常見食用性貝類辨識之研究

Abstract

民以食為天。相信大家都有吃過熱炒或海鮮的經驗，尤其文蛤更是常見的海鮮料理食材，而它們的貝殼有些人喜歡把玩；有些人喜歡收藏。由於一般人對於貝類的認知不是很清楚，若從書上或是網路上比對資料那是相當曠日廢時，故能發展一套系統能夠準確的辨識，不僅可以快速查詢貝殼的種類，也可以減少人力的辨識。本研究針對數位典藏與數位學習成果入口網中的食用貝類進行研究辨識，共44種。 本研究最好的結果為實驗E，其方法首先為輸入貝類影像；其次，將影像轉為灰階圖；第三，對灰階影像做快速傅立葉轉換；第四，選取在四角之低頻頻率，其大小為 矩陣；最後，利用SVM分類即可辨識出為哪種貝類，其準確率有到100％，平均辨識一張貝類影像所花的時間約為0.044秒。

Food is the first necessity of the people. I believe that everyone has the experience of eating stir-fry or seafood, especially clam as a common ingredient of seafood cuisine. Some people like playing with its shell, while some others like collecting. Ordinary people generally don’t know much about shellfish, and it takes too much time to find information from books or the internet. Therefore, a perfect system for accurate identification can not only provide quick query of shell kinds, but can also reduce manpower in identification. This study focused on the research and identification of 44 kinds of edible shellfish recorded on the website of Digital Taiwan – Culture& Nature. The best result of this study was experiment E: first, input the image of shellfish; second, convert the image into grayscale image; third, perform Fast Fourier Transform with the grayscale image; fourth, choose the low frequencies at four corners, which were 7×7 matrixes; finally, get the result of shellfish identification through SVM classification. The accuracy could reach 100% and the average time spent on identifying a shellfish image was about 0.044s.