應用機器視覺的養殖池魚類自動計量分析系統

An Automatic Fish Counting and Size Analysis System Using Machine Vision

林享曇(Shiang-Tan Lin) ； 張嘉孟(Chia-Mon Chang) ； 方煒(Wei Fang) ； 朱元南(Yuan-Nan Chu)

《農業機械學刊》 15卷1期 (2006/03) Pp. 25-36

https://doi.org/10.30062/JAM.200603.0003

摘要

養殖過程中需要隨時掌握魚群的存活率和成長資訊才能有效管理。現有技術還不能以低人力、低干擾方式作全面性的計量，因此本研究應用機器視覺，開發影像收集和自動化處理技術，對於解決魚群計量的需求提出一個可行的方案。本研究的創新之處在針對水道式養殖系統的特性，研發活動式的水中取像平台，可以由單人操作，在魚不離水的情形下完成低重疊性影像的攝影取像，並開發連續性影像的處理軟體，可運用電腦接收影像記錄，完成自動化的計量和重量分析。本研究開發魚類計量分析系統(Fish Counting and Analysis System, FCAS)軟體，將魚體俯視影像經平滑化、二元化、斷開及閉合運算處理，利用偵測線技術區分重複出現的魚隻影像，即可記錄魚群數量。再利用面積與重量的迴歸分析資料可得到養殖池中魚群的總重量和重量分布。四次試驗結果在計數方面可達到平均95.7%的準確率，在總重量方面可達到平均93.5%的準確率，全池魚群的重量分布結果與人工量測結果相當一致。本研究成果可以協助養殖管理者了解魚群生長的狀況，提供管理決策的依據。

關鍵字

機器視覺 ； 魚類計量 ； 水道式養殖池

並列摘要

The survival and growth information of cultured fish is vital to fish culturists for making efficient and timely management decisions. However, apart from current net harvesting method which is largely carried out only at the beginning and end of a cultural season due to high labor requirements and induced fish stress, there is no efficient way to collect fish growth information for the entire stock during the regular growing season. This paper presents a new technique using machine vision to obtain survival and weight distribution data with low labor cost and induced fish stress. Operated by one person, the technique is currently designed to be best used for raceway culture, although it has the potential to apply to wider situations. A movable underwater platform is placed in the raceway so that when fish swim through the shallow water on the platform, low overlapping fish images could be continuously collected by a camcorder and later analyzed automatically in the laboratory to obtain fish number and w eight. The Fish Counting and Analysis System (FCAS) software developed by this research performs a series of image pretreatment operations and could get rid of repeated fish images in consecutive frames by establishing a detection line in the image window and an appropriate image processing algorithm. Four verification experiments were conducted with an average accuracy in fish numbers of 95.7%. The average accuracy for weight estimation was 93.5%. Weight distribution analysis results also corresponded well with manually collected data. The new technique could provide fish culturists with important growth information as a basis for decision support.

並列關鍵字

Machine vision ； Fish counting ； Raceways