Rainfall Recognition Based on Multi-Feature Fusion of Audio Signals
Xueying Li, Yong He, Anlang Peng, Yao Kai-xue · 2023
Rainfall recognition is an important meteorological forecasting task with significant implications for sectors such as agriculture, urban management, and transportation. Traditional rainfall measurement typically employs mechanical rain gauges, which impose strict environmental requirements, often necessitate manual intervention, and offer limited spatiotemporal resolution. To address this challenge, this study employs rain sound for rainfall detection and utilizes feature fusion techniques, combining multiple features of varying scales to capture multiscale information within rainfall data. Furthermore, it introduces dilated convolutional networks to enlarge the receptive field and enhance the model's understanding of global information. In this study, STM32F7 is employed as the processor for rain sound acquisition, and rainfall events are classified using tipping bucket rain gauges, resulting in the creation of a rainfall audio dataset. This dataset is partitioned into training and testing sets, and a deep learning-based classification model is constructed. Experimental results demonstrate that feature fusion and dilated convolutional networks significantly improve the accuracy and performance of rainfall classification.