An integrated convolutional neural network and sorting algorithm for image classification for efficient flood disaster management

Abstract

Drones are used for post-flood disaster management and delivering relief goods to flood-affected areas. Autonomous drones are an alternative means of prioritizing assistance due to the lack of available technology and accessibility in many affected areas during floods. This study proposes a machine-learning approach designed and developed for autonomous drones to identify flood-affected areas with image classification. The proposed integrated approach can be used to deliver relief on a priority basis from the most affected areas to the least affected areas considering distance for efficiency. The proposed system uses a combined convolutional neural network (CNN) and sorting algorithm. The Inception v3 and DenseNet CNN approach can effectively detect flood severity. The Inception v3 shows better performance than DenseNet in terms of image classification. The Inception v3 and DenseNet architectures achieve 83% and 81% accuracy in our self-made flood level dataset, respectively. The integrated algorithm is used to sort the data efficiently. This study demonstrates the efficacy of CNN combined with a sorting algorithm for autonomous decision-making in robotic architecture.