臺灣大學: 生物產業機電工程學研究所林達德陳秋Chen, ChiuChiuChen2013-03-212018-07-102013-03-212018-07-102010http://ntur.lib.ntu.edu.tw//handle/246246/247756本研究之目的為開發一套監測分析蜜蜂覓食行為之影像系統，並在儘量不影響蜜蜂正常行為之前提下，長期監測、記錄與分析蜜蜂進出蜂箱的相關資訊。本系統在硬體的建構上，使用深色不透明的壓克力板製作成系統外殼，並且使用紅外線投光器產生穩定的光源，以取得良好之影像供後端進行影像處理。其他硬體設備還包含蜜蜂通道設計以及CCD攝影機等。另外，本研究使用Borland C++ Builder自行開發軟體，並且搭配開源碼OpenCV與libsvm等作為系統程式共同開發的工具。在運作時，本系統需加裝於蜂箱外部，系統內之蜜蜂通道能夠限制蜜蜂出入蜂箱時的移動路徑，使得每次蜜蜂的進出都在攝影機的取像範圍內。在個別蜜蜂資訊的辨識上，本研究設計一種圓形標籤貼紙黏著於蜜蜂的背上，此圓形標籤可以透過霍爾圓轉換演算法 (Hough circle transform) 被系統程式偵測到。接著利用標籤上的定位黑點，計算阿拉伯數字或字元符號所在的位置並將之由影像中分割出來。偵測到的數字或字元在經過影像前處理後，再使用支持向量機 (support vector machine, SVM) 分類器之辨識以獲得標籤上的數字或字元資訊。最後再透過本研究開發的演算法，判斷並記錄蜜蜂進出蜂箱的時間點。本系統程式對於蜜蜂標籤的辨識率可達97%以上，而判斷蜜蜂進出的正確率則可達到86%以上。This research has developed an imaging system for the monitoring and analyses of honeybees foraging behavior. The system can do long-term detection and monitor for the entering and leaving information of honeybees at the hive entrance under the premise of minimal influence on their normal behavior. The hardware of system are consists of shell, channels and CCD camera. System shell was made by dark opaque acrylic materials and mounted with static IR projector for robust image process. Two major software include in this research are OpenCV and libsvm, system is developed under the environment of Borland C++ Builder. For practical operation, the system must be installed and connected with the beehive, the channels within the system can normalize the movement of honeybees to make sure that each runs are under the field of view of camera. For monitoring the behavior of each honeybee, a circular label sticker with numbers or letters information is applied to this system. The region of sticker was detected by the method of Hough circle transform. One black dot is marked on sticker for normalization of numbers or letters recognition. Arabic numbers or English letters on sticker are then be identified by SVM number classifier. Finally, our proposed system can record the time information of entering and leaving of honeybees. The identification rate of the labeled sticker of the system program is above 97%, and the accuracy rate for the entering and leaving of honeybees is above 86%.2453387 bytesapplication/pdfen-US蜜蜂影像處理機器學習光學字元辨識支持向量機HoneybeeImage processingMachine learningOptical character recognitionSupport vector machinehttp://ntur.lib.ntu.edu.tw/bitstream/246246/247756/1/ntu-99-R97631005-1.pdf