TY - JOUR
TI - Diverse ocean noise classification using deep learning
AU - Mishachandar, B
AU - Vairamuthu, S
T2 - Applied Acoustics
AB - The alarming rise in the ocean noises due to the ramping up of anthropogenic activities had adversely impacted the marine fauna and altered the soundscape of the oceans. In recent times, Deep learning has gained high significance in assessing this pervasive underwater noise data. Studying the underwater soundscape is crucial in conserving the ocean health and in achieving a “quiet ocean”. As an effort to it, in this paper, we present a deep neural network architecture, Convolutional Neural Network-based ocean noise classification cum recognition system capable of classifying vocalization of cetaceans, fishes, marine invertebrates, anthropogenic sounds, natural sounds, and the unidentified ocean sounds from passive acoustic ocean noise recordings. The challenge is to classify these noises amidst the highly non-stationary sound spectrum of short low grunts to long high peaked vocals, limited sensor receiving range, and the need for a large annotated training data. The proposed method can self-learn the features from the training audio data with no need for feature extraction proving better adaptability to complex audio acoustic signals. Experimental results prove that the proposed system is befitting for classifying ocean noises with 96.1% accuracy. Classification helps in distinguishing natural acoustic systems from artificial acoustic systems.
DA - 2021/10//
PY - 2021
VL - 181
UR - https://www.sciencedirect.com/science/article/pii/S0003682X21002358
DO - 10.1016/j.apacoust.2021.108141
LA - English
KW - Wind Energy
KW - Noise
KW - Fish
KW - Invertebrates
KW - Marine Mammals
ER -