Classification of forest vegetation types in Jilin Province, China based on deep learning and multi-temporal Sentinel-2 data

Abstract

The classification of forest vegetation types plays an important role in the land management agencies for natural resource inventory information, especially for federally protected national forests in China. The classification results are widely used in the calculation and inversion of parameters such as forest storage volume, biomass and coverage. Forest canopy density response the extent to which the canopy is connected to each other in the forest. It can be used to observe vegetation growth. In recent years, deep learning convolutional neural networks have made significant progress in the task of remote sensing image classification and recognition. Considering that the spectral characteristics of forests in different seasons in Jilin Province of China are quite different, this paper used the optical image data of Sentinel-2A in summer, spring and autumn as the data source to calculate the normalized difference vegetation index (NDVI), bare index (BI), perpendicular vegetation index (PVI) and shadow index (SI). Next use the four vegetation indexes combined with weighted overlay analysis method to calculate forest canopy density. In this paper, the convolutional neural network (CNN) was used as the forest vegetation type classifier. The classification indexes were the spectral data and the spectral data combined with the forest canopy density information, respectively. The experiment shows that the forest canopy density can significantly improve the classification accuracy and the overall accuracy is increased from 85.58% to 90.41%.