摘要
微藻的种类识别对于水生态监测和保护具有重要意义。传统的藻类识别不仅需要投入大量的人力,且对操作人员的藻类分类学知识和技能要求较高。近年来基于人工神经网络的图像识别技术的快速发展让微藻智能识别成为可能。该研究首次将卷积神经网络模型Inception v3引入微藻种类识别中,通过对25种典型微藻进行迁移学习和参数优化,识别率达到了94.19%。以此为基础开发识别系统,并在无锡太湖、广西牛尾岭水库和福建漳浦某水产养殖厂的水体中进行应用验证,亦取得了较好的识别效果。
The species recognition of microalgae is of great significance for water ecological monitoring and protection.Traditional algae recognition is not only time consuming,but also requires operators to have expertise and skills in algae taxonomy.In recent years,the rapid development of image recognition technology based on artificial neural networks has made intelligent recognition of microalgae possible.In this study the convolutional neural network model Inception v3 was first introduced to the recognition of microalgae species.Through transfer learning and parameter optimization of 25 typical microalgae species,the recognition rate reached as high as 94.19%.Based on this,the recognition system was developed and tested using the microalgae containg water samples from Taihu Lake,Niuweiling Reservoir in Guangxi Province and an aquaculture plant in Zhangpu,Fujian Province for application verification,and good recognition results were also achieved.
作者
杨寿勇
张海阳
李成
李静
张学治
YANG Shouyong;ZHANG Haiyang;LI Cheng;LI Jing;ZHANG Xuezhi(School of Marine Science and Environment Engineering,Dalian Ocean University,Dalian 116023,China;Key Laboratory for Algal Biology,Institute of Hydrobiology,Chinese Academy of Sciences,Wuhan 430072,China)
出处
《环境科学与技术》
CAS
CSCD
北大核心
2020年第S02期158-164,共7页
Environmental Science & Technology
基金
国家重点研发计划政府间国际科技创新合作重点专项(2018YFE0110600)
国家自然科学基金(51909258)
中国科学院知识创新工程重要方向项目(KFJ-STS-ZDTP-055)
中国科学院科学仪器研制项目(YJKYYQ20190055)