Flood Disaster Estimation Using Images and Machine Learning

Abstract

Pakistan is exposed to heavy floods every year, which can cause significant damage to property, infrastructure and loss of life. To minimize the loss, an estimation of the post flood disaster using images of inundated areas and Machine learning is a better solution for improving the efficiency and accuracy of disaster response efforts in the country. The research aims to enhance flood disaster management by utilizing real-time data to measure flood water levels in inundated areas.Estimating floodwater levels is challenging due to different levels and moderate visibility of objects. Building precise flood level maps is crucial for assisting emergency plan activities in the case of a flood. It's crucial to gather data from the disaster region in order to create these maps. In this circumstance, National disaster management authorithy (NDMA) platforms might be helpful information source. In this article, we provide a technique for measuring floodwater using images from (NDMA) sites. If there is no previous experience or understanding of the area where the image was taken, determining how much the items in the image are immersed in water might be one method of estimating the flood level. Several things contribute to the difficulty of this endeavor, including: The size of the items in the photograph may not be understood with certainty;  Different areas of the photographic scene's flood-water may appear at varying heights, and objects may only be just partially apparent since they may be submerged in the water. We provide an approach to address these issues that first identifies classes of items whose sizes are roughly known before using this feature to calculate the water level. We first create a dataset of flood-water images to test the viability of this method, and then we train a deep learning model on it. Finally, we demonstrate how our trained model can accurately estimate flood levels while also recognizing objects. Flood Disaster Estimation using images and machine learning has the potential to make a policy based decision on flood disaster estimation in the country. With the use of this proposed solution one can save the lives and property on bigger scale.