가두리 양식장에서 를 이용한 어류 모니터링을 위한 ROV 고밀도로 사육되는 양식 어류 탐지에 관한 연구

Abstract

Various methods have been proposed for the development of smart aquaculture. however, fish detection models designed for automated fish monitoring encounter challenges such as variations in illumination, frequent fish object occlusion, fish deformation, low contrast, high noise, and dynamic backgrounds. In this study, to improve the detection performance of systems utilizing ROV(Remotely Operated Vehicle) for monitoring fish farmed in sea cages, a dataset of images of the dorsal side of fish was constructed by directly capturing the dorsal side of fish using ROV in cage aquaculture facilities. This supplementary dataset addressed the limitations of existing fish datasets that did not include the images of the dorsal side of fish. The results of training the fish detection model using the proposed dataset were evaluated, and an mAP of 96.7% was achieved for the test dataset. Additionally, the fish detection model successfully detected individual fish even in cases where densely farmed fish were overlapped or grouped together