Hyperspectral image classification using principle component analysis and convelutional neural networks

Abstract

Hyperspectral image (HSI) classification is a hot topic in the field of remote sensing data analysis due to the vast amount of information comprised by this type of images.The highest dimensionality of HSI enhances the computational complexity which affects the overall performance. Hence, the dimensionality reduction plays a vital role to enhance the performance while processing the Hyperspectral images. A Dimensionality reduction technique is proposed by this work as a first approach. This technique is applied using Principle Component analysis (PCA), which extract informative features suitable for data representation and classification. This work was condacted to determine and evaluate the performance of two different methods used for classification of three datasets: Indian Pines dataset, Pavia University dataset and the Salinas dataset. These methods are separated as a traditional Machine Learning based on: SVM with two kernels and a Deep Learning based on: U-Net and a pretrained Tansfer Learning with U-Net. The proposed approach is tested on the three datasetsts, The performance analysis results on the Deep Learning techniques have improved accuracy and performance compared to the Machine Learning techniques. Keywords : Hyperspectral Images, Dimensionality Reduction Technique, PCA, Traditional Machine Learning, Deep Learning, Transfer Learning, SVM, U-Net.