海洋中小型浮游生物原位光学观测关键技术研究

The species composition, abundance distribution and their spatial and temporal changes of marine meso- and micro-plankton are the basic foundation of research and applications on marine ecosystem, marine environmental monitoring and service, marine fishery, as well as military. Traditional survey by net and water sampling is hard to meet the demand of in-situ long-time continuous observation and large-scale fast real-time analysis. So the research on in-situ observation and identification of mesh- and micro-plankton has very important scientific and practical significance. Aiming at the issues of in-situ optical observation for marine meso- and micro-plankton, including the incomplete information, big data, bad effectiveness, hard real-time processing, this proposed research first establishes a micro light field covering large range of transmittance, and acquires the information of large depth of field for underwater micro detection by the way of depth image sequences, so that keeps the complete information of targets; then extracts the effective information from the huge data based on Logical Stochastic Resonance, and constructs the in-situ large-scale visual recognition dataset of meso- and micro-plankton; at last designs the corresponding deep neural network to learn and represent the biomorphic features for taxonomy, yielding the identification, classification and counting of species. This project aims to build the method for optical observation and identification of marine meso- and micro-plankton, in order to acquire the on-site species composition and abundance distribution.

海洋中小型浮游生物的种类组成、丰度分布及其时空变化是海洋生态系统、海洋环境监测和服务、渔业生产及军事等多方面开展研究与应用的基础依据，传统网采或水采调查难以满足原位长期连续观测和大量样品快速实时分析的需求，因此针对其开展原位观测与识别的研究具有重要的科学和现实意义。本项目针对当前海洋中小型浮游生物原位光学观测中存在的信息完整性弱、数据量大、成像有效性差、实时处理困难等问题，通过建立适应宽透射率目标成像的显微光场，以深度图像分层获取的方式，实现对水下显微探测区域的大景深信息采集，从而保障目标信息获取的完整性；再利用逻辑随机共振的方法对海量数据进行有效性提取，构建中小型浮游生物大规模原位图像识别数据集；然后融合分类学形态特征设计相应深度学习神经网络，完成种类识别及分类计数。最终建立一套适合于海洋中小型浮游生物原位光学观测和分类识别的方法，可现场获取浮游生物种类组成和丰度分布。

本项目针对当前海洋中小型浮游生物原位光学观测中存在的信息完整性弱、数据量大、成像有效性差、实时处理困难等问题，建立了一套适合于海洋中小型浮游生物原位光学观测和分类识别的方法，完成了现场获取浮游生物种类组成和丰度分布，研究内容、进度、成果等如期按照计划完成，具体为：通过建立适应宽透射率目标成像的显微光场，以深度图像分层获取的方式，实现了对水下显微探测区域的大景深信息采集；对海量数据进行有效性提取，构建中小型浮游生物大规模原位图像识别数据集；然后融合分类学形态特征设计相应深度学习神经网络，完成了种类识别及分类计数。