Prediction of Soil Carbon and Nitrogen Content Using Hyperspectral Image with A New Feature Selection Algorithm

Abstract

The real-time and rapid detection of soil carbon and nitrogen content is of great significance to promote ecosystem carbon balance, ensure the healthy growth of crops, sustainable land use, prediction and early warning of soil pollution. Hyperspectral image is a promising alternative to predict soil carbon and nitrogen elements. Considering some noise and interference information in hyperspectral image, a new feature selection algorithm, extend successive projections algorithm (ESPA), was proposed. Compared with the prediction using full spectrum, successive projection algorithm (SPA), uninformative variable elimination (UVE), genetic algorithm (GA) and manual selection of feature spectra, the prediction was improved by using ESPA selecting the spectral region. The result indicates that the selected spectral region using ESPA is effective in prediction of soil total carbon (TC) and total nitrogen (TN) content, providing an alternative to predict carbon and nitrogen content in soil using hyperspectral image.