Quick Flash Flood Detection Based on Sound Energy Spectrum and Squeeze-and-Excitation Enhanced Deep Learning

Due to fast global climate changes, flash floods are showing increasing impact all over the world, it has become one of the main natural disasters that lead to casualty and economical losses. Early flash flood warning requires intensive water and rain monitoring stations, which is currently not available in most rural and mountainous areas around the world. Therefore, in this paper, we propose a real-time flash flood detection method based on Squeeze-and-Excitation Module enhanced deep learning and sound energy spectrum. Our proposal could effectively detects whether there is a significant increase in the short-time energy mean value of the sound. The collected audio is converted into the form of energy spectrum by operations such as short-time Fourier transform, and then identified by Squeeze-and-Excitation enhanced deep learning model, a detailed introduction of the framework is provided in this paper. The experimental results demonstrate that the method can provide effective real-time detection of flash flooding.