How many fish in a tank? Constructing an automated fish counting system by using PTV analysis
How many fish in a tank? Constructing an automated fish counting system by using PTV analysis
复制标题
鱼缸里有多少条鱼？
DOI：
10.1117/12.2270627
复制
发表时间：
2017
期刊：
影响因子：
--
通讯作者：
S. Abe ; T. Takagi ; K. Takehara ; N. Kimura ; T. Hiraishi ; K. Komeyama ; S. Torisawa ; S. Asaumi
中科院分区：
文献类型：
--
作者：
Takahiro Taguchi;Satoshi Kubota;Takuma Mezaki;Erika Tagami;Satoko Sekida;Shu Nakachi;Kazuo Okuda;Akira Tominaga;田口尚弘・久保田賢・目﨑拓真・田上恵里香・関田諭子・奥田一雄・富永明;久保田賢・関田諭子・目﨑拓真・田口尚弘・奥田一雄・小西裕子・富永明;田上恵里香・田口尚弘・久保田賢・目﨑拓真・富永明;S. Abe ; T. Takagi ; K. Takehara ; N. Kimura ; T. Hiraishi ; K. Komeyama ; S. Torisawa ; S. Asaumi
由于逃脱网箱和死亡是水产养殖中经常存在的问题，健康控制和设施管理在水产养殖中很重要。特别是，由于太平洋蓝鳍金枪鱼的高市场价值，开发精确的鱼类计数系统一直是太平洋蓝鳍金枪鱼养殖业的强烈愿望。目前的鱼类计数方法涉及人工计数，准确率较低；此外，由于养殖网箱太大，只能在鱼移动到另一个网箱时才能计数，因此方法繁琐。因此，我们开发了一种应用粒子跟踪测速技术(PTV)对网箱内一群游动的鱼进行自动计数的系统。本质上，我们将游动的鱼看作示踪粒子，通过分析相应的运动矢量来估计鱼的数量。提出的鱼类计数系统包括两个主要部分：图像处理和运动分析，其中图像处理部分提取前景，运动分析部分跟踪个体的运动。在这项研究中，我们开发了区域提取和质心计算(RECC)方法和卡尔曼滤波和卡方(KC)检验这两个主要组成部分。为了评估我们的方法的效率，我们构建了一个封闭的系统，在水箱底部放置了一个带有球面曲面镜头的水下摄像机，并拍摄了一群游泳的日本稻鱼(Oryzias Latipe)的360°视角。我们的研究表明，几乎所有的鱼都可以用RECC方法提取，运动矢量可以用KC检验来计算。当在摄像机的画面内观察到超过180个人时，识别率约为90%。这些结果表明，该方法具有作为工厂化水产养殖鱼类计数系统的潜在应用价值。
Because escape from a net cage and mortality are constant problems in fish farming, health control and management of facilities are important in aquaculture. In particular, the development of an accurate fish counting system has been strongly desired for the Pacific Bluefin tuna farming industry owing to the high market value of these fish. The current fish counting method, which involves human counting, results in poor accuracy; moreover, the method is cumbersome because the aquaculture net cage is so large that fish can only be counted when they move to another net cage. Therefore, we have developed an automated fish counting system by applying particle tracking velocimetry (PTV) analysis to a shoal of swimming fish inside a net cage. In essence, we treated the swimming fish as tracer particles and estimated the number of fish by analyzing the corresponding motion vectors. The proposed fish counting system comprises two main components: image processing and motion analysis, where the image-processing component abstracts the foreground and the motion analysis component traces the individual’s motion. In this study, we developed a Region Extraction and Centroid Computation (RECC) method and a Kalman filter and Chi-square (KC) test for the two main components. To evaluate the efficiency of our method, we constructed a closed system, placed an underwater video camera with a spherical curved lens at the bottom of the tank, and recorded a 360° view of a swimming school of Japanese rice fish (Oryzias latipes). Our study showed that almost all fish could be abstracted by the RECC method and the motion vectors could be calculated by the KC test. The recognition rate was approximately 90% when more than 180 individuals were observed within the frame of the video camera. These results suggest that the presented method has potential application as a fish counting system for industrial aquaculture.
开发 PTV 新算法
DOI：
10.2208/jscej.1996.533_107
发表时间：
1996
期刊：
Doboku Gakkai Ronbunshu
影响因子：
--
作者：
K. Takehara;T. Etoh;S. Murata;K. Michioku
通讯作者：
K. Michioku
一种自动追踪粒子的新算法
DOI：
10.2208/prohe.34.689
发表时间：
1990
期刊：
影响因子：
--
作者：
T. Etoh;K. Takehara
通讯作者：
K. Takehara
通过人工神经网络自动计算鱼群数量
影响因子：
4.5
作者：
Paul F. Newbury;P. Culverhouse;D. Pilgrim
通讯作者：
D. Pilgrim