基于计算机视觉的鱼类异常行为建模与识别研究

Fish behavior can provide physical health status and environmental suitability information,is a focus on aquaculture.The conventional manual observation or biological test method is very difficult to achieve timely, accurately find the fish abnormal behavior,has slightly carelessness will cause a great loss.Modeling of fish behavior and abnormal behavior recognition is the key to realize stress state warning.The study uses zebrafish as research animal.Swimming,feeding,clustering,avoidance and breeding behaviors under environmental stress(density,temperature,dissolved oxygen,and ammonia)are studied by computer vision and image analysis technologies.The quantitative representation of the individual and group behaviors of fish is explored.The key characteristics of specific behavior are extracted by the use of high-dimensional data analysis method.The variation between physiological indicators reflect the degree of stress and behavioral patterns is revealed.Then,fish normal behavioral patterns and common abnormal behavioral patterns are established by mathematical modeling methods.And a method for fast, accurate, automatic detection and identification of fish abnormal behavior is realized.This study takes advantage of aquaculture and information technology,a new way of quantitative research of fish behavior is presented.And the theoretical basis and methodological basis for the development of automatic detecting equipment and early warning system for fish stress is provided.

鱼的行为能够提供生理健康状况和环境适宜程度等重要信息，是水产养殖中重点关注对象。常规人工观察或生物测验方式很难做到及时、准确地发现鱼类异常行为，稍有不慎就会造成重大损失。鱼类行为的建模与异常行为识别是实现鱼类应激状态预警的基础和关键。本项目以斑马鱼为研究对象，采用计算机视觉和图像视频分析技术研究鱼在环境胁迫（密度、温度、溶解氧和氨氮）下的游泳、摄食、聚群、规避和繁殖行为，探究鱼类个体和群体行为的数量化表示方法，利用高维数据分析方法进行鱼类行为数据降维，提取表征特定行为的关键特征，揭示反映应激程度的生理指标与行为特征的变化规律，建立能区分鱼类正常行为模式和常见异常行为模式的数学模型，实现快速、准确、无应激的鱼类异常行为自动检测和识别。本项目集水产养殖和信息技术交叉特长，给鱼类行为学定量研究提供了一种新的思路和方法，为研制鱼类应激状态自动检测装备和预警系统提供理论依据和方法基础。

本项目以斑马鱼、大菱鲆等鱼为研究对象，采用计算机视觉和图像视频分析技术研究鱼在环境胁迫（密度、温度、溶解氧和氨氮）下的游泳、摄食、聚群、规避和繁殖行为，探究鱼类个体和群体行为的数量化表示方法，利用高维数据分析方法进行鱼类行为数据降维，提取表征特定行为的关键特征，揭示反映应激程度的生理指标与行为特征的变化规律，建立能区分鱼类正常行为模式和常见异常行为模式的数学模型，实现快速、准确、无应激的鱼类异常行为自动检测和识别。研究期间发表科研论文 7 篇，其中 SCI 检索杂志 1 篇，核心检索6篇，申请发明专利2项。