Deep Learning-Based Semantic Segmentation Techniques and Their Applications in Remote Sensing

Abstract

The recent few years have witnessed advancements in the field of computer vision owing to the development of more precise deep learning algorithms, the availability of abundant data, and more computing power. This has led to newer developments in various fields related to computer vision, and remote sensing is one of them. Satellite imagery is a vast source of information about the earth and its environment, but it is also inherently challenging to draw useful conclusions from it owing to the presence of complex spectral and spatial details. Satellite image analysis demands pixel-level accuracy as there is a semantic meaning associated with each pixel in the image. But most of the work on semantic segmentation has been performed on natural images. This chapter presents the state-ofthe-art techniques for semantic segmentation based on deep learning and their applications in satellite imagery. The various challenges faced by the researchers in the application of these techniques to satellite imagery are also discussed.