Aerial Image Analysis using Convolutional Neural Networks based Multiscale Transformer Fusion Network for Disaster Management

In recent years, disaster management has become an important area in geographically vulnerable regions like southern coast of Java and Indonesia. However, traditional approaches for disaster management had faced challenges like inefficient early warning systems and hazard mapping. To overcome these challenges, this research proposes Convolutional Neural Network based Multiscale Transformer Fusion Network namely (CMTFNet) for disaster management. Initially, data is collected from airborne sensors, which captures radar reflections from the ground through Synthetic Aperture Radar (SAR) and High-resolution optical images. Then, this collected data is preprocessed by using normalization to standardize pixel intensity values among images. After that, semantic segmentation is done by using CMTFNet which accurately classifies each pixel in the image by enabling precise disaster impact assessment. Finally, the segmented output is analyzed for identifying affected areas, support efficient disaster response and recovery planning. The proposed CMTFNet effectively achieved better results in terms of accuracy (96.25%) and False Positive rate (0.08%) respectively when compared with existing Long Short-Term Memory (LSTM).