Enhancing emergency vehicle access in dense settlements of Mumbai using high-resolution satellite imagery: A deep learning approach

Abstract

Unplanned urbanization in developing cities has led to haphazard development, increasing disaster risks in densely populated areas. Despite being more disaster-prone, these dense settlements often lack essential emergency services. Delivering emergency services requires a high-resolution geodatabase of minor roads and their attributes, such as road width. Such information can be a crucial input for quantifying emergency vehicles (EVs) access in an area. To the best of our knowledge, no research has addressed this problem statement, making this paper the first to propose quantifying road width and using it to develop EVs accessibility maps. We propose a two-stage process to achieve these goals. In the first stage, we designed a deep neural network (DNN) to accurately organize built-up features, including minor roads using true-colour high-resolution satellite imagery of Pleiades-1A for Mumbai. In the second stage, we propose an algorithm to calculate road width and then compare it with EVs dimension (width) details to develop high-resolution accessibility maps. The model classified features with around 91.88% accuracy along with 90% Kappa coefficient, while the classification accuracy of minor roads in informal settlements was about 85.80%. We found that informal settlements have much less access to various EVs compared to formal settlements due to mixed signatures. The outcomes presented in this paper can be used as a decision-making tool to develop a geodatabase of road widths and EVs accessibility maps for efficient resource planning, which can ultimately lead to the development of disaster-resilient cities.