HybridSSCN: Analysis Of Hierarchical Feature Learning Architecture Using Blended Conv3D And DepthwiseConv2D For Hyperspectral Image Classification

Abstract

Convolutional neural networks (CNNs) have made it possible to conduct various Hyperspectral Image (HSI) feature extraction and classification tasks at the forefront of technological innovation and development. This paper proposes a PHL framework where P, H, and L, denote PCA (Principal Component Analysis), HybridSSCN (Hybrid Spectral-Spatial ConvNet), and LinearSVC (Linear Support Vector Classifier). The HybridSSCN model is a novel deep learning (DL) architecture that incorporates a Conv3D for spectral-spatial feature learning followed by a DepthwiseConv2D layer for spatial feature learning. HybridSSCN helps in learning efficient complex and hierarchical features and aids in lowering computational costs. The features derived from HybridSSCN are classified using the LinearSVC classifier, which achieves 100% accuracy for all three benchmark datasets with 30% and 10% limited training and uneven data compared to the existing contemporary models.