Comparative study of feature dimension reduction algorithm for high-resolution remote sensing image classification

The high-resolution remote sensing image classification is an important research topic in pattern recognition, and its computational complexity grows exponentially with the increase of the dimension. Hence, it is necessary to perform feature dimension reduction. This paper presents a comparative study on state-of-the-art feature selection and feature transformation methods for the task of high-resolution remote sensing image classification. We conduct a group of experiments on mRMR, PCA and KPCA for their applicability. Comparison results show that nonlinear dimension reduction method based on feature transformation is more suitable for the task at hand. What´s more, appropriate kernel function and kernel parameters are also essential. It is vital to reduce the dimension, which can alleviate the computational cost greatly and improve accuracy.