Abstract:

The development of high-resolution remote sensing imaging technology provides a new way to the large-scale estimation of forest canopy density. The traditional inversion methods for canopy density only use spectral or topographical features of remote sensing images. However, due to the existence of the different thing with same spectrum and the same thing with different spectrum phenomena, it is difficult to improve the estimation accuracy of canopy density. Based on spectrum and other traditional features, this paper combines texture features of remote sensing images to estimate canopy density. Firstly, the gray level co-occurrence matrix (GLCM) texture features are computed using object-based method. Then, prinicipal component analysis (PCA) method is applied in correlation analysis and dimension reduction of texture features. Finally, spectrum and topographical features together with texture features are introduced into stepwise regression model to estimate canopy density. The experimental results showed that compared with the traditional method only based on spectrum or topographical features, the method combined with texture features greatly improved the estimation accuracy. The coefficient of determination (adjusted R²) increased from 0.737 to 0.805. The estimation accuracy increased from 81.03% to 84.32%.

Key words: canopy density, texture, gray level co-occurrence matrix (GLCM), block-oriented, principal component analysis (PCA), stepwise linear regression