Comparative Analysis of Hyperspectral Image Classification Methods

Hyperspectral imaging is an effective tool in various areas of application such as medical, agricultural, Satellite imagery, defense and weather forecasting. The latest hyperspectral image (HSI) sensor technology is able to cover large areas with spatial, visual and temporal adjustment. In order to extract important spectral-spatial information from such HSI data, there is a need of space and time complexity algorithms. In the last few decade different development has been made in this hyperspectral imaging. In this field, HSI classification has key challenges such as size of data, insufficient samples in training and spectral variability. While extracting important information from HSI data, there is a need of efficient methods to extract significant features and the appropriate selection of filters to achieve reliable result. In this paper experimental work done on standard publically available databases Indian Pines for HSI classification using SVM, KNN and CNN. We have proposed wavelet-CNN method where factor analysis is used for reduction of extracted transform domain features. The proposed methodology reduces the space and time complexity of the HSI classification model. The proposed algorithm improves overall accuracy of classification of HSI data up to 99.86% and it also reduces the time complexity to 3.56 seconds.