基于卷积神经网络的微藻种类识别  

Identification of microalgae species based on convolutional neural network

作　　者：崔雪森[1] 田晓清[2] 康伟[2] 朱浩朋 张胜茂[1] JOE Silke 戴阳[1] 樊成奇[2] CUI Xuesen;TIAN Xiaoqing;KANG Wei;ZHU Haopeng;ZHANG Shengmao;JOE Silke;DAI Yang;FAN Chengqi(Key Laboratory of Oceanic and Polar Fisheries,Ministry of Agriculture and Rural Affairs,East China Sea Fisheries Research Institute,Chinese Academy of Fishery Sciences,Shanghai 200090,China;East China Sea Fisheries Research Institute,Chinese Academy of Fishery Sciences,Shanghai 200090,China;Marine Institute of Ireland,Ireland 999014)

机构地区：[1]中国水产科学研究院东海水产研究所农业农村部远洋与极地渔业创新重点实验室,上海200090 [2]中国水产科学研究院东海水产研究所水产品质量安全与加工实验室,上海200090 [3]Marine Institute of Ireland,Ireland 999014

出　　处：《上海海洋大学学报》2021年第4期710-717,共8页Journal of Shanghai Ocean University

基　　金：中央级公益性科研院所基本科研业务费专项(2018GH13);中国水产科学研究院基本科研业务费专项(2020TD68)。

摘　　要：对微藻的光学图像进行采样,并结合国内外专家对微藻鉴定的经验知识,制作了微藻图像数据集,并进行数据增强处理。借助深度学习的原理和方法,构建基于卷积神经网络结构的深度学习模型(AlexNet),对模型进行训练,并利用5折交叉验证方法确保模型的稳定性。结果表明,模型的训练精度为97.86%±1.63%,测试精度为85.86%±0.80%,达到了预期效果。利用AlexNet模型训练得到的参数,对预留的280个样本图像进行实际测试,7个藻种的平均精确度、平均召回率和调和平均数分别为83.3%,84.4%和83.3%,表明深度学习方法是鉴定微藻的一种有效方法。In this study, optical images of 7 microalgae were sampled. Based on the experience and knowledge of experts at home and abroad on identification of marine microalgae, an image data set labeled with algae names was made and data enhancement was carried out.With the help of the principles and methods of deep learning, the AlexNet model based on the structure of convolutional neural network was designed and trained.The 5-fold cross validation method was applied to ensure the stability of the model.The results showed that the average training accuracy of the model can reach 97.86%±1.63%and the average testing accuracy can reach 85.86%±0.80%. By using the parameters obtained from AlexNet model training, the reserved 280 sample images were actually tested.The average accuracy, average recall rate and average F1 Score of the 7 algal species were 83.2%,84.4% and 83.3%, respectively.It was indicated that the deep learning method is an effective way to identify marine toxic algal species.

关 键 词：微藻 卷积神经网络 自动识别 

分 类 号：S946.3[农业科学—水产养殖]