基于深度学习VGG网络模型的海洋单细胞藻类识别算法  

Recognition algorithm of marine single-cell algae based on deep learning VGG network

作　　者：王羽徵 程远 毕海 于秋玉 刘丹 WANG Yuzheng;CHENG Yuan;BI Hai;YU Qiuyu;LIU Dan(Coloege of Information Engineering, Dalian Ocean University, Dalian 116023, China;Offshore (Dalian) Ecological Development Company Limited, Dalian 116085, China;Smart Vision Dalian Research Institute, Dalian 116085, China)

机构地区：[1]大连海洋大学信息工程学院,辽宁大连116023 [2]近海(大连)生态发展有限公司,辽宁大连116085 [3]智慧视通大连研究院,辽宁大连116085

出　　处：《大连海洋大学学报》2021年第2期334-339,共6页Journal of Dalian Ocean University

基　　金：辽宁省自然科学基金(20180550573);辽宁省博士科研启动基金(2019-BS-031);辽宁省教育厅科研项目(JL201918)。

摘　　要：为更好地对海洋中单细胞藻类进行有效识别,本研究提出了基于改进式VGG16网络模型的单细胞藻类识别算法—AlgaeNet,在传统VGG网络模型基础上,通过减少卷积核数量,并添加BatchNormalization层进行神经网络模型加速。结果表明:在相同试验条件下,本研究中提出的AlgaeNet算法在训练过程中的损失值收敛速度及对测试集样本(卵形小球藻Chlorella ovalis与小等刺硅鞭藻Dictyocha fibula Ehrenberg)的预测准确率上升速度较传统VGG、AlexNet网络模型优势明显,识别准确率可达99.317%。研究表明,基于改进式VGG16网络模型的单细胞藻类识别算法AlgaeNet在单细胞藻类识别领域具有较好的分类识别性能,可实现海洋中藻类的准确识别。In order to better identify marine single-celled algae,a single-cell alga recognition algorithm is developed based on improved VGG16 network—AlgaeNet.This algorithm is derived from the traditional VGG network,reducing the number of convolution kernels,and adding BatchNormalization layer to accelerate the neural network.The convergence rate of the loss value during the training process and the increase rate of the prediction accuracy of the test set samples(Chlorella ovalis and Dictyocha fibula Ehrenberg)are found to be better under the same experimental conditions,compared with the traditional VGG network and the AlexNet network,with prediction accuracy of 99.317%.Experimental results showed that the algorithm had better classification and recognition performance in the field of single-celled alga recognition,with accurate recognition of marine single-celled algae.

关 键 词：卵形小球藻 小等刺硅鞭藻 VGG 深度学习 识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] S931.1[自动化与计算机技术—计算机科学与技术]