Satellite Image Classification Using Deep Learning for Big Earth Data

Satellite images have shown a variety of applications in the last two decades which includes health monitoring of crops, disaster management, urban planning etc. With a rapid growth in the development of artificial intelligence and deep neural networks, there has been a significant growth in research of satellite images using neural networks. There are multiple variants of deep neural networks reported in satellite image research for different purposes. Out of all the networks, the use of convolutional neural networks is quite efficient and popular. This paper reviewed the deep neural networks for satellite image classification reported in the last decade i.e. from 2013 to 2022 with a special focus on recent development. The review is organized from the perspective of evolution technologies in deep neural networks such as use of transfer learning, attention, and transformers. There are still open challenges in the satellite image research such as handling the large dataset, quality of data, and availability of labeled data etc. The paper sketches the methodological advancement and limitation of different deep neural networks for satellite image classification and highlights the possible future research in the domain.