Assessment and Impact of Feature Extraction Methods for Hyperspectral Remote Sensing Image Classification Based on Deep Convolutional Neural Networks

HSI (Hyperspectral Image) consists of more spectral bands, used for the classification of various objects on earth. However, these huge numbers of spectral bands possess redundant information and decrease classification accuracy. To perform classification efficiently, dimensionality reduction approaches are applied. PCA is the frequently used feature reduction technique for data having a huge no of dimensions. This research work has proposed a PCA and Factor Analysis for dimensionality reduction. After the implementation, the extracted features of HSI data from PCA and Factor Analysis to be compared. Also, CNN(Convolutional Neural Networks) with various layers of Convolutional, Pooling, and Fully Connected Layers after decreasing the features to segregate the HSI data. To check the effectiveness of the developed method, testing will be done with benchmarks of HSI data sets like Indian pines, SalinasA Scene, Pavia University Scene, and Kennedy Space Center(KSC).