Hyperspectral Image Classification Using SWT and CNN

The new paradigm of remote sensing captures huge volume of hyperspectral images of earth concern to various fields in human kind. Processing of such high-volume multidi-mensional image datasets, object detection, feature extraction, prediction and hyperspectral image classification are some of the active issues in current computing arena. So far many methods have been developed to classify hyperspectral images. This paper investigates latest approaches and proposes a model using stationary wavelet transform(SWT), principal component analysis(PCA) and convolutional neural networks(CNN). SWT is used to extract the meaningful features from the hyperspectral data. Later, PCA selects a subset of transformed coefficients for classification. A CNN has been designed to classify the selected transformed coefficients. Experiment analysis of the model is performed with two popular online hyperspectral image datasets, where the model based classification accuracy outperforms to other models being discussed in the paper.