Feature reduction using the minimum noise fraction and principal component analysis transforms for improving the classification of hyperspectral images

Abstract

Dimensionality reduction is an important milestone in the preliminary process of higher dimensional data analysis. Most of research in the hyperspectral image field deals with data extraction techniques. Each feature extraction technique has its own uniqueness. Though each feature extraction technique has its advantages and disadvantages, using a specific technique may result in significant data loss.  To avoid such problems, mixed reduction techniques are utilized in this research. In the current study, dimensionality reduction was done using PCA, MNF, and a combined PCA-MNF method. Image classification using a minimum distance (MC) method was performed after a dimensionality reduction technique. The results showed that our proposed method increased the accuracy of image classification, outperforming PCA and MNF with an accuracy of 80.77%. The accuracy of image classification using PCA is 40.37%, while it was 77.21% using MNF.