Search

Classification of remote sensing Hyperspectral Image (HSI) using spectral-spatial feature extraction methods has attracted researchers worldwide, giving prodigious achievement. We propose a PSL framework where P, S, and L denote Principal Component Analysis (PCA), SWTNet, and Linear Support Vector Classification (SVC), respectively. SWTNet model includes a 2D Stationary Wavelet Transform (SWT) for extracting shallow spectral-spatial texture features followed by multiple input multiscale 3D Convolutional Neural Networks (CNNs) model for deep feature extraction. The SWTNet is a two-stage combined shallow and deep feature extraction model that extracts spectral-spatial discriminative features with a strong correlation. The multiple-input of the CNN model helps to perform cross-correlation with the input data. The multiscale property eases the problem of loss of details during the convolution process. We classify the extracted features using the Linear SVC classifier. Our experimental results showcase the potential of our proposed PSL framework, which outmatch accuracy results compared to the other state-of-the-art models when examined using limited training samples. [ABSTRACT FROM AUTHOR]