Human Activity Classification in Underwater using Sonar and Deep Learning

Human Activity Classification in Underwater using Sonar and Deep Learning

연구 분야: Infrastructure

학회: WUWNet '19: Proceedings of the 14th International Conference on Underwater Networks & Systems

In this paper, we study the classification of human activity on the surface of a body of water using sonar. In particular, we investigate the classification of three different swimming styles; freestyle, butterfly, and backstroke. Experiments are conducted in a swimming pool to capture acoustic micro-Doppler signatures produced by the different swimming styles. Two acoustic hydrophones are used underwater; one to transmit a single tone signal in the direction of a swimmer and the other to receive the reflected waveform from the swimmer's body. We apply joint time-frequency analysis on the received acoustic signal to extract the micro-Doppler signatures present in the spectrogram. Each of these swimming style activities presents their own unique micro-Doppler signatures. To classify the acoustic micro-Doppler signatures, we explore a deep convolution neural network (DCNN) algorithm. Spectrogram can be considered as an image in which case applying DCNN can serve well for feature recognition purposes. We show that using the spectrogram images the DCNN algorithm can classify different swimming styles performed on the surface of the water with fairly high accuracy. Using the collected data set, we performed experiments where we used 80% of the data for training and the remaining 20% for validation purposes. The DCNN algorithm averaged 93.7% accuracy during training while it had a 90.8% average validation accuracy.