Flooded Area Segmentation on Remote Sensing Image from Unmanned Aerial Vehicles (UAV) using DeepLabV3 and EfficientNet-B4 Model

Climate change caused by global warming results in increased rainfall and has the potential to cause flooding. Floods are natural disasters that often occur in Indonesia and can cause significant losses. In handling floods, effective monitoring of flood-affected areas is needed. Flood monitoring using remote sensing technology, such as the Unmanned Aerial Vehicle (UAV), is compelling. However, developing an accurate semantic segmentation method to map flood areas is still necessary. This research aims to perform semantic segmentation using DeepLabv3 with various pre-trained backbone models such as ResNet50, EfficientNet-B4, and MobileNet on the FloodNet dataset of 398 data. The measurement of segmentation performance will be evaluated using miou, accuracy, precision, recall metrics, and f1-score. The best evaluation results were obtained using the EfficientNet-B4 model with a miou score of 0.481 and an accuracy of 0.90.