Segmented-Incremental-PCA for Hyperspectral Image Classification

Abstract

The Hyperspectral Image (HSI) through remote sensing contains crucial data about the land objects via adjacent cramped spectral wavelength bands. The classification performance does not appear to be adequate when all of the original HSI features (bands) are used. To mitigate this, band (dimensionality) reduction strategies via feature extraction and feature selection techniques are typically employed to enhance classification efficiency. Despite the widespread usage of Principal Component Analysis (PCA) for HSI feature reduction, it frequently struggles to assess the local beneficial properties of the HSI since it analyzes only the HSI’s global statistics. Therefore, Segmented-PCA (SPCA) and Incremental-PCA (IPCA) are presented to supersede the classical PCA. In this paper, we propose the Segmented-Incremental-PCA (SIPCA) feature extraction approach to exploit the amenities of both SPCA and IPCA. In particular, SIPCA first segments the entire HSI into a number of strongly correlated bands subgroups and then apply the classical IPCA on each subgroup separately. We analyze the proposed SIPCA through experimenting utilizing a per-pixel Support Vector Machine (SVM) classifier over the Indian Pines mixed agricultural HSI classification. Based on the classification accuracy, we manifest that our proposed SIPCA technique (91.22%) outperforms the entire original bands of HSI (87.61%), PCA (88.78%), IPCA (89.171%) and SPCA (90.878%) feature extraction methods.