Hyperspectral feature selection and classification with a RBF-based novel Double Parallel Feedforward Neural Network and evolution algorithms

Band selection is an important preprocessing procedure for analysis of hyperspectral data, which suffers from the vast amount of data and Hughes phenomenon. In recent years, band (feature) selection using Neural Network such as Multi-layer Forward Neural Network (MLFNN), Radial Basis Function Neural Network (RBFNN) and Double Parallel Feedforward Neural Network (DPFNN) becomes a promising method for dimensionality reduction. In the paper, we present a novel feature selection method using the improved DPFNN (IDPFNN), where it is the RBFNN not MLFNN to keep parallel connection with the Single-layer Feedforward Neural Network (SLFNN). The algorithm combines the procedure of structure optimization with that of feature selection using a hybrid algorithm based on Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). An AVIRIS data set is used for evaluation of the performance of the proposed method through the experiments of feature selection and classification. The experimental results show the better effectiveness of the algorithm of IDPFNN.