Novel Applications of Deep Learning in Remote Sensing Satellite Imagery: Natural Hazards and Disasters Risk Management

Abstract

The remote sensing satellite images based applications at worldwide has an important vision to implement and develop advanced space research technology usages and practically natural resources information and monitoring at local scales, district and national levels and provide correct inputs on time for an accurate disaster management and modelling e.g.: fire-forest assessment, floods, crop estimation, geomorphology, natural water resources information, urban changes etc. are simply carried out at high resolution spatial, spectral as well as temporal pixel resolutions. Remote sensing technology is effectively acquiring satellite-based information about the land surface, ocean and earth’s atmosphere using advanced remote sensing-based satellites such as Sentinel series of Sentinel type-1, Sentinel type-2, LANDSAT’s, Worldview -2, Worldview -3 etc. platforms. Remote sensing technology has widely used in military and civic usages. Accurate analysis of remote sensing satellite images is difficult due to the complicate nature of the satellite images. Development of suitable system for the accurate analysis of remote sensing satellite images is very essential. Traditional methods such as manual detection and identification of images and objects from remote sensing technology of the satellite images is very arduous, time consuming and costly. Advanced machine learning algorithms and remote sensing satellite images aids in various applications requires high resolution of spectral and spatial data / information includes agriculture, climate change, disaster management, ecology, environment, forestry, oceanography, transportation, weather, and so on. Machine learning algorithms such as Neural Networks (NN), Support Vector Machine (SVM), Random Forests (RF) etc. through advanced computer vision technology and deep learning algorithms includes CNNs, faster R-CNN, YOLO series to accurately identify and collect relevant features with actual accuracy and high speed.