Deep Learning Algorithms for Sonar Imagery Analysis and Its Application in Aquaculture: A Review

With the rapid development of underwater sensor techniques, using acoustic techniques to obtain and analyze underwater images is an effective approach due to the superior nature of sonar, especially in turbidity and dim underwater environment. In recent years, sonar image processing based on deep learning (DL) is being developed and applied to automatic target recognization tasks. This article outlines research of DL-based algorithms for sonar imagery, including denoising, feature extraction, classification, detection, and segmentation. This review shows that sonar image classification is the most studied and transfer learning has put into spotlight. Besides, datasets and applications are listed that can be used in aquaculture. Currently known applications include behavior analysis, biomass estimation, species identification, and habitat survey. Furthermore, smart aquaculture, as one of the important application fields of sonar technique, the challenges and future research directions in aquaculture are discussed. The purpose of this work aims to help researchers find effective DL algorithms for sonar image processing, especially the potential application in aquaculture.