Pemanfaatan Model U-Net Berbasis Deep Learning untuk Segmentasi Citra dalam Identifikasi Area Genangan Banjir

Abstract

Abstract— Climate change and increased rainfall intensity are causing a higher frequency and magnitude of flood disasters in various regions. This situation necessitates an accurate, automated, and efficient monitoring system for the rapid detection of flood inundation areas to support emergency response. This study aims to implement and evaluate a U-Net-based image segmentation model for identifying flood inundation areas using aerial imagery. The dataset utilized in this study comprises 3,041 RGB images and 3,041 corresponding ground truth masks. The research methodology includes data acquisition, preprocessing (comprising image resizing, pixel value normalization, and data augmentation techniques such as rotation and flipping to enhance data variance), and data partitioning into training, validation, and testing sets. The model was trained using a batch size of 16, a learning rate of 0.001, and for 30 epochs. The model's performance was comprehensively evaluated using standard segmentation metrics: accuracy, precision, recall, Intersection over Union (IoU), and the Dice coefficient. The results indicate that the U-Net model can segment flood inundation areas with excellent performance. The model achieved a global accuracy of 0.908, precision of 0.895, recall of 0.883, an IoU of 0.800, and a Dice coefficient of 0.889. Analysis of the training and validation curves demonstrated stable convergence without significant indications of overfitting. Moreover, the visual segmentation results indicated a high degree of correspondence between the model's predictions and the actual ground truth masks. Consequently, it is concluded that the U-Net model is effective and robust for identifying flood areas from aerial imagery. It also holds significant potential for integration into artificial intelligence-based flood monitoring systems for disaster mitigation.