摘要： 金枪鱼围网渔船在寻找金枪鱼鱼群过程中，所产生的燃油、人力及物力成本很高，提高鱼群搜索效率、降低入渔成本是当前面临的主要问题。本研究基于YOLOv3模型的计算机视频智能辅助分析与识别程序，实现对金枪鱼鱼群特征的自动识别，以减少寻鱼时间，提高捕捞效率。以上海开创远洋渔业有限公司围网船队提供的金枪鱼鱼群特征视频作为研究对象，对特征视频进行前期处理，构建金枪鱼鱼群特征数据集，并对数据集进行识别训练，将训练好的模型部署在具有推断任务的计算机上进行金枪鱼鱼群模拟识别。结果显示：该模型的识别准确率为68.6%。研究表明，基于YOLOv3的特征识别模型在金枪鱼渔情预报中具有实际的应用价值。本研究结果可为金枪鱼鱼群特征识别提供参考。

关键词: 金枪鱼围网, 鱼群搜索, 目标检测, 特征识别, 深度学习

Abstract: In the process of tuna seiners seeking tuna schools, it incurs high cost of fuel, manpower and material resources. Currently, it has become a common concern to improve fish searching efficiency and reduce the fish catches cost. In this study, computer video intelligent aided analysis and recognition program based on YOLOv3 model was used to realize the automatic identification of tuna shoal characteristics, thus to reduce the time of finding fish and improve fishing efficiency. A characteristic data set of tuna shoal was constructed by pre-processing the characteristic video provided by the Shanghai Kaichuang Deep Sea Fishing Co. LTD and the data set was trained for recognition. The trained model was deployed on a computer with an inference task for simulation identification of tuna shoal. The results show that the recognition accuracy of the simulation was 68.6%. The study proves that the YOLOv3 based characteristic recognition model has a practical application value in the forecast of tuna fishing conditions, and the study results could be referred to for future tuna shoal features recognition.

Key words: tuna seiner, fish search, target detection, feature recognition, deep learning