摘要
针对水产养殖中的精准投喂问题,以大西洋鲑(Salmo salar)为研究对象,提出一种基于鱼体运动特征和图像纹理特征的鱼群摄食活动强度量化方法,进行鱼类摄食行为识别研究。利用自适应背景差分及光流法得到运动鱼体的速度、转角,并通过信息熵统计速度和转角的分布,之后通过灰度共生矩阵提取能量、熵、对比度、相关性和逆差距5个图像纹理特征值。最后,结合鱼体运动特征及图像纹理特征,对鱼类摄食行为进行识别和检测。实验结果表明,该方法的识别准确率达到了94.17%,相较于单一特征检测本研究的检测精度更高。
In view of the problem of precise feeding in aquaculture,a method for quantifying the fee-ding activity intensity of shoal was proposed to recognize feeding behavior of fish based on the motion feature of fish body and image texture with the help of rainbow trout.Firstly,the speed and angle of the moving fish are obtained by adaptive background subtraction and optical flow.Moreover,the distribution of the speed and angle are studied by information entropy.Then five image texture features,energy,entropy,contrast,correlation and homogeneity,are extracted by gray-level co-occurrence matrix.Finally,feeding behavior of fish is recognized and detected by combining the motion feature of fish body with image texture.The experimental results showed that the recognition accuracy of this method reaches 94.17%,higher than that of single feature.
作者
黄志涛
何佳
宋协法
Huang Zhitao;He Jia;Song Xiefa(College of Fisheries, Ocean University of China, Qingdao 266003, China)
出处
《中国海洋大学学报（自然科学版）》
CAS
CSCD
北大核心
2022年第1期32-41,共10页
Periodical of Ocean University of China
基金
国家重点研究发展计划项目(2017YFD0701701,2019YFD0900503)资助。
关键词
鱼类行为
摄食行为
计算机视觉
光流
图像纹理
灰度共生矩阵
fish behavior
feeding behavior
computer vision
optical flow
image texture
gray-level co-occurrence matrix
