Hybrid gene expression programming-based sensor data correlation mining

This paper deals with the reflectance estimation model issue to improve the estimation accuracy. We propose a model containing two core procedures: dimensionality reduction and model mining. First, the dimensionality reduction algorithm of hyperspectral data based on dependence degree (DRND-DD) is proposed to reduce the redundant hyperspectral band. DRND-DD solves the selection of suitable hyperspectral band via rough set theory. Furthermore, to improve the computation speed and accuracy of the model, based on DRND-DD, this paper proposes reflectance estimation model mining of leaf nitrogen concentration (LNC) for hyperspectral data by using hybrid gene expression programming (REMLNC-HGEP). Experimental results on three datasets demonstrate that the DRND-DD algorithm can obtain good results with a very short running time compared with principal component analysis (PCA), singular value decomposition (SVD), a dimensionality reduction algorithm based on the positive region (AR-PR) and a dimensionality reduction algorithm based on a discernable matrix (AR-DM), and REMLNC-HGEP has low average time-consumption, high model mining success ratio and estimation accuracy. It was concluded that the REMLNC-HGEP performs better than the regression methods.