Flood Region Segmentation from Digital Image with Deep Learning Scheme
Abin Antony, Ibrahim Mohammad Khrais, K. Vijayakumar, D Keerthivasan, E. Christopher Siddarth, A Sangeetha, Md. Tabil Ahammed
Abstract
Most environmental monitoring systems use a computer algorithm to process data from different sensors. The results of this processing help to send out alerts when there is a major problem. The goal of this project is to create an autonomous flood region segmentation system that uses deep learning (DL) to analyses digital photos taken by an unmanned aerial vehicle (UAV). This approach helps figure out how bad the water level is in a certain area. This work looked at the VGG-UNet (VUNet) scheme to make the implementation easier and the detection more accurate. The different parts of this tool are: collecting and resizing images and masks; training the UNet using images and masks; dividing the flood area into segments; and comparing the segmented binary section to the mask on a pixel-by-pixel basis to get the metrics needed. We compared the results of VUNet with those of selected UNet models, and the overall results of this study show that the suggested VUNet gives superior Jaccard (95.03±0.23), Dice (98.34±0.02), and accuracy (98.71±0.01) than other selected DL-methods. This proves that the method used works better on the specified database.