A novel AI-based model for real-time flooding image recognition using super-resolution generative adversarial network

Abstract

Intensified climate change in recent years has had a global impact, leading to increased precipitation events of short duration and high intensity. This phenomenon poses a severe challenge to urban and underground infrastructure. Accurate detection and location of floodplains and water bodies are essential to ensure informed decision-making and implement proactive measures to minimize risks and losses. Therefore, an automated and efficient flood image detection system is the need of the hour. This study proposes a deep learning–based flood detection system. Images taken by surveillance cameras at intersections are used as input data, making the system well-suited to urban applications. Data augmentation techniques are used to improve the model performance. We demonstrated the practicality of this model by applying it to street surveillance images taken in Taiwan. The developed model quickly and successfully identified the extent and location of flooding with high precision and reliability. The developed model can be used to provide valuable insights for flood management and disaster management agencies. The test findings obtained from this study demonstrate the superior performance of the DeepLabv3 + model compared to the Mask R-convolutional neural network model, which was further enhanced using a super-resolution generative adversarial network. The model achieved remarkable precision with a precision metric score of 84 %, which is also complemented by a recall rate of 91 %. Most notably, the mean Intersection over Union (mIoU) metric reached an impressive accuracy level of 85.8 %. The results of this study highlight the importance of developing advanced flood imagery detection models aiming to considerably reduce the risks and losses incurred by flooding. The application of such a flood image detection system could help increase the ability of a city to cope with flood events caused by climate change.