摘要
使用体外标记技术可对仿刺参(Apostichopus japonicus)进行种群和个体尺度上的时空行为学研究、种群动态研究、良种繁育、高效采捕方法的研究。由于仿刺参体壁柔软,排异能力较强,使得传统的侵入式标记方法留存率较低;且传统标记对体壁的破坏会导致伤口溃烂,影响仿刺参的正常生活。为研发非侵入性的仿刺参识别技术,本研究利用深度学习中的卷积神经网络模型,对仿刺参图像进行特征提取,该特征能够表征个体独特的体表纹理模式。对50 d内连续拍摄的仿刺参图像进行特征提取并训练分类器后,发现分类器在测试集上最高可达到0.996±0.011的精度;而传统的侵入式标记方法最高只能达到约0.75的精度。对实验仿刺参个体进行个体识别跟踪,使用前25 d的仿刺参图像进行特征提取并训练模型,对后25d的图像进行预测,可达到0.946±0.058的精度。实验结果表明,使用ResNet50卷积神经网络可有效地对仿刺参进行预测,并在时间追踪任务中取得优于传统标记方法的精度。
External tagging methods facilitate the study of the spatio-temporal behavior of Apostichopus japonicus at population and individual levels,as well as population dynamics,breeding,and efficient harvesting methods.However,due to the soft body wall and strong exclusion ability of Apostichopus japonicus,the retention rate of traditional invasive tagging methods is low.Moreover,damage caused to the body wall by traditional tagging methods often leads to wound ulceration,which affects the quality of life of Apostichopus japonicus.In this study,a non-invasive identification method for Apostichopus japonicus is developed by processing images using a deep convolutional neural network model to extract features and characterize the unique body texture patterns of individual specimens.After feature extraction and the training of the classifier on consecutive images of Apostichopus japonicus taken in a period of 50 d,the classifier achieved a maximum accuracy of 0.996±0.011 on the test set compared to traditional invasive tagging methods that only achieved an accuracy of up to 0.75.For individual temporal tracking recognition,feature extraction and model training were performed using images taken in a period of 1 to 25 d.The classifier achieved an accuracy of 0.946±0.058 on the test set consisting of images taken in the periods of 26-50 d.These results indicate that the use of the ResNet50 convolutional neural network can effectively predict the categories of Apostichopus japonicus and achieve a better accuracy than traditional tagging methods in the temporal tracking task.
作者
刘洋
王扬帆
胡景杰
包振民
丁君
常亚青
杨建敏
侯虎
LIU Yang;WANG Yangfan;HU Jingjie;BAO Zhenmin;DING Jun;CHANG Yaqing;YANG Jianmin;HOU Hu(Key Laboratory of Marine Genetics and Breeding,Ministry of Education,College of Marine Life Sciences,Ocean University of China,Qingdao 266003,China;College of Fisheries and Life Science,Dalian Ocean University,Dalian 116023,China;School of Agriculture,Ludong University,Yantai 264025,China;Shandong Huachun Fishery Co.,Ltd,Dongying 257236,China;College of Food Science and Engineering,Ocean University of China,Qingdao 266003,China)
出处
《中国水产科学》
CAS
CSCD
北大核心
2022年第10期1487-1499,共13页
Journal of Fishery Sciences of China
基金
国家重点研发计划蓝色粮仓科技创新专项(2018YFD0901)
国家自然科学基金项目(32072976,31772844)
教育部科学技术研究项目(2021ZL09)
关键词
仿刺参
非侵入式标记
深度学习
疣足
复杂表型
Apostichopus japonicus
non-invasive tagging method
deep learning
papillae
complex trait
作者简介
刘洋(1995-),男,博士研究生,研究方向为全基因组选择育种和生物信息学.E-mail:769277594@qq.com;通信作者:王扬帆,教授,研究方向为基因芯片设计,全基因组选择育种.E-mail:yfwang@ouc.edu.cn;通信作者:胡景杰,教授,研究方向为热带海洋生物遗传学、种质资源评价与开发、良种培育.E-mail:hujingjie@ouc.edu.cn
