影像監測技術評估魚類在生態渠道之行為

Abstract

本研究以雙鏡頭佐以簡單數學運算，設計出能量測魚體長度與魚隻吻部XYZ座標之非侵入性監測方法。 室外實驗地點為苗栗穿龍圳東河排水之生態工程渠段，使用雙鏡頭法對不同河段進行魚種監測，計算各採樣點之魚通量(隻/m3•sec)；魚通量的概念將不同能見度之各採樣點能相互比較，其值受到棲地物理條件、拍攝時段等影響。 貝氏有限混合分佈理論以魚通量作權重，繪製不同魚種之最適福祿數曲線，尋找交點得到不同魚種之適合福祿數範圍。研究發現，福祿數為0至0.01之間為大肚魚優勢區；福祿數0.01至0.20為台灣石[魚賓]佔優勢；福祿數0.20至0.31為大肚魚佔優勢，超過福祿數0.31則台灣馬口魚佔優勢，當福祿數大於0.58，台灣馬口魚之福祿數出現機率乘上權重值已小於百分之ㄧ。魚隻成熟度會影響其魚種在不同福祿數下之表現。 野外各點在影片中由人工觀測魚類棲所行為判定為棲息地的變異係數CV值皆小於1，也就是標準差小於平均值之情況為魚類棲息地的可能性較高；反之若標準差大於平均值之情況則為廊道的可能性較高。 室內實驗以雙鏡頭法於水工實驗渠道中量測台灣馬口魚與吳郭魚(尼羅口孵魚)之長度與游速，魚隻長度平均誤差百分比最高為9.47%(1.23公分)；最低為4.19%(0.18公分)。誤差來源除了濁度與亮度影響人工辨識讀值外，魚體是否彎曲、魚尾巴是否彎曲、與魚身投影線是否與雙鏡頭平行都會影響人工判讀的結果。同時計算魚隻於不同流速下之魚尾擺動頻率，以推算在野生環境下，不同魚種適應福祿數之最高值；研究發現體長較小的於需要更高的魚尾擺動頻率來維持穩定，而台灣馬口魚比吳郭魚更適合高流速環境。最後以增加流速法(Increased Velocity Tests)，量測兩種魚之臨界游速(Critical Swimming Speed)：台灣馬口魚之臨界流速介於0.39m/s(總體長=5.3cm)至0.74 m/s(總體長=14.7cm)之間；吳郭魚之臨界流速介於0.30 m/s(總體長=8.3cm)至0.46 m/s(總體長=13.8cm)之間，以上結果可提供本土魚種ㄧ個設計參數依據。

This study using bi-camera method and simple mathematical operation as a non-invasive monitoring method to develop a XYZ-coordinates to calculate the fish length and fish swimming speed. In the field, the experiment used bi-camera method to monitor different ecological engineering channel sections in Chuan-Long channel in Miaoli County, to recognize the fish species and calculate the fish flux (fish/m3•sec). The fish flux could compare with different sites with different visibility. The fish flux index was affected by different factors such as habitats and different filming time. By using Bayesian finite mixture distribution as the method and the fish flux index as weight values, the experiment estimated the optimum-Froude number curves in different species in order to find the intersection points. The results showed that the Gambusia affinis affinis predominated over other species when the Froude number was 0－0.01. The Acrossocheilus paradoxus predominated over other species when the Froude number was 0.01－0.20. The Gambusia affinis affinis predominated over other species when the Froude number was 0.20－0.31. The Candidia barbata predominated over other species when the Froude number was higher than 0.31. In addition, when the Froude number exceeded 0.58, the appearing probability of Candidia barbata multiplied by its weighted was less than 1%. The fish maturity affected the performances in different Froude number. The coefficients of variance of fish flux were smaller than 1 in the study sites which were judgment artificially to be habitats by behavior of fishes. That was to say, when the standard deviations were smaller than means of fish flux, the sites might more likely be fish habitats but corridors. In the laboratory, the lengths and swimming speeds of Candidia barbata and Oreochromis niloticus niloticus were estimated by bi-camera method in the modeling channel. The maximum error was 9.47%(1.23 cm), and the minimum error was 4.19%(0.18cm). The error occurs not only in turbidity and brightness, but also in the curvature of the fish body and the curvature of the fish tail. Moreover, the accuracy was affected whether the fish was parallel with the bi-camera or not. The laboratory experiment determined the tail-beat frequency in different velocities to estimate the maximum adapting Froude number for two fish species when they were in the field. The result showed that the small fish needed high tail-beat frequency to make the body in a stable situation. In addition, Candidia barbata was more suitable for the fast flow than Oreochromis niloticus niloticus. The laboratory experiment measured the critical swimming speed of two fish species by Increased Velocity Tests. The critical swimming speeds of Candidia barbata were 0.39m/s(total length=5.3cm)－0.74 m/s(total length=14.7cm), and the critical swimming speeds of Oreochromis niloticus niloticus were 0.30m/s(total length=8.3cm)－0.46 m/s(total length=13.8cm). This result can be designing parameters for field fish restoration.