Deep Learning Approaches for the Classification of IoT-based Hyperspectral Images

Abstract

The advancement of unique and efficient knowledge-based expert systems has been significantly influenced by the progress in the field of computer-aided learning and testing, which have demonstrated promising results in a wide range of practical applications. Deep learning techniques, in particular, have been used extensively to detect, predict, and classify patterns in remotely sensed urban land cover areas. Machine learning is already lending a hand in monitoring and tracking changes undergoing on the Earth’s surface. However, deep learning enables a higher level of abstraction and provides better extraction of spectral and spatial features in satellite-based imaging analysis. This chapter primarily deals with the application of Internet of Things and deep learning in hyperspectral imaging analysis. For this purpose, some well-known publicly available datasets are used for the classification of different classes present in these datasets. Four main deep neural networks, especially convolution neural network, recurrent neural network (long short-term memory and gated recurrent unit), auto-encoders, and generative adversarial network have been used for the experimentation purpose. A comparative analysis of these classification techniques used for finding the accuracy is made. In the end, certain challenges in deep learning are analyzed, along with some of the emerging future research axes.