Separation and identification of mixed signal for distributed acoustic sensor using deep learning

With the application of Distributed Acoustic Sensors (DAS) across various infrastructures, it will play a pivotal role in shaping smart cities in the future. However, the current single-source detection and identification technology might struggle to meet the high precision needs in the intricate environmental conditions of mixed multi-source interference. We propose a new deep neural network-based multi-source signal separation method for DAS and accomplish the separation performance of this method under practical applications. In addition, a new evaluation metric for the separation method is proposed in conjunction with the separation and identification of DAS mixed signals. For mixed signals with different source numbers, the recognizable rate of separated signals can reach 98.33% on average. This study provides a promising solution to the multi-source mixed interference problem faced by DAS in complex environments.