Research Square (Research Square), Год журнала: 2022, Номер unknown

Опубликована: Авг. 12, 2022

Abstract Underwater acoustic object recognition is becoming attractive given the critical information available. However, this comes at expense of large-scale annotated data, which expensive to collect and annotate. This paper proposes a semi-supervised learning approach SE_RseNet_Decoder recognizing insufficient sample underwater targets. Given goal, we introduce network containing supervised unsupervised modules. Firstly, leverage module recognize labeled signals reduce dimensional feature extraction unlabeled samples. Then, designed as an auxiliary optimize network, uses low-dimensional features restore high-dimensional samples enhance classification ability network. We especially ReLU activation function connect modules that can help find balanced relationship between regression tasks for signals. Extensive experiments on multiple benchmark datasets demonstrate superiority our framework showing proposed achieves best accuracy compared with other approaches few Moreover, experimental results optimal combination variables effect method under variables.

Язык: Английский