Convolutional Recurrent Neural Network-Based Boat Detection Method for Wind Noise Condition

Convolutional Recurrent Neural Network-Based Boat Detection Method for Wind Noise Condition

연구 분야: Artificial Intelligence

학회: 2022 IEEE 11th Global Conference on Consumer Electronics (GCCE)

We have been working on boat detection from environmental sound using a convolutional neural network (CNN). However, it had a problem with accuracy degrading when strong winds blew. In this study, we propose a method for boat detection using deep learning from environmental sound in strong wind conditions. Our proposal method was boat detection via convolutional recurrent neural network using the difference in duration between the boat and wind noise as a cue. The improvement of the proposed method was 0.03 points higher on the average of F-measure than the CNN.