基于视觉注意结合生物形态特征的海洋浮游植物显微图像分析

项目名称： 基于视觉注意结合生物形态特征的海洋浮游植物显微图像分析

项目编号： No.61301240

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 无线电电子学、电信技术

项目作者： 郑海永

作者单位： 中国海洋大学

项目金额： 26万元

中文摘要： 依据藻细胞的生物形态特征通过人工镜检进行分类和计数是目前公认的藻类监测方法。然而，该方法存在专业水平高、分类人员断层、耗时费力等问题而无法应用于大量样品和现场及原位实时快速分析。而目前国内外基于计算机视觉的藻细胞显微图像分析研究存在藻种类别及数据集单一化且难以有效检测和分割细胞目标、提取表示细节特征等难题。本项目针对大量类别形态各异海洋浮游植物显微图像，结合形状、大小、纹理以及角毛、横纵沟、尖顶刺等生物形态分类视觉特征，构建自底向上与自顶向下相结合的视觉注意计算模型检测并提取藻种细胞显著性区域，基于视觉层次感知机制实现藻种细胞目标识别和分类计数。 此项研究结合海洋生物学家的藻种分类知识进行视觉注意建模，模拟藻类专家人工镜检的视觉感知过程，可能成为浮游植物图像分析的一种有效方法，有望实现浮游植物藻种快速、有效地鉴定，为藻类现场及原位实时监测奠定基础。

中文关键词： 视觉注意；显著性检测；图像分割；图像识别；深度神经网络

英文摘要： Currently the authoritative method of phytoplankton monitoring mainly relies on well-experienced phytoplankton operators using microscope for cell identification and classification as well as counting based on characteristics of biological morphology, which need high level professional experts that are becoming less and less. Besides, it is also time-consuming and laborious. So it can not be applied to a large number of samples as well as on-site and in situ real-time analysis. However, the current research on automatic analysis of phytoplankton microscopic images based on computer vision technology only contains a single or a small amount of species and data sets, and is difficult to detect and segment cell target, extract and represent special characteristics effectively. The proposed research focuses on marine phytoplankton microscopic images with a large number of species and various morphologies to build a computational model of visual attention combining bottom-up and top-down mechanisms. The model combines with biomorphic visual characteristics such as shape, size, texture, seta, cingulum and sulcus to detect and extract cell salient regions. Then object recognition and classification counting can be realized by a hierarchical computational model in visual cortex. The proposed research combines the knowle

英文关键词： visual attention；saliency detection；image segmentation；image recognition；deep neural network