Hyperspectral Image Classification Using Deep Learning Technique

Abstract

Today, the classification process is demanded for modern city planning, agriculture and environmental monitoring, and many other applications. The optimum classification degree is still insufficient so far. The present classification methods for remote-sensing images are grouped according to the features they use into: manual feature-based methods, unsupervised feature learning methods, and supervised feature learning methods. In recent times, the supervised deep learning approaches are extensively introduced in various remote-sensing applications, such as object detection and land use scene classification. In this article, an experiment is conducted using one of the widespread deep learning models, Convolution Neural Networks (CNNs), specifically, AlexNet architecture on a standard sounded hyper spectral dataset, Pavia University (PaviaU). The model achieved an overall accuracy of 91% ± 0.01. A comparison with other different techniques is also introduced.