碩士 / 國立臺灣大學 / 生物產業機電工程學研究所 / 91 / Monitoring fish growth in ponds is labor intensive and could cause fish stress. This reserch developed an automatatic fish counting and weighing system using machine vision, which could count the total number and the gross weight of fishes and individual fish weights. A dynamic recognition and measurement algorithm is developed to recognize each fish passing a monitoring window and get individual areas of the fish to calculate its weitht hased on a relationship between the weight and area obtained experimentally. The coefficient of determination for a linear and exponential equations is 96.91% and 97.9%, respectively. A platform set at 20cm underwater with a slope is used to guide the fish to swim through the platform at ease while fish images are taken. Because water is shallow at the platform fish images had little overlap. The stress of the fish is also reduced since they do not leave water during the measurement. Four experiments were conducted using 588 tilapia (Tilapia nilotica) in a re-circulating raceway aquaculture system. The fishes swimmed through the platform in twenty minutes. Average accuracy of counting was 97.8%. The estimations of fish weight had accuracies of 92.6% and 93.6%, respectively, for the linear and exponential equations hased on a random sampling of 138 fishes.
| Identifer | oai:union.ndltd.org:TW/091NTU00415008 |
| Date | January 2003 |
| Creators | Hsiang-Tan Lin, 林享曇 |
| Contributors | Yuan-Nan Chu, 朱元南 |
| Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
| Language | zh-TW |
| Detected Language | English |
| Type | 學位論文 ; thesis |
| Format | 101 |
Page generated in 0.0099 seconds