Automatic Segmentation of Flood Region in Otsu’s/Kapur’s Threshold Enhanced Images using Deep-Learning Scheme

Artificial Intelligence (AI) supported data analytics is adopted in variety of domains to process the data with a guaranteed accuracy. The application of the AI-schemes, like MachineLearning (ML) and Deep-Learning (DL) are commonly considered when a faster and accurate image examination is necessary. Hence, AI techniques are frequently utilized to process gray/RGB images. This research aims to propose a DL-supported segmentation tool to examine the Flood Monitoring Image (FMI) data. The developed system encompasses the following phases: (i) image collection and resizing, (ii) image pre-processing utilizing the Butterfly Algorithm (BA) and Otsu’s/Kapur’s based multi-threshold, (iii) executing DL-segmentation to extract the flood region from the selected image, and (iv) comparing segmented area with the binary mask (BM), and calculating the essential image metrics to validate tool’s efficacy. This study validates the merit of DL-tool on the unprocessed and preprocessed images. The experimental results of this study demonstrate that the VGG-UNet yields superior segmentation outcomes, with better mean value of Jaccard-index (>93%), Dice-coefficient (>95%), and accuracy (>95%) in comparison to other DL-schemes employed in this research