Research On The Rapid Extraction Method Of Eucalyptus Forest Information Based On GF-6 Satellite Remote Sensing Images

Eucalyptus is one of the major plantation tree species in southern China,and its fast-growing and productive characteristics have created good economic benefits,triggering large-scale rapid expansion and disorderly planting.The large-scale planting behavior of eucalyptus poses a major threat to soil moisture,soil nutrients,and biodiversity,so monitoring information on the spatial distribution of eucalyptus plantations is of great significance for sustainable development and ecological protection of regional plantation forests.At present,the traditional eucalyptus forest monitoring methods mainly rely on the survey of forestry resources,which is inefficient.With the development of remote sensing technology,remote sensing data has been widely used in forest resource inventory and other aspects,which is helpful for large-scale monitoring of plantation forest structure,but the lack of spectral information limits its application effect in forestry remote sensing.With the successful launch of China’s first satellite with the "red-edge" band,the red-edge band is more sensitive to vegetation features,which brings greater application prospects for eucalyptus forestry remote sensing extraction.Therefore,this study adopts the CART(Classification and Regression Tree)multi-scale hierarchical classification extraction method and SNIC(Simple Non-Iterative Clustering)superpixels + SVM(Support Vector Machines)algorithm for object-oriented eucalyptus forestry remote sensing extraction based on the features of GF-6 satellite images.The main research contents and results of this paper are as follows.(1)Starting from the theory of object-oriented multiscale segmentation,the optimal segmentation parameters are selected for multiscale segmentation and the class hierarchy for eucalyptus forest extraction is constructed.Based on the GF-6 band spectral information to define the red-edge index features,the SPM(Salford Predictive Modeler)model is used to construct the CART decision tree to realize automatic feature screening and threshold setting in the classification extraction process.The overall accuracy of CART object-oriented eucalyptus forest information extraction based on the multi-scale hierarchical segmentation method was 91.75%,with the Kappa coefficient of 0.84 and F1-Score of 0.92.The consistency verification results were highly consistent and significantly better than the pixel-based method,which verified the feasibility of applying this method to eucalyptus forest information identification and extraction.(2)The classification algorithm combining SNIC superpixels and SVM is proposed,and the whole process of multi-feature extraction and classification recognition of GF-6 images is implemented by code programming in the GEE cloud computing environment.the study of eucalyptus forest information extraction under SNIC superpixels algorithm has achieved good results,and the highest overall accuracy obtained by the feature combination scheme combining red-edge information is 91.74%,and the F1-Score reaches 0.91.The SNIC superpixel-based method has higher completeness and accuracy compared with the pixel-based method.The object-oriented SNIC+SVM algorithm proposed in this study can be executed quickly and can identify the plantation area and spatial distribution of eucalyptus forests accurately and efficiently,and is suitable for application in forestry departments and ecological protection agencies for effective monitoring and management of large-scale plantation of eucalyptus forests.(3)The new bands of GF-6 satellite data were introduced and the red-edge index features based on their derivation were constructed.The comparative analysis of the multi-feature combination scheme showed that the GF-6 red-edge data effectively improved the classification extraction accuracy and enhanced the vegetation information,and had significant contributions to the eucalyptus forest information extraction.Meanwhile,the effectiveness of the GF-6 satellite’s red-edge band as a sensitive spectral band of vegetation spectra is verified,which can effectively reflect the differences in eucalyptus plantation forest characteristics.Therefore,the domestic GF-6 satellite broadband multispectral data has good applicability and reliability in eucalyptus forestry resource management and remote sensing monitoring.