文献类型：期刊文献

英文题名：MTAFFB: a multi-task active learning framework for analyzing fish school feeding behavior

作者：Zhou, Tianyu[1];Peng, Xiaohong[1];Zhang, Ying[1,2]

机构：[1]Guangdong Ocean Univ, Fac Math & Comp Sci, Zhanjiang 524088, Peoples R China;[2]Southern Marine Sci & Engn Guangdong Lab, Zhanjiang Bay Lab, Zhanjiang 524000, Peoples R China

年份：2025

卷号：12

外文期刊名：SMART AGRICULTURAL TECHNOLOGY

收录：ESCI(收录号：WOS:001584762100008)、Scopus(收录号：2-s2.0-105015166324)、WOS

语种：英文

外文关键词：Feeding behavior analysis; Feeding intensity classification; Residual bait counting; Multi-task active learning

外文摘要：In aquaculture, analyzing fish feeding behavior is a key focus in both academia and industry. Fish feeding behavior involves not only the fish themselves but also the bait they ingest. Based on an in-depth review of existing studies, this paper proposes a multi-task active learning framework that integrates multi-task learning and active learning to analyze fish feeding behavior. First, a laboratory-based experimental environment for fish feeding behavior was established, and images of fish feeding were collected for subsequent model training. Then, a multi-task active learning framework (MTAFFB) was developed to simultaneously analyze fish feeding intensity and count residual bait. Experimental results show that the proposed framework not only enables comprehensive analysis of fish feeding behavior but also effectively reduces data annotation costs while improving model performance. In the classification task, the accuracy reaches 92%, with precision and recall both at 94.12% and an F1-Score of 0.941. In the counting task, the mean absolute error (MAE) is reduced to 22.44 and the mean squared error (MSE) to 27.93. This method provides a feasible solution for analyzing fish feeding behavior in multi-task scenarios.