[3Yin2] Interactive session 1

[3Yin2-36] Fishing spots estimation from sea surface temperature image using key-point detection model

〇Shuma Nakata¹, Masaaki Iiyama¹ (1.Shiga University)

Keywords:key-point detection, deep learning, fishing spots estimation

In this paper, we propose a pattern-recognition-based method for estimating fishing spots from sea surface temperature (SST) images observed by satellites. Using location and quantity of fish catches as training set, we consider the fishing spots estimation as a key-point detection problem from images, and propose a key-point detection model, which is originally used in human pose estimation. Unlike conventional key-point detection, only some of the key-points in the image are included in the training set, which results in under-detection of the fishery location during inference. To solve this problem, we use a new model with“ ReLU Loss ”in addition to the usual L2 Loss. As a result of performance evaluation using pelagic skipjack catch data, we confirmed that the proposed method outperforms the conventional method.