Klasifikasi Area Terdampak Banjir Rob Menggunakan Model Vision Transformer (ViT) Berbasis Deep Learning

Abstract

This research focuses on the development of a classification system for tidal flood- affected areas using the Vision Transformer (ViT) model based on deep learning. Tidal floods are natural disasters that frequently occur in coastal regions such as Belawan, North Sumatra, significantly affecting the local population. The ViT model is employed to classify satellite imagery of the area into two categories: flood and no-flood. The dataset used comprises annotated satellite images that are then converted into mask PNG, processed into labelled patches. These patches are then augmented to enhance model generalization. The training results show a validation accuracy of 99%, while no-flood on unseen data yields an accuracy of 90.78% with an F1-score of 0.9065. These results indicate that ViT has strong potential in detecting tidal flood-affected areas automatically and efficiently. The system is implemented as a web application, where users are able to upload images and receive classification results.