New method of mussel survey by using high-resolution acoustic video camera-ARIS and deep learning
Fan Zhao, Katsunori Mizuno, Shigeru Tabeta, Takato Asayama, Hiroki Hayami, Yasufumi Fujimoto, Tetsuo Shimada
Abstract
Due to water transparency, water depth, and higher labor demand, conventional methods for the underwater survey (e.g., optical sensing and quadrat survey) have their limitations. Thus, to overcome these barriers, this paper proposes a method of acoustic sensing which uses the high-resolution acoustic video camera-ARIS to visualize the lake bottom and investigate the distribution of mussels. Newly underwater sensing method produces near-video quality acoustic images for constructing the map by Image Mosaic Operation, which can be helpful for assessing the status of mussels. Convolutional Neural Network(CNN) shows its help in the detection and classification of mussels in this study. Meanwhile, the accuracy and efficiency of the well-trained deep learning model manage to improve this research. Through the field survey, the proposed method successfully obtained the distribution maps of mussels in Lake Izunuma.