Satellite image classification methods and techniques: A survey

At the age of artificial intelligence, remote sensing and especially satellite imagery is gaining widespread interest among computer science community in their effort to give machines the ability to recognize their environment through satellite image classification. Imaging satellites provide images of Earth that are collected, analyzed, and processed for civil and military purposes. Indeed, satellite images have many applications in the fields of meteorology, oceanography, fisheries, agriculture, biodiversity, geology, cartography, land use planning, and warfare, etc. Satellite image classification aims at transforming satellite imagery into useable information rather than having an image of a place. It relies on many different approaches and methods that can be applied according to specific circumstances and conditions. These approaches and methods fall into five categories which are supervised classification, unsupervised classification, pixel-based classification, object-oriented classification, and Convolutional neural network classification or CNN classification. The goal of this paper is to present a survey which will review these methods based on their algorithms and techniques, datasets, image resolutions, and image types, and will show and discuss their strengths and weaknesses.

I. Introduction

Satellite imagery system [1] obtains and measures electromagnetic (EM) energy emanating from or reflected by terrestrial objects. The electromagnetic spectrum is a combination of different wavelengths that include visible light, infrared, microwave, and ultraviolet. The main source of electromagnetic energy is the sun [1]. Every object on earth has different reflective forces at different wavelengths. The strength of reflection or radiation helps to identify various objects such as water, trees, buildings, etc. in the satellite image [1].