On the 8th, the Ministry of Oceans and Fisheries National Institute of Fisheries Science announced that it has developed technology to automatically identify 10 species of major fish living in the waters around Dokdo.
Through joint research with Pusan National University, the National Institute of Fisheries Science built learning data for 41 species of fish using approximately 130,000 underwater video images taken at five points near Dokdo. By applying the object detection model (YOLOv11), the institute conducted identification experiments on 10 major fish species inhabiting Dokdo and confirmed a high accuracy with an average detection rate of 84.8%. The object detection model is a deep learning-based model that quickly detects the location and type of multiple objects in images.
As a result of analyzing underwater videos taken from 2014 to 2024 using artificial intelligence (AI) technology, the identification accuracy by species showed that the surmullet and black rockfish were at 94.8%, the rock bream at 93.7%, the mullet at 92.6%, and the flatfish at 92.2%, achieving over 90% accuracy in major fish species.
The task of identifying fish species visually in underwater videos is challenging due to various constraints such as light transmission, the irregular movements of fish, and camouflage, consuming a lot of time and effort. The AI-based fish identification technology developed this time overcomes these limitations and achieves high accuracy, significantly enhancing the level of domestic underwater video analysis technology.
Choi Yong-seok, head of the National Institute of Fisheries Science, noted, "Based on the AI-based species identification technology, we plan to expand this technology not only for species identification but also for quantitative surveys such as estimating the density of fishery resources." He added, "We will continue research so that AI underwater video analysis technology can be widely used in various fields of fishery resource surveys."