Inspiration

After consulting with researchers, we understood it takes huge number of well-trained people like fishing observer involved to collect biological measurements. For other boats without observer, we can only get the information by the hand-write logbook. So find an easy way anyone can help to collect the fishing data can be really helpful to understand our current marine resource.

經由跟研究人員討論,才瞭解到漁業作業過程有各種不同的方式,但一直缺乏一個快速有效的方式接收漁船上魚類的基礎測量資料。目前的方式倚靠魚類觀察員或者船長提供的漁獲日誌,都需要大量的人力而且相關的資訊通常無法有效進入漁業的資料庫供後續的追蹤研究。

What it does

We designed a total solution for the fishing logbook system, including auto fish counting and recognizing. First we will setup a camera system which captures the fish photos. There are two modes: "one fish" and "all fish". In the "one fish" mode, the system could recognize the species of the fish and colleting basic information including length and weight. In the "all fish"mode, the system would count how many fish we grab in the photo. In the same time the system could get the date log, GPS info altogether. Users could collect the data by just take shots in the camera system. Other data will automatically input in the system, and update to the server’s database on the fishing boat temporarily. Users could check the data on the server by any mobile devices. When boat is in shore, the server on the boat will synchronize to the research global fishing database automatically. Those data could be access by everyone.

我們設計了一套完整方案,首先會在船上甲板架設攝影機,負責拍攝捕撈上來的漁獲照片, 一張是單隻魚的照片,系統會幫我們辨識這是哪隻魚; 一張是滿滿的魚照片,系統會幫我們計算數量。 當漁獲置於重量感測器的位置時,同時也會顯示重量。 使用者可以將這些自動蒐集到的資料,合併人工撰寫漁獲其他資訊,傳送到本漁船上的server存放,使用者隨時都可以藉由手持裝置看到我們的漁獲資料。 當船靠岸的時候,船上的server會自動同步到網際網路上的研究中心資料庫,讓全世界的使用者都可以自由存取。

Challenges we ran into

We meet some image recognition challenges. First, in the “all fish” mode, we would count the total number of fish. When fish cover other fish, the system could not judge that is a fish. It will cause the counting error. Solution: We need to put the fish one by one on the desk.

Second, The fish scales will reflect the light. That couldn't know the feature on the fish. Solution: It needs to find some adapt threshold for binarize method.

Third, in fish recognition, we use the algorithm Local binary pattem(LBP) to extract the fish features, and combine the machine learning to training fish feature data and recognizing fish species.

  1. 我們遭遇到了一些影像辨識的問題,比如計算滿滿的魚的照片的數目時,魚身會彼此互相遮住特徵,所以會無法判定是一隻魚,所以會少算。 解決方法:魚都要平均擺放在甲板上,不能夠重疊。
  2. 魚鱗可能會因為反光而在影像上看不出特徵 解決方法:未來要找到適當的取threshold的方法。

3.在魚類辨識的部分,我們使用Local binary pattem(LBP)演算法來進行魚類特徵的建立,再藉由機器學習的方式加以訓練及辨識。

Built With

Share this project:
×

Updates