor is used to execute the method as described in the first aspect or any possible implementation of the first aspect.

[0057] Fourthly, embodiments of this application also provide a computer-readable storage medium storing computer instructions that, when executed by a computing device, cause the computing device to perform the method involved in the first aspect and its possible implementations.

[0058] Fifthly, embodiments of this application also provide a computer program product including computer instructions that, when executed by a computing device, cause the computing device to perform the methods described in the first aspect and its possible implementations.

[0059] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0060] This application's technical solution integrates multispectral remote sensing imagery, Sentinel-1 radar data, topographic data, and temporal variation features, combined with machine learning models (such as random forests), to achieve high-precision tree species identification. The core advantage of this method lies in its comprehensive utilization of multi-dimensional data, enabling precise capture of the temporal variation characteristics of vegetation. For example, by incorporating temporal dynamic features into dense temporal imagery for large-scale tree species identification, it can amplify the phenological differences between various tree species, thereby improving the accuracy of forest tree species identification and achieving rapid and effective monitoring of tree species diversity. Furthermore, by automating tree species classification, it is suitable for remote sensing monitoring and vegetation resource assessment in large-scale areas. Attached Figure Description

[0061] Figure 1 A flowchart illustrating the tree species identification method based on dense temporal images provided in this application embodiment;

[0062] Figure 2 The RVI based on the reference sample is shown. s A schematic diagram of the rules for distinguishing evergreen forests from deciduous forests using frequency histograms;

[0063] Figure 3 The study area's dominant tree species seasonal characteristics are shown.

[0064] Figure 4 The NDVI time-series spectral curves of the dominant tree species after SG filtering transformation are shown.

[0065] Figure 5 The NDVI time-series spectral curves of the dominant tree species after SG filtering and differential transformation are shown.

[0066] Figure 6The radar charts showing the identification accuracy of various dominant tree species based on different feature combinations are shown (SI: spectral index; SG: SG filter transform; FD: first-order differential transform; SD: second-order differential transform).

[0067] Figure 7 A flowchart illustrating the application scenario provided in Embodiment 1 of this application:

[0068] Figure 8 This application illustrates a tree species identification device based on dense temporal images, according to an embodiment of the present application.

[0069] Figure 9 An embodiment of the present application is shown. Detailed Implementation

[0070] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0071] In the description of the embodiments of this application, the words "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the words "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a specific manner.

[0072] In the description of the embodiments in this application, the term "and / or" is merely a description of the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, B existing alone, and A and B existing simultaneously. Furthermore, unless otherwise stated, the term "multiple" means two or more. For example, multiple systems refer to two or more systems, and multiple terminals refer to two or more terminals.

[0073] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and their variations all mean "including but not limited to," unless otherwise specifically emphasized.

[0074] In the description of the embodiments in this application, "some embodiments" are mentioned, which describe a subset of all possible embodiments. However, it is understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0075] In the description of the embodiments of this application, the terms "first, second, third, etc." or module A, module B, module C, etc. are used only to distinguish similar objects and do not represent a specific ordering of objects. It is understood that, where permitted, a specific order or sequence can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0076] In the description of the embodiments of this application, the reference numerals for the steps, such as S101, S102, etc., do not necessarily indicate that the steps will be executed in this manner. Where permissible, the order of the steps can be interchanged or executed simultaneously.

[0077] Quantitative information on forest tree species is crucial for achieving sustainable development and ecological environmental protection. Timely and accurate forest type mapping is an important task in forest resource inventory. Furthermore, tree species classification mapping is significant for terrestrial vegetation carbon storage accounting and carbon trading. Forests play a key role in reducing carbon balance and are an important source of carbon sinks. Methods for estimating carbon storage by establishing individual tree biomass models for different tree species based on factors such as tree diameter or volume are essential for regional carbon storage. Accurate and detailed tree species classification mapping is indispensable for establishing individual tree diameter-at-breast mass models for major tree species.

[0078] Currently, only 12% of the world's primary forests are protected, and many more face severe risks of degradation. Studies show that human activities impact global forest structure, but whether the pressures of human activities have significantly extended to protected forests and primary forests that are generally considered less affected by humans remains unknown. For example, mining operations cause significant damage to existing forest vegetation, while the reclamation of coal mining areas plays a crucial role in rebuilding carbon sinks and improving ecosystems. Clearly defining forest types and tree species information provides a reliable foundation for regional coordinated planning and sustainable protection and utilization of forest resources.

[0079] Rapid and effective monitoring of tree species diversity can promote biodiversity conservation and research, as well as sustainable forest management. Forest biodiversity levels play a crucial role in maintaining and functioning forest ecosystems, and are key factors influencing ecosystem stability, productivity, and carbon storage. Rapid and effective monitoring of tree species diversity is an important aspect of forest biodiversity conservation efforts and is of great significance for promoting biodiversity conservation and research, as well as sustainable forest management. Furthermore, clear spatial information on large-area tree species composition helps in better understanding tree species ecology, such as community dynamics and species contributions to ecosystem functions and services. Other environmental studies, such as wildlife habitat mapping and forest insect abundance estimation, also benefit from tree species information. In addition, knowledge about tree species distribution can also influence logging and government management policies.

[0080] However, existing methods for identifying dominant tree species in regional forests have the following main problems:

[0081] (1) Existing research on tree species extraction mainly relies on hyperspectral and high spatial resolution remote sensing data sources. However, the above data are expensive and costly, and cannot meet the requirements of large-scale and high-precision tree species identification and mapping.

[0082] (2) When extracting tree species over a large area based on time series images, due to the limitation of image spatial resolution and the common pixel mixing phenomenon during the extraction process, the classification process is easily affected by different forest stand types and surrounding background information, resulting in unsatisfactory classification results.

[0083] (3) Most of the current vegetation remote sensing classification and mapping methods rely on field data of a single time phase, but field sampling often requires a lot of manpower and resources. For long-term vegetation classification and dynamic mapping research, reliable historical samples need to be obtained. The existing vegetation classification label samples cannot meet the needs of tree species classification.

[0084] (4) Classify according to the differences in phenological trajectory characteristics of each tree species. The overall spectral characteristics of vegetation have high similarity, but dense time series data can accurately reflect the phenological information of each tree species (branching, leaf unfolding, flowering and leaf fall, etc.), but there is currently a lack of effective spectral transformation methods to highlight the differences in phenological trajectories;

[0085] The aforementioned problems make it difficult for existing natural forest area tree species methods to meet the practical needs of regional forestry resource surveys and monitoring.

[0086] In summary, precise tree species classification is fundamental to forest management planning and disturbance monitoring. Accurate tree species classification provides crucial basic information and scientific basis for local forest resource management. It should be noted that image features are a vital basis for extracting image information, primarily including spectral, spatial, and temporal features. The spatial frequency of an image represents the rate of change of different details within the image, i.e., the degree of change in grayscale values. Different tree species have complex and varied structures and backgrounds, and their texture features lack good stability; therefore, tree species identification based solely on images from a single time period results in low extraction accuracy.

[0087] Therefore, expanding the phenological differences among tree species and exploring the dynamic differences in canopy spectral phenology of different tree species under complex backgrounds are key to tree species identification. This application fully utilizes spectral information from multispectral remote sensing, topographic data, and radar data to explore the impact of spectral filtering and differential variations on the identification results of regional forest tree species, and proposes a tree species identification method, device, and equipment based on dense temporal imagery. Specifically, this application provides a tree species identification method, device, and equipment based on dense temporal imagery. By combining multiple remote sensing data sources (multispectral imagery, Sentinel-1 radar data, topographic data, etc.) and various data processing methods (such as vegetation index, NDVI transformation, linear spectral mixture model, etc.), it can effectively improve the accuracy of tree species identification, especially in large-scale, diverse forest areas. By utilizing temporal dynamic characteristics to expand the phenological differences among tree species, it solves the problem of "different species sharing the same spectrum," demonstrating strong adaptability and higher efficiency and accuracy compared to traditional methods.

[0088] Figure 1 A flowchart illustrating the tree species identification method based on dense temporal images provided in this application embodiment (see also the flowchart framework diagram). Figure 7 ).like Figure 1 As shown, the specific steps are as follows:

[0089] S101. Acquire dense long-term multispectral remote sensing images and topographic data of the study area; the dense long-term multispectral remote sensing images include all available remote sensing forest images with cloud cover of less than 20%.

[0090] Among them, dense long-term multispectral remote sensing images include all bands of the original images, which can be obtained by downloading all available remote sensing satellite images of the study area in the corresponding year (e.g., 2019-2022) from the remote sensing satellite image download platform.

[0091] For example, remote sensing satellite imagery download platforms can include: Google Earth Engine (GEE) platform (https: / / code.earthengine.google.com / ), U.S. Geological Survey (https: / / www.usgs.gov / ), European Space Agency (https: / / scihub.copernicus.eu / ), Geospatial Data Cloud (http: / / www.gscloud.cn / ), etc., and remote sensing satellite imagery can include: Landsat TM / ETM+ / OLI, Sentinel-2MSI, etc.

[0092] In this embodiment, the available bands for dense long-term multispectral remote sensing images, including the original images, specifically include: using the spatial resolution of 10m for blue (B2: 490nm), green (B3: 560nm), red (B4: 665nm), near-infrared (B8: 842nm) and the red edge region (B5: 705nm; B6: 740nm; B7: 783nm); narrow NIR (B8a: 865nm); and SWIR (B11: 1610nm; B12: 2190nm).

[0093] The terrain data used is the NASA ADEM (NASA Digital Elevation Model) dataset. NASA ADEM data is a global digital elevation model (DEM) jointly released by NASA and the U.S. Geological Survey (USGS). It is generated based on NASA satellite data (such as SRTM—the Shuttle Radar Topography Mission) and other satellite measurement data, providing global terrain height information. Specifically, the NASA ADEM dataset is a reprocessing of digital elevation model data generated from Shuttle Radar Topography Mission (SRTM) data, incorporating data primarily from the Ice, Cloud, and Land Elevation Satellite (ICESat), the Earth Science Laser Altimeter System (GLAS), and the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) instrument.

[0094] In some implementations, this step also involves acquiring forest resource data from other sources in the study area, including Sentinel-1 radar data, forest inventory data, and field survey data.

[0095] Sentinel-1 radar data corresponds to Sentinel-1 data with a spatial resolution of 10m for the corresponding year. Forest inventory data can be recent forest land change survey data provided by the local forestry bureau that has passed acceptance inspection. Forest inventory data mainly uses forest plots as survey units, including tree species composition, age structure, tree density, canopy closure, height, diameter at breast height, etc. Forest inventory data is used to gain a preliminary understanding of the vegetation type and distribution in the study area, providing a more accurate survey range for subsequent field surveys. Field survey data is obtained by conducting field investigations under the guidance of technical personnel from the local forestry bureau for areas where tree species are unclear or boundaries are ambiguous in the existing forest inventory data.

[0096] For example, in complex areas where field surveys are difficult to conduct, and in areas with poor network signals that make sampling challenging, drones equipped with multispectral sensors can be used for aerial photography-assisted sampling under the guidance of local forestry bureau technicians to obtain field survey data. Simultaneously, several representative dominant tree species forests (North China larch, Chinese pine, poplar, Liaodong oak, and other species) should be selected as sample areas (at least 40 for each dominant species). The differences in forest age structure (the age distribution of trees), understory cover (the degree of vegetation cover on the ground), and stand density (the number of trees per unit area) should be comprehensively considered, and each training sample area should meet the standard of at least 30*30m in size. Furthermore, a random starting point systematic sampling strategy should be adopted, with average diameter and height measurements taken for trees with a diameter at breast height (DBH) greater than 4m, and tree species recorded. Simultaneously, Simultaneous Localization and Mapping (SLAM) equipment should be used to scan the sample plots to accurately extract tree information and provide strong support for tree species identification.

[0097] S102. Construct vegetation spectral indices by fusing image bands based on dense long-term multispectral remote sensing images; vegetation spectral indices are used to indicate the annual dynamic changes in vegetation growth characteristics.

[0098] In some implementations, step S102 can be achieved through the following steps:

[0099] S1021, forest growth images are obtained by preprocessing dense long-term multispectral remote sensing images and topographic data.

[0100] Preprocessing for dense, long-term multispectral remote sensing images includes: time series and cloud cover filtering, radiometric correction, geometric correction, band synthesis, cloud and shadow masking, and image cropping.

[0101] Cloud cover filtering of acquired dense long-term multispectral remote sensing images is performed to remove cloud-covered portions, as clouds can affect the quality of ground observations and lead to inaccurate data. Cloud detection algorithms (such as the Fmask algorithm) are typically used to detect and remove clouds and their shadows. The selection of a time series helps to obtain more comprehensive information on vegetation dynamics. Radiometric correction converts the raw radiometric values ​​of dense long-term multispectral remote sensing images into usable ground reflectance data. The raw data acquired from dense long-term multispectral remote sensing images are affected by atmospheric conditions, sensor performance, and other factors, therefore radiometric correction is necessary for accurate analysis. Geometric correction is the process of spatially aligning the image to ensure it matches the geographic coordinate system and eliminates spatial errors caused by factors such as Earth curvature, sensor angle, and orbital variations. Data from different bands are synthesized, typically by combining bands of interest based on research needs. The Sentinel-2 satellite has 13 bands, some of which have different spatial resolutions. For example, the pixel resolution of bands B5, B6, B7, B8A, B11, and B12 is 20m, while the resolution of bands B2, B3, B4, and B8 is 10m. Cloud detection results from the image are used to remove cloud and cloud-shadowed areas. This is crucial for subsequent data analysis because clouds and shadows affect the true reflectance information of ground features. The remote sensing image is cropped according to the boundaries of the study area, removing irrelevant regions and retaining only the areas of interest. This reduces computational load and improves processing efficiency.

[0102] It should be noted that dense long-term multispectral remote sensing images (hereinafter referred to as satellite remote sensing images) are time-series images. Due to the different resolutions of different bands of Sentinel-2, in order to unify the data format, it is usually necessary to resample the 20m resolution bands (such as B5, B6, etc.) to 10m resolution. Specifically, the pixel resolution of Sentinel-2 bands B5, B6, B7, B8A, B11, and B12 is 20m. In this embodiment, bicubic interpolation is used to resample these bands to 10m. Bicubic interpolation has an edge enhancement effect and can better preserve the fine structure of the image, but the computational load is very large. Using the GEE platform for calculation can effectively solve the weakness of this method. Next, the B2, B3, B4, and B8 bands and their resampled bands are reprojected onto the EPSG:4326 coordinate system and fused into a new forest growth image.

[0103] For example, the preprocessing platform for satellite remote sensing images can be Google Earth Engine, ENVI software, Sen2Cor software, ArcGIS software, SNAP software, etc.

[0104] For example, the NASADEM dataset undergoes preprocessing, including image registration, radiometric correction, geometric correction, resampling, image mosaicking, and cropping. For instance, the NASADEM dataset, used as topographic data, is resampled to a spatial resolution of 10m using linear sampling, and its slope and aspect are calculated in ArcGIS. Resampling the NASADEM data to 10-meter resolution provides more refined topographic information, suitable for high-precision analysis. Calculating slope and aspect further quantifies topographic features, facilitating land use classification of dense long-term multispectral remote sensing images of the study area in subsequent steps, thus improving classification accuracy.

[0105] It's important to note that slope refers to the degree of inclination of terrain, usually expressed as an angle. Slope is a crucial topographic parameter in environmental studies, land use planning, and landslide prediction. In ArcGIS, the "Slope" tool can be used to calculate slope. This tool calculates the slope value of each pixel based on its elevation value in the Digital Elevation Model (DEM) by analyzing the elevation changes of surrounding pixels. Neighborhood-based calculation methods, such as the 8-neighbor method, are commonly used. Aspect represents the orientation of a terrain surface in the horizontal direction, usually expressed in degrees. In ArcGIS, the "Aspect" tool can be used to calculate aspect. This tool determines the horizontal orientation of each pixel by analyzing its relative elevation information (0° represents north, 90° represents east, 180° represents south, and 270° represents west). Slope and aspect provide additional topographic information for tree species identification, helping to improve the classification accuracy and reliability of the model. These topographic features can reflect the growth environment of trees. Combined with spectral data from remote sensing images, they can better understand the spatial distribution and growth patterns of different tree species, providing more accurate support for tree species identification, ecological assessment, forest management and other tasks.

[0106] For example, preprocessing for drone aerial photography sampling equipped with multispectral sensors includes image registration, radiometric correction, geometric correction, image stitching, and cropping, which helps eliminate the influence of different times and locations and ensure data consistency. If the processing level is Collection2Level-2, it indicates that the product has already undergone basic preprocessing such as geometric correction, radiometric correction, and atmospheric correction, and no further processing is required.

[0107] S1022, based on the different band information of forest growth images, image band calculations are performed to obtain the vegetation spectral index table of tree species.

[0108] In this step, vegetation indices include: normalized vegetation index, improved red-edge normalized vegetation index, ratio vegetation index, red-edge chlorophyll index, enhanced vegetation index, vegetation attenuation index, greenness index, improved red-edge ratio vegetation index, novel inverted red-edge chlorophyll index, modified soil-adjusted vegetation index, ground chlorophyll index, chlorophyll absorption index, etc., as shown in Table 1.

[0109] Table 1. Spectral indices (12) based on Sentinel-2 remote sensing imagery.

[0110]

[0111] As shown in Table 1, the above vegetation indices, used as bands in subsequent classifications, can effectively reflect the annual dynamic changes in crop growth. Specifically, the Normalized Difference Vegetation Index (NDVI) is widely used to reflect vegetation density and health status by using the ratio of red light to near-infrared bands. The Inverted Red-Edge Chlorophyll Index (IRECI) uses the red-edge band (the transition zone between red light and near-infrared) to more accurately describe the growth status of vegetation. The Ratio Vegetation Index (RVI) uses the ratio of red light to near-infrared bands to reflect vegetation growth. The Red-edge Chlorophyll Index (CI) red-edgeNDVI: Calculated in the red-edge band, this vegetation index better adapts to the spectral characteristics of different plants. Enhanced Vegetation Index (EVI): For dense vegetation areas, EVI is more adaptable than NDVI and better reduces atmospheric influences. Plant Senescence Reflectance Index (PSRI): Assesses vegetation degradation by comparing the ratio of red to green light bands. Green Leaf Index (GLI): Reflects the greenness of plants and is suitable for analyzing plant growth status. Soil Adjusted Vegetation Index (SAVI): Adds a soil adjustment coefficient to NDVI to eliminate the influence of soil background. These vegetation indices reflect the dynamic changes of crops and vegetation under different time and environmental conditions, and are particularly suitable for monitoring and analyzing vegetation growth status and land cover types. The Modified Red Edge Simple Ratio Index (MSR) is a vegetation index based on the red edge band, commonly used to enhance sensitivity to vegetation chlorophyll content, nitrogen status, or stress. The Meris terrestrial chlorophyll index (MTCI) is used to estimate the chlorophyll content of the vegetation canopy. This index is sensitive to medium- to high chlorophyll concentrations and is suitable for agricultural, ecological, and forest health monitoring. The Modified Normalized Difference Vegetation Index with Red Edge (mNDVI) is an optimized version of the vegetation index incorporating the red edge band, designed to enhance sensitivity to vegetation chlorophyll content, nitrogen status, and stress, while reducing the saturation problem of traditional NDVI in high biomass areas. The Modified Chlorophyll Absorption in Reflectance Index 2 (MCARI2) is an improved vegetation index used to enhance sensitivity to chlorophyll content while reducing the influence of soil background and canopy structure. It is an optimized version of the original MCARI and is widely used in fields such as agricultural remote sensing and vegetation physiological status monitoring. By calculating these vegetation indices and inputting them as feature variables into subsequent classification models, the accuracy of classification can be effectively improved, especially in the accurate identification and analysis of forest tree species.

[0112] S103. Using a linear spectral mixture model combined with topographic data, land use classification was performed on land features in dense long-term multispectral remote sensing images of the study area. The data were divided into building, water body and other land features, as well as forest land datasets. Based on the extracted forest land datasets, pre-classification was performed, dividing the forest land datasets into evergreen forests and deciduous forests.

[0113] In this step, the Linear Spectral Mixture Model (LSMM) assumes that the spectral reflectance of each pixel in the mixed spectrum is composed of multiple endmembers (basic components) within the pixel and their abundance (weights). Specifically, selecting appropriate forest endmember spectra can significantly improve the accuracy of spectral decomposition. Linear spectral decomposition analysis can be applied to all multispectral data, but a prerequisite for LSMM is that the number of endmembers must be less than or equal to the number of bands in the sensor plus one.

[0114] For example, a linear spectral blending model refers to the spectral reflectance of a pixel in a certain wavelength band, which is a linear combination of the reflectances of each component constituting the pixel with their area proportions as weighting coefficients, expressed as:

[0115]

[0116] Where: R iλ Let be the spectral reflectance of the i-th pixel in the λ band (in other words, it represents the color of a pixel in the image at a specific λ band); r kλ f is the spectral reflectance of the k-th basic component in the u-th band; ki ξ is the abundance of the k-th endpoint element in the i-th pixel (or the proportion of that endpoint in the pixel; the abundance ranges from 0 to 1, representing the relative proportion of that endpoint in the pixel); iλ is the residual value, representing the multiple reflections and transmissions of light between pixel components, which has a non-linear mixing effect (this residual term is mainly used to compensate for the multiple reflections and transmissions of light within the pixel, which causes some non-linear changes in the spectral features of the image); n is the number of terminal elements, and m represents the number of basic components that make up the pixel.

[0117] Specifically, the linear spectral mixture model treats the spectral reflectance of each pixel as a linear combination of multiple land cover components (such as water bodies, vegetation, buildings, etc.), relying on the spectral characteristics of these components and their relative abundance in that pixel. Endmembers: These are the spectral reflectances representing different land cover types. For example, vegetation, water bodies, and bare soil can all be considered endmembers. Abundance: This refers to the relative proportion of each endmember in the pixel, usually estimated using optimization algorithms (such as least squares, nonnegative matrix factorization, etc.). Residual: Due to the nonlinear mixing effect of multiple components in a pixel, the model considers residual values ​​to better fit the spectral data. The goal of this model is to classify land use by inverting the abundance of each component in each pixel. By using the linear spectral mixture model and combining it with topographic data, remote sensing imagery can be decomposed into datasets for buildings, water bodies, woodlands, etc.

[0118] It is worth mentioning that in this step, a hierarchical classification strategy can be used to pre-classify the extracted forest land dataset, dividing it into evergreen forests and deciduous forests. The hierarchical classification strategy involves first dividing the vegetation into evergreen and deciduous forests, and then identifying the various tree species categories within each of the evergreen and deciduous forests. For example, after first dividing the vegetation into evergreen and deciduous forests, species such as North China larch, Chinese pine, aspen, Liaodong oak, and others can be extracted from the evergreen forests to achieve hierarchical classification.

[0119] In some implementations, deciduous forests and evergreen forests can be distinguished based on the Ratio Vegetation Index (RVI). RVI is a vegetation index designed based on the spectral reflectance characteristics of plants in remotely sensed imagery. The RVI provides important information about vegetation reflectance and is a crucial indicator for assessing vegetation abundance, especially in cases of high-density cover, where it is highly sensitive to vegetation conditions. It describes the growth status of ground vegetation by calculating the reflectance ratio of specific spectral bands. RVI is typically calculated using the following formula shown in Table 1:

[0120] RVI = (NIR - Red) / (NIR + Red)

[0121] Here, NIR represents the reflectance in the near-infrared band, and Red represents the reflectance in the red band. For vegetation, a higher RVI value generally indicates higher vegetation cover, and vice versa. By repeatedly adjusting the threshold and referring to previous studies, a threshold range that can effectively extract the gradient change region of evergreen forests can be determined.

[0122] For example, the forest dataset can be divided into evergreen forest and deciduous forest through the following steps:

[0123] S1031. Randomly select sample points from the winter and summer images of deciduous forests in the field survey data, perform ratio processing, and obtain the minimum value of the relatively stable RVI ratio between the two images as the separation threshold:

[0124]

[0125] Among them, RVI S RVI serves as a reference sample for the separation threshold. summer RVI is the RVI value of summer image pixels. winter This represents the RVI value of the winter image pixels.

[0126] In this step, the RVI value of deciduous forests is higher in summer because plant growth is more vigorous; while in winter, the RVI value of deciduous forests is lower because most deciduous plants have lost their leaves or are dormant. Therefore, the RVI... S The RVI ratio should be able to stably characterize the differences between deciduous and evergreen forests. The RVI ratio for evergreen forests should be relatively stable and unaffected by significant seasonal variations, especially for coniferous forests. While the RVI ratio for shrub vegetation may fluctuate, the RVI ratio for evergreen forests changes less between summer and winter. Broadleaf forests show relatively greater variation, and the threshold for separating evergreen and deciduous forests can be determined by calculating the mean and standard deviation of the RVI ratio.

[0127] S1032. Obtain a reference sample RVI for deciduous forests. S The mean M RVI and standard deviation D RVI .

[0128] In this step, the reference sample RVI can be calculated using the following formula. S The mean M RVI and standard deviation D RVI :

[0129]

[0130] Among them, M RVI It is the mean of the RVI ratio, D RVI is the standard deviation, and n is the sample size.

[0131] S1033, Based on the mean M RVI and standard deviation D RVI and multiple reference samples RVI S Plot the frequency histogram to obtain the reference sample RVI that follows a normal distribution. S Frequency and RVI S Normal distribution curve.

[0132] In this step, the reference sample RVI is plotted based on the selected reference sample.S The frequency histogram. Assuming these reference samples follow a uniform distribution, then based on the assumption of a normal distribution, the distribution curve of the RVI ratio can be obtained; see [reference needed]. Figure 2 .

[0133] S1034, according to RVI S Frequency and RVI S The normal distribution curve is used to set the discrimination rules, and the deciduous forest and evergreen forest are determined by the discrimination rules.

[0134] Figure 2 The RVI based on the reference sample is shown. S A schematic diagram of the rules for distinguishing evergreen and deciduous forests using a frequency histogram. (Example) Figure 2 As shown, due to the selection of reference sample RVI S Following the principle of uniform distribution, the reference sample RVI is plotted. S The frequency distribution follows a normal distribution. In this embodiment, deciduous forest sampling points are selected as reference samples, and the discrimination rule is that if RVI S ∈[M RVI -D RVI M RVI +D RVI If the pixel identifier indicates a deciduous forest, then the forest feature is identified as such. If RVI S ∈[0, M RVI -2D RVI If the pixel identifier indicates a forest feature, then the forest feature is identified as evergreen forest. The next interval for determination is [M]. RVI -2D RVI M RVI -D RVI ] and [M RVI +D RVI M RVI +2D RVI The reclassified intervals are then used as training samples, employing a hybrid pixel decomposition method combined with texture features for further classification. Texture features: In remote sensing image classification, texture features are used to capture spatial variations and patterns in an image, which helps to further distinguish different forest types (e.g., evergreen forests and deciduous forests). In this embodiment, to achieve more accurate classification, texture features and a hybrid pixel decomposition method are combined to re-analyze the already classified pixels. Hybrid pixel decomposition methods (e.g., the linear spectral mixing model described above) can handle multiple components in mixed pixels, thereby improving classification accuracy.

[0135] S104. Based on the forest dataset, perform Savitzaky-Golay filtering transformation and spectral differential transformation to obtain the NDVI transformed image set; the NDVI transformed image set indicates the time-based remote sensing phenological characteristics of plants, which in turn indicate the growth and development status and periodic changes of plants in different seasons.

[0136] Understandably, time-series-based remote sensing phenological characteristics refer to information obtained through remote sensing technology that reflects the temporal changes in plant growth cycles and ecological characteristics. These characteristics mainly include changes in reflectance spectra during plant growth, flowering, and withering processes, and are typically used to analyze plant phenological stages (such as spring germination, summer vigorous growth, and autumn leaf fall). Plant phenology refers to the timing and duration of seasonal stages (such as germination, flowering, and fruiting) within a plant's growth cycle. Time-series remote sensing phenological characteristics, through remote sensing indices such as NDVI, help researchers monitor plant growth status, particularly changes in vegetation under different seasons or climatic conditions. For example, NDVI values ​​are typically higher during spring vegetation recovery, while they decrease during winter or drought periods. By analyzing this time-series data, it is possible to infer plant growth cycles and suitable habitats.

[0137] In some implementations, step S104 can be achieved through the following steps:

[0138] S1041. Based on the phenological information of the dominant tree species in the study area, the Savitaky-Golay algorithm is used to perform polynomial filtering transformation on the NDVI time series data to obtain smoothed and denoised NDVI time series spectral curves of the dominant tree species.

[0139] The phenological information includes seasonal phase characteristics of dominant tree species. These characteristics indicate canopy changes during the bud opening period, peak leaf expansion period, peak growth period, full leaf color change period, and end of leaf fall for each species. For details, please refer to [link / reference needed]. Figure 3 . Figure 3 The seasonal characteristics of the dominant tree species in the study area are shown.

[0140] In this step, the Savitzky-Golay algorithm (or SG filtering algorithm) performs polynomial filtering transformation on the NDVI (Normalized Difference Vegetation Index) time series data to smooth and denoise the NDVI time series data, thereby enhancing the signal characteristics of the NDVI time series data and thus better classifying tree species.

[0141] It should be noted that the SG filtering algorithm is a filtering method based on local polynomial least squares in the time domain, widely used for smoothing and denoising data streams. In the time-series data analysis of this embodiment, the SG filtering algorithm is used to remove noise while preserving the main characteristics of the signal. The key to the SG filtering algorithm is to weight the load points within the window. The weights are obtained by least squares fitting of a given high-order polynomial. It can more effectively preserve the characteristics of the signal while filtering and smoothing. After verification with actual sampling points, this embodiment ultimately adopts the SG filtering algorithm with a smoothing window of 4 and a convolution dimension of 2.

[0142] For example, the Savitaky-Golay algorithm is used to perform polynomial filtering transformation on NDVI time series data, including:

[0143] For NDVI time series data Y(t), (t=0,1…,Y N-1 Where N is a positive integer, for its 2k+1 neighboring data points (where k is a positive integer), fit an l-th order polynomial within the local region:

[0144] σ(t)=a0+a1t+a2t 2 +…+a l t l

[0145] In the formula, σ(t) is the fitted polynomial, representing the fitted value at time point t (i.e., the result of the polynomial fitting); a l (l = 0, 1, ..., l) are the coefficients of the polynomial, which need to be obtained through fitting; l is the order of the polynomial (a positive integer), representing the power of the highest-order term in the polynomial.

[0146] In this model, suppose we have a time series dataset Y(t), where t represents time (from 0 to N-1), and Y(t) is the observation at time t. For each data point Y(t) in the time series, we consider its neighboring data points. Specifically, we choose 2k+1 data points, where k is a positive integer representing how many neighboring data points are taken to the left and right of each data point. In other words, if we want to fit a data point Y(t), we will use the k points to its left and right (including itself, a total of 2k+1 points) to perform a polynomial fit.

[0147] The polynomial coefficients a are obtained by fitting local polynomial least squares method. l This minimizes the squared error between the fitted polynomial and the original data within the local neighborhood:

[0148]

[0149] In the formula, E is the squared error, and Cj These are the weighting coefficients, Y(t+j) is the actual observed value at time point t+j, and σ(t+j) is the actual fitted value at time point t+j calculated using polynomial fitting.

[0150] The goal of fitting is to find the coefficients a0, a1, ..., a of these polynomials. l This allows the fitting polynomial within the local region to minimize the fitting error. A common approach is to obtain these coefficients using the least squares method. The error is defined as the squared difference between the fitted curve σ(t+j) and the actual observed value Y(t+j) for each data point Y(t+j) within the local region, and the best fit is obtained by minimizing this squared error.

[0151] Figure 4 The NDVI time-series spectral curves of the dominant tree species after SG filtering transformation are shown. Figure 4 As shown, through the above polynomial fitting process, the smoothed data points Y can be obtained. (t) These smoothed data points Y (t) That is, the data after filtering and denoising, connected to the above data points Y. (t) The NDVI time-series spectral curves of the dominant tree species after SG filtering transformation can then be obtained.

[0152] S1042. Based on first-order and second-order differential transformations, the NDVI time-series spectral curves of the dominant tree species after extracting different spectral parameters are obtained.

[0153] In this step, the formulas for calculating the first-order and second-order differential transforms are as follows:

[0154]

[0155] In the formula, R(λ) i R(λ) is the spectral reflectance value at band i. i )′,R(λ i )″ represent the first and second order spectral values ​​between bands i and i+1, respectively, and Δλ is the step size between adjacent bands.

[0156] Understandably, spectral differential transformation involves mathematically simulating the reflectance spectrum and calculating differential values ​​of different orders to extract different spectral parameters. Spectral differential transformation can eliminate some atmospheric effects and the influence of vegetation environments such as soil, better reflecting the essential characteristics of plants. Specifically, the spectral curve after first-order differentiation enhances the slope changes of the spectral curve, reflecting rapid changes between different bands and highlighting the temporal shift differences of long-term spectral curves. That is, the trend of spectral curve changes becomes more obvious over time, which helps in analyzing the dynamic changes of the plant growth cycle. Second-order differential transformation effectively reflects the inflection points and curvature characteristics of spectral curve changes, enhancing the differences between various spectral curves. Through second-order differential transformation, detailed features of the spectral curve (such as dramatic changes in reflectance) can be effectively extracted, thereby enhancing the differences between different spectral curves and improving the accuracy of plant feature extraction. Second-order differentiation can reveal the changing trends of the curve, such as the peaks and troughs of spectral reflectance values, suitable for analyzing more complex vegetation changes. Third-order differential transformation can theoretically further extract detailed characteristics, such as more subtle changes in reflectance. It can capture more minute changes and enhance the characteristic details of the spectrum. However, the calculation result of the third-order differential is related to the deflection of the original space vector (rotation of the curve), which may lead to some unstable changes. Therefore, in this embodiment, the third-order differential transformation will not be discussed in detail.

[0157] Figure 5 The NDVI time-series spectral curves of the dominant tree species after SG filtering and differential transformation are shown. Figure 5 As shown, spectral differential transformation can eliminate spectral interference caused by atmospheric and soil factors, thereby enhancing the reflectance characteristics of the plant itself.

[0158] S105. Based on the vegetation spectral index and NDVI transformed image set as classification features, the tree species in the study area are classified using a random forest model.

[0159] In this step, vegetation spectral indices and NDVI-transformed image sets are used as input variables for the random forest model to output tree species identification results, including the identification of various tree species in deciduous forests and evergreen forests. The Random Forest (RF) algorithm is an algorithm that integrates multiple trees using the Bagging concept of ensemble learning; its basic unit is the decision tree. A decision tree is a tree-like structure that divides data into different categories through nodes and branches. Each node represents a conditional judgment for a feature, branches represent different decision results, and the final leaf nodes represent the classification result.

[0160] For example, the random forest model randomly selects N samples from the original training set as training samples using the bootstrap method. In this embodiment, a 7:3 ratio is used for training sample selection, and the remaining 30% of the samples are called out-of-bag (OOB) data, which can be used to estimate the performance of the random forest model. During the training process of each tree, due to the random selection characteristic of the bootstrap method, each training sample may be selected multiple times, and some samples may not be selected. These unselected samples are the out-of-bag data.

[0161] Optionally, the training samples include multi-seasonal phenological features, Sentinel-1 radar data, and red-edge region bands. Multi-seasonal phenological features: Extracting multi-seasonal vegetation growth characteristics from time-series images, such as NDVI values ​​and curve morphology in different seasons. Sentinel-1: Combining Sentinel-1 SAR data (such as synthetic aperture radar images), utilizing the scattering characteristics of radar signals to improve classification accuracy. Red-edge region bands: Combining Sentinel-2 red-edge bands, using the red-edge bands to perform sensitivity analysis on vegetation and distinguish different tree species. These features are input into a random forest model for training; ensemble classification using multiple decision trees effectively improves classification accuracy.

[0162] In some implementations, the tree species identification method based on dense temporal images provided in this application further includes S106, which verifies the accuracy of the tree species identification results.

[0163] Figure 6 The diagram shows radar charts illustrating the identification accuracy of various dominant tree species based on different feature combinations (SI: spectral index; SG: SG filter transform; FD: first-order differential transform; SD: second-order differential transform). For example... Figure 6 As shown, accuracy verification can evaluate tree species identification results from overall accuracy (OA) and kappa coefficient, producer accuracy (PA), and user accuracy (UA). The calculation formulas for each parameter are as follows:

[0164]

[0165] In the formula, h i X is the number of correctly classified samples in class i; i T represents the total number of pixels in the i-th class; i is the total number of real pixels in class i; N is the total number of pixels in the classification.

[0166] The kappa calculation result is -1 to 1, but usually the kappa falls between 0 and 1. It can be divided into five groups to represent different levels of consistency: 0.0 to 0.20 very low consistency (slight), 0.20 to 0.40 fair consistency (fair), 0.40 to 0.60 moderate consistency (moderate), 0.60 to 0.80 substantial consistency (substantial), and 0.80 to 1 almost perfect consistency (almost perfect).

[0167] The tree species identification method based on dense temporal images provided in this application includes, but is not limited to, the following application scenarios:

[0168] (1) Dynamic updating of forest resources. Tree species identification is an important part of forestry resource surveys, and accurate tree species identification is crucial for natural resource management. However, conventional tree species identification mainly relies on field surveys, which are labor-intensive, time-consuming, and resource-intensive. This study uses remote sensing technology to make the acquisition of forest resources faster, more accurate, and more efficient, and it plays a very important role in the dynamic monitoring of forestry resources and the classification and identification of tree species.

[0169] (2) Accurate estimation of carbon storage. Accurate estimation of tree and forest biomass is crucial for assessing and tracking the current status and rate of change of forest carbon pools, predicting the role and feedback of forests in regional and global carbon cycles, and providing an important basis for policy-making in response to climate change. Tree species diversity has different impacts on carbon storage for different types of tree species. The calculation of carbon storage requires the establishment of a set of equations based on different tree species. Therefore, clarifying the distribution of tree species in a region is the foundation for accurate estimation of carbon storage.

[0170] (3) Real-time monitoring of precious tree species. Tree species mapping involves screening for precious tree species, clarifying and updating the spatial location of precious tree species in real time, providing real-time information for forest protectors, and further facilitating the timely and effective formulation of protection measures.

[0171] (4) Forest tree species diversity monitoring. Remote sensing technology is gradually becoming an emerging means of large-scale and rapid monitoring of forest biodiversity, providing a strong guarantee for the rapid extraction of spatial pattern information of tree species diversity, and large-scale tree species mapping provides data basis for regional forest species diversity detection.

[0172] Example 1

[0173] Figure 7 A flowchart illustrating the application scenario provided in Embodiment 1 of this application. For example... Figure 7As shown, five typical tree species (Chinese pine, Liaodong oak, poplar, North China larch, and other tree species) in a certain region were selected as the subjects. The images used for tree species extraction included 66 Sentinel-2 remote sensing images from 2019 to 2022. The methods used for image preprocessing and tree species identification included "Linear Spectral Mixture Model (LSMM)," "RVI thresholding method," "Savitzaky-Golay filter transform," "spectral differential transform," and "random forest algorithm." The specific steps are as follows:

[0174] S201. Obtain raw data.

[0175] For example, the raw data includes Sentinel-2, topographic data, sample data, and Sentinel-1. The original Sentinel-1 and Sentinel-2 remote sensing imagery data were both sourced from the European Space Agency, with 66 Sentinel-2 images of the Huodong mining area acquired through the GEE platform for the year of evaluation. The topographic data used was the NASADEM dataset. NASADEM is a reprocessed digital elevation model (DEM) dataset generated from the Space Shuttle Radar Terrain Mission (SRTM) data. By incorporating data primarily from the Ice, Cloud, and Land Elevation Satellite (ICESat), the Geoscience Laser Altimeter System (GLAS), and the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) instruments, NASADEM provides more accurate global topographic data. The sample data came from existing forest inventory data and field survey data, with 70% used as training samples to train the classification model and 30% used as validation samples to verify the accuracy of the random forest model. Field survey data were used as on-site data. A systematic sampling strategy with random points was adopted. Square plots of 30m*30m were designed. Trees with a diameter at breast height (DBH) greater than 4m were measured and their height was recorded. At the same time, handheld LiDAR (SLAM) was used for scanning. A total of 412 plots were collected.

[0176] S202, Preprocessing.

[0177] First, the Sentinel-2 time-series imagery was preprocessed. Since the pixel resolution of Sentinel-2 bands B5, B6, B7, B8A, B11, and B12 is 20m, bicubic interpolation was used to resample these bands to 10m. Bicubic interpolation has an edge enhancement effect and can better preserve the fine structure of the image, but it is computationally intensive. Using the GEE platform effectively overcomes this weakness. Next, bands B2, B3, B4, and B8, along with their resampled bands, were reprojected onto the EPSG:4326 coordinate system and fused into a new image.

[0178] To fully utilize the high spectral resolution of Sentinel-2, vegetation indices were constructed sequentially using information from different image bands (Table 1), and then combined with 10 preprocessed Sentinel-2 bands to construct 12 feature variables as shown in Table 1. These feature variables help improve the accuracy of tree species classification.

[0179] For example, the raw remote sensing image data can also be preprocessed via a satellite platform, including cloud removal, terrain correction, and image cropping. Optionally, cloud removal can employ the Simple Cloud Score algorithm; terrain correction can employ the SCS+C correction method based on the Lambertian reflectance model of the DEM; and image cropping can employ the Quality Mosaic algorithm.

[0180] S203. Using a linear spectral mixture model combined with topographic data, land use classification was performed on the land features in the study area, which were divided into forest land, buildings, water bodies and other land features.

[0181] S204. Based on the extracted forest land data, the regional forest land is pre-classified using the RVI thresholding method. This involves identifying 280 pure deciduous forest sample points and plotting the RVI values. S Frequency histogram and calculate RVI S The mean and standard deviation, and according to Figure 2 The discrimination rules are used to classify evergreen forests and deciduous forests.

[0182] S205, Perform Savitzaky-Golay filter transformation and spectral differential transformation.

[0183] All cloud-removed imagery from 2019-2022 was used. The NDVI time-series spectral curves of the canopy for the five dominant tree species, after SG algorithm filtering and spectral differential transformation, are shown below. Figure 4 and Figure 5 .

[0184] S206. Tree species classification and accuracy evaluation.

[0185] Based on step S205 Figure 4 and Figure 5 The dominant tree species canopy NDVI time-series spectral curve information is used as input variables for the random forest model to identify various tree species in deciduous and evergreen forests.

[0186] The accuracy of the tree species classification results was verified. The tree species identification results were evaluated based on overall accuracy, kappa coefficient, producer accuracy, and user accuracy.

[0187] In this embodiment, the overall accuracy and Kappa coefficient of tree species classification based on the dominant tree species feature combinations under different time-series transformation data are shown in Table 2:

[0188] Table 2. Overall accuracy and Kappa coefficient of advantageous tree species feature combinations based on different time-series transformed data.

[0189]

[0190]

[0191] According to the Kappa coefficient of tree species extraction accuracy in Table 2, the method of this application has a high degree of consistency in the tree species classification results.

[0192] It is understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. Furthermore, in some possible implementations, each step in the above embodiments may be selectively executed according to actual circumstances; it may be partially or fully executed, without limitation here. Additionally, all or part of any feature of any of the above embodiments can be freely and arbitrarily combined without contradiction; the combined technical solution is also within the scope of this application.

[0193] like Figure 8 As shown in the illustration, this application also provides a tree species identification device 300 based on dense time-series imagery, comprising: an acquisition module 301 and a processing module 302. The acquisition module 301 is used to acquire dense long-term multispectral remote sensing images and topographic data of the study area; the dense long-term multispectral remote sensing images include all available remote sensing forest images with cloud cover less than 20%, for example, from 2019 to 2022; the processing module 302 is used to construct a vegetation spectral index by performing image band fusion based on the dense long-term multispectral remote sensing images; the vegetation spectral index is used to indicate the annual dynamic change characteristics of vegetation growth; and a linear spectral mixture model combined with topographic data is used to classify land features in the dense long-term multispectral remote sensing images of the study area into land use categories, namely buildings, water bodies and other land features, and forest land. The dataset was pre-classified based on the extracted forest land dataset. The RVI thresholding method combined with a linear spectral mixture model was used to divide the forest land dataset into evergreen forests and deciduous forests. Savitzaky-Golay filtering and spectral differential transformation were performed on the forest land dataset to obtain an NDVI transformed image set. The NDVI transformed image set indicated the time-series-based remote sensing phenological characteristics of the plants, which indicated the growth and development status and cyclical changes of the plants in different seasons. The tree species in the study area were classified using a random forest model after combining the vegetation spectral index and the NDVI transformed image set as classification features.

[0194] It should be understood that both the acquisition module 301 and the processing module 302 can be implemented in software or in hardware. For example, the implementation of the acquisition module 301 will be described below. Similarly, the implementation of the processing module 302 can refer to the implementation of the acquisition module 301.

[0195] As an example of a software functional unit, module 301 may include code running on a computing instance. The computing instance may include at least one of a physical host (computing device), a virtual machine, or a container. Furthermore, the aforementioned computing instance may be one or more. For example, module 301 may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code may be distributed within the same region or in different regions. Further, the multiple hosts / virtual machines / containers used to run the code may be distributed within the same availability zone (AZ) or in different AZs, each AZ including one or more geographically proximate data centers. Typically, a region may include multiple AZs.

[0196] Similarly, multiple hosts / virtual machines / containers used to run this code can be distributed within the same Virtual Private Cloud (VPC) or across multiple VPCs. Typically, a VPC is set up within a region. Communication between two VPCs within the same region, as well as between VPCs in different regions, requires a communication gateway to be set up within each VPC to enable interconnection between VPCs.

[0197] As an example of a hardware functional unit, the acquisition module 301 may include at least one computing device, such as a server. Alternatively, the acquisition module 301 may also be a device implemented using an Application-Specific Integrated Circuit (ASIC) or a Programmable Logic Device (PLD). The PLD may be implemented using a CPLD, FPGA, GAL, or any combination thereof.

[0198] The multiple computing devices included in the acquisition module 301 can be distributed in the same region or in different regions. Similarly, the multiple computing devices included in the acquisition module 301 can be distributed in the same Availability Zone (AZ) or in different AZs. Likewise, the multiple computing devices included in the acquisition module 301 can be distributed in the same Virtual Private Cloud (VPC) or in multiple VPCs. These multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.

[0199] It should be understood that the above-described device is used to execute the methods in the above embodiments. The implementation principle and technical effect of the corresponding program modules in the device are similar to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.

[0200] Figure 9 A computing device provided in the embodiments of this application, such as Figure 9 As shown, the computing device 70 includes: at least one memory 702 for storing a program; and at least one processor 701 for executing the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the tree species identification method based on remote sensing imagery as described in any of the above embodiments.

[0201] This application provides a computer storage medium 703, which stores instructions that, when executed on a computer, cause the computer to perform a tree species identification method based on remote sensing images as described in any of the above embodiments.

[0202] It is understood that the processor in the embodiments of this application may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor may be a microprocessor or any conventional processor.

[0203] This application also provides a computer-readable storage medium. The computer-readable storage medium stores computer program instructions that, when executed on a computing device, cause the computing device to perform the data processing methods described in the above embodiments. The computer-readable storage medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive).

[0204] This application also provides a computer program product containing instructions. The computer program product may be a software or program product containing instructions, capable of running on a computing device or stored on any usable medium. When the computer program product is run on at least one computing device, it causes the at least one computing device to perform a data processing method.

[0205] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application. Those skilled in the art should understand that although this application has been described in detail with reference to the foregoing embodiments, modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions in the embodiments of this application.

Claims

1. A tree species identification method based on dense temporal images, characterized in that, include: Acquire dense long-term multispectral remote sensing images and topographic data of the study area; the dense long-term multispectral remote sensing images include all available remote sensing images with cloud cover of less than 20%. The vegetation spectral index is constructed by image band fusion based on the dense long-term multispectral remote sensing images; the vegetation spectral index is used to indicate the annual dynamic change characteristics of vegetation growth. Land use classification was performed on land features in dense long-term multispectral remote sensing images of the study area using a linear spectral mixture model combined with topographic data. The data were divided into building, water body and other land features, and forest dataset. Based on the extracted forest dataset, pre-classification was performed. The RVI threshold method combined with the linear spectral mixture model was used to divide the forest dataset into evergreen forest and deciduous forest. Based on the forest dataset, Savitzaky-Golay filtering and spectral differential transformation are performed to obtain an NDVI transformed image set. The NDVI transformed image set indicates the time-based remote sensing phenological characteristics of plants, which indicate the growth and development status and periodic changes of plants in different seasons. The NDVI transformed image set includes NDVI time-series spectral curves, which can be obtained by extracting different spectral parameters of dominant tree species based on first-order and second-order differential transformations. The tree species in the study area are classified by combining the vegetation spectral index and NDVI transformed image set as classification features and then using a random forest model. The vegetation spectral index includes the ratio vegetation index (RVI); the pre-classification based on the extracted forest land dataset, dividing the forest land dataset into evergreen forest and deciduous forest, includes: The RVI values ​​of sample points from winter and summer images of deciduous forests were randomly selected from field survey data and their ratios were processed. The minimum value of the relatively stable RVI ratio between the two images was used as the separation threshold. in, As a reference sample for the separation threshold, The RVI value of the summer image pixels. RVI values ​​for winter image pixels; Obtain a reference sample of deciduous forests mean and standard deviation ; Based on the mean and standard deviation and multiple reference samples Plot the frequency histogram to obtain a reference sample that follows a normal distribution. frequency and Normal distribution curve; in accordance with frequency and A discrimination rule is set based on the normal distribution curve, and the deciduous forest and evergreen forest are determined by the discrimination rule; Based on first-order and second-order differential transforms, the NDVI time-series spectral curves of the dominant tree species after extracting different spectral parameters were obtained, including: The formulas for calculating the first-order and second-order differential transforms are as follows: In the formula, It is a band Spectral reflectance value at that location, , bands and The first and second order spectral values ​​between them, This represents the step size between adjacent bands.

2. The method according to claim 1, characterized in that, The step of constructing vegetation spectral indices by image band fusion based on the dense long-term multispectral remote sensing images includes: Forest growth images are obtained by preprocessing the dense long-term multispectral remote sensing images and topographic data. The preprocessing includes image registration, resampling, time series and cloud cover filtering, radiometric correction, geometric correction, band synthesis, cloud and shadow masking, and image cropping.

3. The method according to claim 2, characterized in that, The resampling includes: The terrain data was resampled to a spatial resolution of 10m using a linear sampling method. The dense long-term multispectral remote sensing image was resampled to 10m using bicubic interpolation for bands with a pixel resolution of 20m.

4. The method according to any one of claims 1-3, characterized in that, The method of using a linear spectral mixture model combined with topographic data to classify land use features in dense long-term multispectral remote sensing images of the study area includes: A linear spectral blending model refers to the spectral reflectance of a pixel in a certain wavelength band. It is a linear combination of the reflectances of the components that make up the pixel, weighted by their area proportions, and is expressed as: in: For the i-th pixel in Spectral reflectance of the band; It is the spectral reflectance of the k-th basic component in the u-band; It is the abundance of the k-th endpoint element in the i-th pixel; is the residual value, representing the multiple reflections and transmissions of light between pixel components, exhibiting a non-linear mixing effect; n is the number of terminal elements, and m represents the number of basic components constituting the pixel.

5. The method according to claim 4, characterized in that, The vegetation spectral index includes time-series data of the Normalized Difference Vegetation Index (NDVI); the step of performing filtering and spectral differential transformation based on the forest dataset to obtain the NDVI-transformed image set includes: Based on the phenological information of dominant tree species in the study area, the Savitaky-Golay algorithm was used to perform polynomial filtering transformation on the NDVI time series data to obtain smoothed and denoised NDVI time series spectral curves of dominant tree species. The phenological information includes seasonal characteristic information of dominant tree species, which indicates the canopy changes of each tree species during the leaf bud opening period, peak leaf expansion period, peak growth period, full leaf color change period, and end of leaf fall period. Based on first-order and second-order differential transformations, the NDVI time-series spectral curves of dominant tree species after extracting different spectral parameters were obtained.

6. The method according to claim 5, characterized in that, The process of performing polynomial filtering transformation on NDVI time series data using the Savitaky-Golay algorithm includes: For NDVI time series data Where N is a positive integer, for its 2k+1 adjacent data points, where k is a positive integer, fit an l-th order polynomial within the local region: In the formula, σ(t) is the fitted polynomial; The coefficients of the polynomial; Polynomial coefficients are obtained by fitting local polynomial least squares method. This minimizes the squared error between the fitted polynomial and the original data within the local neighborhood: In the formula, The error is the squared error. These are weighting coefficients. It is a point in time. The actual observed value at that location, The time points were calculated using polynomial fitting. The actual fitted value at that location; Through the above polynomial fitting process, smoothed data points can be obtained. These smoothed data points That is, the data after filtering and denoising, to obtain the NDVI time-series spectral curve of the dominant tree species after SG filtering transformation.

7. The method according to any one of claims 1-3, characterized in that, The method further includes: verifying the accuracy of the tree species identification results, including: The tree species identification results were evaluated from two aspects: overall accuracy and kappa coefficient. The overall accuracy evaluation involves comparing the pixel-level tree species identification results of long-term remote sensing image data with ground truth values ​​to assess the accuracy of the classification results. The formula for calculating the overall accuracy is as follows: Overall accuracy = (TP + TN) / (TP + TN + FP + FN) TP represents the number of samples that are truly positive and correctly predicted as positive, TN represents the number of samples that are truly negative and correctly predicted as negative, FP represents the number of samples that are truly negative but incorrectly predicted as positive, and FN represents the number of samples that are truly positive but incorrectly predicted as negative. The Kappa coefficient is a statistical indicator used to evaluate the consistency between classifiers or evaluators. The formula for calculating the Kappa coefficient is as follows: Where Po is the sum of the number of correctly classified samples in each class divided by the total number of samples, which is the overall classification accuracy; assuming the number of true samples in each class are a1, a2, ..., a C The predicted number of samples for each class are b1, b2, ..., b C If the total number of samples is n, then: The kappa calculation result is -1 to 1, but kappa falls between 0 and 1, and can be divided into five groups to represent different levels of consistency: 0.0 to 0.20 very low consistency, 0.20 to 0.40 general consistency, 0.40 to 0.60 medium consistency, 0.60 to 0.80 high consistency, and 0.80 to 1 almost perfect consistency.

8. A tree species identification device based on dense temporal images, characterized in that, include: The acquisition module is used to acquire dense long-term multispectral remote sensing images and topographic data of the study area; the dense long-term multispectral remote sensing images include all available remote sensing forest images with cloud cover of less than 20%. The processing module is used to construct a vegetation spectral index by performing image band fusion based on the dense long-term multispectral remote sensing image; the vegetation spectral index is used to indicate the annual dynamic change characteristics of vegetation growth. Land use classification was performed on dense long-term multispectral remote sensing images of the study area using a linear spectral mixture model combined with topographic data. The data was categorized into buildings, water bodies and other land features, and forest land. Based on the extracted forest land dataset, pre-classification was performed, and the forest land dataset was divided into evergreen forest and deciduous forest using the RVI thresholding method combined with the linear spectral mixture model. Savitzaky-Golay filtering and spectral differential transformation were performed on the forest land dataset to obtain an NDVI transformed image set. The NDVI transformed image set indicates the time-based remote sensing phenological characteristics of plants, which indicate the growth and development status and periodic changes of plants in different seasons. The NDVI transformed image set includes NDVI time-series spectral curves, which can be obtained by first-order and second-order differential transformations, extracting NDVI time-series spectral curves of dominant tree species after different spectral parameters. The tree species in the study area were classified by combining the vegetation spectral index and the NDVI transformed image set as classification features using a random forest model. The vegetation spectral index includes the ratio vegetation index (RVI); the pre-classification based on the extracted forest land dataset, dividing the forest land dataset into evergreen forest and deciduous forest, includes: The RVI values ​​of sample points from winter and summer images of deciduous forests were randomly selected from field survey data and their ratios were processed. The minimum value of the relatively stable RVI ratio between the two images was used as the separation threshold. in, As a reference sample for the separation threshold, The RVI value of the summer image pixels. RVI values ​​for winter image pixels; Obtain a reference sample of deciduous forests mean and standard deviation ; Based on the mean and standard deviation and multiple reference samples Plot the frequency histogram to obtain a reference sample that follows a normal distribution. frequency and Normal distribution curve; in accordance with frequency and A discrimination rule is set based on the normal distribution curve, and the deciduous forest and evergreen forest are determined by the discrimination rule; Based on first-order and second-order differential transforms, the NDVI time-series spectral curves of the dominant tree species after extracting different spectral parameters were obtained, including: The formulas for calculating the first-order and second-order differential transforms are as follows: In the formula, It is a band Spectral reflectance value at that location, , bands and The first and second order spectral values ​​between them, This represents the step size between adjacent bands.

Images