Muh. Taufik Hidayat, Fhatiah Adiba, Muhammad Yahya, Dyah Darma Andayani, Andi Baso Kaswar
Flooding is one of the most common natural disasters, especially during the rainy season. Flooding on highways can cause material losses and casualties, as well as disrupt community mobility and adversely affect the economic sector. Early flood detection can help reduce the impact of these natural disasters. This research was designed to detect the occurrence of flooding on highways by comparing the performance of two Convolutional Neural Network (CNN) architectures, namely MobileNet and VGG16, for the detection and classification of flood images on highways. This research used a dataset consisting of 2,000 images, with 1,000 images for the flood class and 1,000 images for the non-flood class. The results showed that the MobileNet model performed better than the VGG16 model. The MobileNet model has an accuracy of 99% and a shorter computation time, which is only 9 seconds, while the VGG16 model has an accuracy of 96% and takes 69 seconds for computation time. Based on the results of this study, it can be concluded that the MobileNet model can be used for early detection of flooding on highways. It is proven to have high accuracy and a shorter computation time; therefore, it can be used in real-time flood detection systems. © 2024 IEEE.
Universitas Negeri Makassar, Department of Computer Engineering, Makassar, Indonesia; Universitas Negeri Makassar, Department of Automotive Engineering Education, Makassar, Indonesia
