A tree species identification method based on spectral orthogonal transformation and multivariate filtering

Abstract

The present application relates to a kind of tree species identification method based on spectral orthogonal transformation and multivariate filtering, comprising the following steps: S1, the high spatial resolution remote sensing image of forest area is collected, and pretreatment is carried out;S2, preliminary interpretation is carried out to the image after pretreatment, and tree species sample is obtained in combination with field investigation, tree species sample is marked on the image after pretreatment, and is divided into training sample and verification sample;S3, the spectral band of remote sensing image is orthogonally transformed, and a set of two two orthogonal feature layers are generated;S4, multivariate filtering processing is carried out to the layer after orthogonal transformation, and filter feature is extracted, and hybrid feature set is constructed;S5, based on hybrid feature set, tree species is identified using machine learning classification method, and classification result and precision statistics are output.The present application reduces the correlation between layers by orthogonal transformation of multispectral data, then highlights the difference of tree species by four filters of texture, low-pass, median and enhanced Lee, to promote the identification of tree species.

Description

Technical Field

[0001] This invention relates to the field of remote sensing identification technology, specifically to a tree species identification method based on spectral orthogonal transformation and multivariate filtering. Background Technology

[0002] Tree species identification based on remote sensing imagery is one of the key technologies for forestry resource surveys and ecological monitoring. Current research trends mainly focus on data source selection, image time series determination and methodological innovation, and feature mining.

[0003] Effective image feature mining is a hot topic, as effective feature transformation can significantly improve the accuracy of tree species identification. However, highly correlated spectral bands can hinder tree species identification. Current research generally employs techniques such as principal component analysis and minimum noise separation to circumvent the correlation between bands, and then uses newly generated feature layers to identify tree species. However, because these techniques extract a limited number of uncorrelated layers and have poor effectiveness, they are insufficient to deeply characterize the subtle differences between tree species. Applying them to identify tree species does not significantly improve accuracy, and may even reduce it.

[0004] To overcome the shortcomings of existing technologies in handling highly correlated variables, this invention proposes a tree species identification method based on spectral orthogonal transformation and multivariate filtering, so as to significantly improve the identification accuracy of tree species. Summary of the Invention

[0005] To address the above problems, this invention proposes a tree species identification method based on spectral orthogonal transform and multivariate filtering. The specific technical solution is as follows: A tree species identification method based on spectral orthogonal transform and multivariate filtering includes the following steps: S1. Collect high spatial resolution remote sensing images of the forest area and perform preprocessing; S2. Perform preliminary interpretation on the preprocessed images, obtain tree species samples in conjunction with field surveys, mark the tree species samples on the preprocessed images, and divide them into training samples and validation samples. S3. Perform orthogonal transformation on the spectral bands of the remote sensing image to generate a set of pairwise orthogonal feature layers; S4. Perform multivariate filtering on the orthogonal transformed layer, extract the filtered features, and construct a hybrid feature set; S5. Based on the hybrid feature set, the tree species are identified using machine learning classification methods, and the classification results and accuracy statistics are output.

[0006] Furthermore, in S2, the sample partitioning adopts a stratified random sampling method.

[0007] Furthermore, in S3, the orthogonal transformation adopts the Gram-Schmidt orthogonal transformation method.

[0008] Furthermore, during multivariate filtering in S4, texture filters, low-pass filters, median filters, and enhanced Lee filters are used to extract filtering features.

[0009] Furthermore, the features extracted by texture filtering include mean, variance, coherence, entropy, correlation, contrast, dissimilarity, and second moment of angle.

[0010] Furthermore, in S4, the window size for filtering is 9×9 pixels.

[0011] Furthermore, machine learning classification methods in S5 include Random Forest (RF) or XGBoost.

[0012] The beneficial effects of this invention are as follows: 1. Orthogonal transformation of multispectral data helps reduce the correlation between layers, generate purer features that are more capable of distinguishing tree species, and improve the independence and discriminative ability of features. By fusing four filters of orthogonal layers—texture, low-pass, median, and enhanced Lee—the differences in canopy texture, macroscopic morphology, and edge preservation characteristics can be fully reflected, which helps in the identification of tree species.

[0013] 2. The implementation process is relatively simple, time-consuming, and highly accurate, making it valuable for identifying trees in urban forests, nurseries, and other similar locations. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a high spatial resolution image of the labeled tree samples as described in the embodiment; Figure 2 It is the second high spatial resolution image of the labeled tree samples described in the embodiment; Figure 3 This is the optimal classification result for the high spatial resolution image described in this invention; Figure 4 This is the optimal classification result for the high spatial resolution image II described in this invention. Detailed Implementation

[0016] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0017] The present invention provides the following specific implementation schemes: This invention provides a tree species identification method based on spectral orthogonal transform and multivariate filtering, comprising the following steps: S1. Collect high spatial resolution remote sensing images of the forest area; then perform preprocessing on the collected remote sensing images, such as radiometric calibration, fusion, atmospheric correction, and cropping, so that the preprocessed images can meet the needs of tree species surveys and experimental tests.

[0018] S2. Perform preliminary interpretation of the preprocessed remote sensing images to determine the image colors suitable for tree species surveys and print paper images. In this embodiment, color display is performed according to the RGB432 band combination, and paper images are printed. Carry the paper images to conduct field surveys of tree species, interactively matching typical trees in the field environment with the trees in the images. Then, outline the canopies of the surveyed trees on the paper images and label the tree species. Stop the survey after collecting a sufficient number of samples from various tree species to be identified. After the survey is completed, use ENVI software to mark the obtained tree species samples on the preprocessed remote sensing images (e.g., ...). Figure 1 , 2 As shown in Tables 1 and 2), stratified random sampling is used to extract training and accuracy verification samples from the labeled tree species samples according to a certain proportion.

[0019] Table 1. Tree species IDs, Latin names of tree species / bamboo, and number of pixels in training and accuracy validation samples in high spatial resolution images. Table 2. Tree species numbers, Latin names of tree species, and number of pixels in training and accuracy validation samples in high spatial resolution image 2. S3. The Gram-Schmidt orthogonal transform method is used to orthogonally transform the spectral bands of the remote sensing data. The algorithm design is shown in formulas (1)-(3). Let α1, α2, ..., α n Let be remote sensing data with n bands (layers), and let: (1) (2) (3) The new layers β1, β2, ..., β are generated by transforming all the original spectral bands using the above algorithm. n These are a set of pairwise orthogonal layers. Compared to the original spectral bands, these newly generated layers are able to more independently characterize the differences between tree species.

[0020] S4. Based on the layer obtained by orthogonal transformation, under a suitable window size (9×9 pixels in this embodiment), four types of filters are used to extract filtering features: texture (extracting mean, variance, coherence, entropy, correlation, contrast, dissimilarity, and second moment of angle), low-pass, median, and enhanced Lee. The four filtering features are combined to construct a hybrid feature set for tree species identification.

[0021] S5. Based on the multivariate filtered hybrid feature set obtained in S4, the training samples and accuracy verification samples obtained in S2 are called to identify tree species within the coverage area of ​​remote sensing images. With the support of machine learning classification methods (random forest, XGBoost, etc.), the result map of tree species classification and various accuracy statistics are obtained.

[0022] The tree species classification results obtained using two classification methods, Random Forest and XGBoost, are shown in the figures below. Figure 3 , 4 As shown in Tables 3 and 4, the mean and standard deviation of the overall precision obtained by repeating the operation 20 times for each dataset are respectively shown in Tables 3 and 4.

[0023] Table 3. Mean and standard deviation of overall accuracy obtained from the original spectral bands and multivariate filter layer classification of high spatial resolution images 1 and 2. Table 4. Overall accuracy mean and standard deviation obtained from the classification of high spatial resolution images 1 and 2 using orthogonal transformation layers and multivariate filtering layers. A comparison of Tables 4 and 3 reveals that orthogonal transformation achieves higher tree species identification accuracy than spectral bands; each type of filtering feature in orthogonal data achieves higher tree species identification accuracy than each type of filtering feature in spectral data; and the combination of four types of filtering features in orthogonal data achieves higher tree species identification accuracy than the combination of four types of filtering features in spectral data. These findings are consistent across both high spatial resolution image 1 and high spatial resolution image 2 data types, as well as between random forest and XGBoost classifiers, demonstrating that the tree species identification method combining spectral orthogonal transformation with multivariate filtering significantly improves tree species identification accuracy.

[0024] In this invention, orthogonal transformation of multispectral data helps reduce correlations between layers, generating purer features with better tree species discrimination capabilities, and improving the independence and discriminative power of the features. By fusing four filters from the orthogonal layers—texture, low-pass, median, and enhanced Lee—it can fully reflect the differences in canopy texture, macroscopic morphology, and edge preservation characteristics, which is helpful for tree species identification. Furthermore, the implementation process is relatively simple, time-efficient, and highly accurate, making it highly valuable for identifying trees in urban forests, nurseries, and other similar locations.

[0025] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A tree species identification method based on spectral orthogonal transform and multivariate filtering, characterized in that, Includes the following steps: S1. Collect high spatial resolution remote sensing images of the forest area and perform preprocessing; S2. Perform preliminary interpretation on the preprocessed images, obtain tree species samples in conjunction with field surveys, mark the tree species samples on the preprocessed images, and divide them into training samples and validation samples. S3. Perform orthogonal transformation on the spectral bands of the remote sensing image to generate a set of pairwise orthogonal feature layers; S4. Perform multivariate filtering on the orthogonal transformed layer, extract the filtered features, and construct a hybrid feature set; S5. Based on the hybrid feature set, the tree species are identified using machine learning classification methods, and the classification results and accuracy statistics are output.

2. The tree species identification method based on spectral orthogonal transform and multivariate filtering according to claim 1, characterized in that: In S2, the sample partitioning uses a stratified random sampling method.

3. The tree species identification method based on spectral orthogonal transformation and multivariate filtering according to claim 1, characterized in that: In S3, the orthogonal transformation adopts the Gram-Schmidt orthogonal transformation method.

4. The tree species identification method based on spectral orthogonal transform and multivariate filtering according to claim 3, characterized in that: In S4, multivariate filtering is performed using texture filters, low-pass filters, median filters, and enhanced Lee filters to extract filtering features.

5. The tree species identification method based on spectral orthogonal transform and multivariate filtering according to claim 1, characterized in that: Features extracted by texture filtering include mean, variance, coherence, entropy, correlation, contrast, dissimilarity, and second moment of angle.

6. The tree species identification method based on spectral orthogonal transform and multivariate filtering according to claim 1, characterized in that: In S4, the window size for filtering is 9×9 pixels.

7. The tree species identification method based on spectral orthogonal transform and multivariate filtering according to claim 1, characterized in that: In S5, machine learning classification methods include Random Forest (RF) or XGBoost.

Images