Study On Vegetation Classification In Central Xiao Hinggan Mountains Based On Improved Spatio-temporal Fusion Model

Vegetation is not only an important part of the environment,but also one of the best indicators of human living environment,so the study of vegetation is of great significance.The important premise of vegetation research is to classify vegetation.While traditional vegetation classification method through the field investigation and sampling as the research data,the need of the heavy workload and cost,it is difficult to effectively realize extraction and research on the spatial distribution of vegetation,with the constant innovation of science and technology,remote sensing technology can achieve for the synchronous observation of large area and timeliness strong,has been widely used in the vegetation classification research.Because of the continuity and similarity of phenological characteristics of vegetation,it is difficult to classify vegetation with single image.In addition,it is difficult to obtain high quality remote sensing images with high spatial and temporal resolution from images of bad weather.Based on this,this paper improved the ESTARFM spatio-temporal fusion model.Landsat images and GF-1 WFV images were fused with MODIS images respectively through the improved ESTARFM model to construct NDVI time series data sets,and classified remote sensing images in the central Xiao Hinggan Mountains by object-oriented decision tree.Based on the central area of Xiao Hinggan Mountains for research,the first study of ESTARFM model and its improvement,through improved ESTARFM space-time fusion model of Landsat image and MODIS image data fusion,the space-time and GF-1 and MODIS image data for data fusion of time and space,and the fused image and the original image correlation analysis,select the best fusion data,build vegetation NDVI time-series data sets,and connecting with the actual sample point vegetation index NDVI curve of different vegetation types,construction of classification decision tree,Classification of NDVI data sets based on time series and remote sensing images without time series data sets were conducted in the research area by object-oriented decision tree method,and the classification results were compared,and the following results were obtained:1.In this study,the original ESTARFM is studied firstly,then the sliding window is reduced,and the conversion coefficient algorithm is adjusted to improve the original model.The image fusion experiment of the improved model shows that the image fusion results before and after the model improvement are smaller than the real value,but the average difference of the improved model results is more connected than the fusion results of the original model.In terms of root mean square error and mean absolute error,the fusion result of the improved model is less than that of the original model;in the correlation coefficient part,the correlation coefficient is significantly improved in both red and near infrared bands,in which the correlation coefficient is increased from 0.7248 to 0.7635 in red band and from 0.8275 to 0.8556 in near infrared band;and in the model,the correlation coefficient is increased from 0.8275 to 0.8556 in near infrared band.In terms of operation speed,the improved ESTARFM model takes half less time than the original model,which proves the practicability of the improved ESTARFM model.2.Based on the improved ESTARFM space-time fusion model,this study fuses Landsat image and MODIS image,GF-1 WFV image and MODIS image in the RD band and NIR band.The correlation coefficients of the fusion results of Landsat image and GF-1 WFV image in NIR band are 0.7635 and 0.4474 respectively,but the GF-1WFV data deviate from 1:1 straight line obviously.In RED band,the correlation coefficient of Landsat image is significantly higher than that of GF-1 WFV image;and in program operation,the time required for Landsat image fusion is the same as that of GF-1 WFV image;based on the above reasons,this paper chooses the fusion result of Landsat image and MODIS image.3.The NDVI time series curves of different terrain types are extracted from NDVI spatial-temporal data sets,and based on the sampling points selected from the field measurements and Google high-definition maps.Although the NDVI time series curves of deciduous forest,coniferous forest and coniferous-broad-leaved mixed forest vary roughly,there are obvious differences in different time intervals,which provide a basis for the subsequent classification of this paper.4.Object-oriented classification of images is carried out.Firstly,the optimal segmentation scale is selected for classified images.This paper chooses the optimal segmentation scale of classified images by ESP tool combined with ROC-LV image,and applies this result to the later classification.The classification of remote sensing images based on time series NDVI sequence and the classification of remote sensing images without time series data are carried out respectively.After the accuracy verification,it is proved that the classification of remote sensing images is based on time series NDVI sequence and time series data.When sensory image is classified,the classification result based on time series data is more accurate,and the phenomenon of vegetation misclassification and omission is obviously reduced.