Certain investigations on sar image segmentation and classification using modified deep learning

Abstract

In the past 30 years, Synthetic Aperture Radar (SAR) is widely applied newlinein a multitude of application fields, such as geoscience research, climate newlinechange research, the monitoring of the earth system and environment, and newlineeven planetary exploration. SAR is utilized in the area of earthquake disaster newlinemonitoring and mapping the global surface of other planets, such as Venus, newlinewhich has a thick atmosphere. SAR is considered an active microwave remote newlinesensing instrument. It transmits electromagnetic waves sequentially, collects newlineechoes reflected from ground targets, and stores data to process the images. newlineCompared with optical imaging systems SAR provides high-resolution newlineimages without consideration of daylight, cloud coverage, and weather newlineconditions. This is particularly significant in some high latitude areas like newlinepolar nights and some bad weather conditions. SAR represents a type of newlineactive remote sensing technology that uses microwave electromagnetic newlineradiation to produce and send data to the surface of a target location. SAR newlineimaging is frequently used in national security applications since it is newlineunaffected by weather, geographical location, or time. newlineThe first approach is to process SAR images, that experiments four newlineprominent image segmentation algorithms, and a hybrid method to segment newlineSAR images that shows better efficiency is proposed. First among them is the newlineOTSU thresholding method which segmented the image with less interest in newlinethe texture component. Further, Adaptive Thresholding (ADT), Artificial Bee newlineColony (ABC) segmentation, and Discrete Wavelet Transforms (DWT) are newlineexperimented. Wavelet transforms give better segmentation results compared newlineto thresholding methods. newline