Underwater Acoustic Signal Recognition with Small Transfer Learning-based Training Samples

Le résumé fourni par la source

To address the challenges of underwater acoustic signal recognition under small sample training conditions, we propose a deep learning model based on transfer learning. The proposed model integrates a pretrained VGGish system with bidirectional long short-term memory (BiLSTM) to achieve high-performance recognition of underwater acoustic signals. Several experiments conducted on the DeepShip dataset demonstrate that the proposed model outperforms the existing deep learning models in accuracy. With 20% of the dataset used as the training set, the proposed model achieved an accuracy of 84.18%, compared with the baseline accuracies of 68.82%, 56.17%, and 43.59%. Moreover, comparisons regarding the use of transfer learning in model construction and the implementation of BiLSTM highlight the superiority of the proposed transfer learning approach compared with BiLSTM in recognizing underwater acoustic signals.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.