SEGMENTATION OF HIGH-RESOLUTION SATELLITE IMAGERY USING A HYBRIDIZATION OF COLOUR-BASEDAND K-MEANS CLUSTERING TECHNIQUES FOR RESCUE MISSION

Abstract

Rapid response to natural disasters, like floods, is essential to reducing casualties and suffering. Rescue teams must have quick access to reliable data. Satellite imagery offers a lot of information that can be analysed to help identify disaster-affected areas. In order to detect and manage natural disasters, segmentation analysis of satellite images is becoming an increasingly crucial component of environmental and climatic monitoring. Image segmentation, which separates a single image into several homogeneous fragments, enhances pattern recognition. Object placement, lighting, shadow, and other factors can all affect how effective an image segmentation technique is. There is no one method that can be used to segment all imagery, although some strategies have been more successful than others. Individual segmentation techniques have flaws like region rising, initial seed selection, noise and low intensity transition, but combining two or more techniques reduces these flaws and boosts segmentation accuracy. This study proposes a method for flooded area identification and segmentation based on a combination of colour-based and k-mean clustering (KC) segmentation techniques. When comparing the proposed technique (colour-based KC model) with commonly used segmentation techniques like Colour thresholding (CT), Region-based Active Contour (RAC) and Edge-based Active Contour (EAC) segmentation, the proposed method achieved better performance metrics with a 0.8234 Jaccard Index, 0.9234 Dice similarity coefficient, 0.9589 precision, 0.9078 recall and 0.9327 BFscore, which was higher than the other four segmentation techniques and previous works. The results obtained indicated that the proposed technique performed better than existing techniques in detecting and segmenting flooded areas in satellite images. Future work can explore other segmentation methods and test the analysed techniques on diverse satellite images with varying scenarios.