Land Cover Classification of Hyperspectral Image Using Deep Learning Techniques

Abstract

The use of hyperspectral data for land cover categorization is gaining popularity. One of the most significant applications of remotely sensed data is creating accurate land cover maps using satellite pictures. For image categorization, the spectral and spatial information obtained by HSI is crucial. Deep learning (DL) models have just lately emerged as a strong tool for tackling a variety of machine learning (ML) issues. For many image classification applications, convolutional neural networks (CNNs) are the current state-of-the-art. When we applied Deep Neural Network (DNN) to the Salinas and Pavia University datasets we got classification accuracy 88.75 % and 96.20 % respectively. When we applied Convolution Neural Network (CNN) to the Indian pines datasets we got classification accuracy 95.08 %. From these results we conclude that, In DNN technique number of iteration is more compare to CNN technique, DNN takes more time compare CNN to classify the land cover of hyperspectral image. In general we can say that CNN is better DL technique compare to SVM and KNN.