Research on Individual Recognition and Matching of Whale and Dolphin Based on EfficientNet Model
Siwen Wu, Jiale Wang, Yihan Ping, Xuan Zhang
Abstract
Due to the particularity and complexity of the marine environment, the identification and detection of marine organisms has been the focus of researchers all over the world. In recent years, with the rapid development of marine biological imaging technology and biometric recognition technology, it is possible to recognize and match individual organisms, such as fingerprint and face recognition has been widely used. In this paper, a retrieval model combining EfficientNet and Additive Angular Margin loss function is proposed to track individual whales and dolphins over time. In the process of data preprocessing, the key feature parts are trimmed by YOLOV5 model, and the data set is supplemented by geometric transformation and color transformation. Experimental result show that EfficientNet B6 model achieves the best performance, while ensuring the accuracy, the operation efficiency is relatively high. Specifically, the MAP@5 value of EfficientNet-B6 model matching result is 0.768, which is 1.59% and 1.05% higher than EfficientNet-B0 and EfficientNet-B5, respectively.