Classification of Low Quality Underwater Objects Using Convolutional Neural Networks and Transfer Learning
Arun A. Balakrishnan, Bijoy M. S, M. H. Supriya · OCEANS 2022 - Chennai · 2022
Classification of low quality, degraded underwater images using deep learning based architectures are proposed in this paper. Deep learning based algorithms outperform traditional algorithms due to the vast availability of data. Performance evaluation of the proposed models indicates the transfer learning approach has better accuracy of 86% compared to CNN. Xception is the base model used for transfer learning. Data augmentation and Early stopping mechanisms are also incorporated to get the best performance from the proposed models.