請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/40722
標題: | 應用頻譜影像分析蜜蜂色覺 Analyses of Honeybee Color Vision Using Spectral Images |
作者: | Shou-Shan Chiang 蔣壽山 |
指導教授: | 林達德(Ta-Te Lin) |
關鍵字: | 色彩視覺,色彩編碼,色彩拮抗,分類,蜜蜂,多光譜影像, Color vision,Color coding,Color opponency,Honeybee,Classification,Multi-spectral image,Affinity propagation, |
出版年 : | 2008 |
學位: | 碩士 |
摘要: | 研究使用影像處理技術,模擬並分析蜜蜂色彩視覺神經處理機制,以解釋在蜜蜂色彩視覺系統的色彩拮抗編碼(color opponent coding, COC)模型中,神經電生理實驗記錄與色彩分辨行為實驗結果之間的對應關係,前者已知有超過十九種的組合,後者卻僅有兩種。在影像擷取的方面,透過可接受紫外光與可見光的CCD攝影機搭配依照蜜蜂光感受器對波長敏感度特製的濾鏡,拍攝蜜蜂在大自然中所見的花朵或是蜘蛛,得到在紫外光、藍光以及綠光範圍的影像。在影像處理與分析部份,根據蜜蜂色彩視覺的神經處理機制合成影像。在二十七種可能的神經處理方式中,我們專注於多光譜影像的神經處理機制,因此剔除屬於窄域神經的七種組合方式,針對廣域神經與色彩拮抗神經的二十種組合,並且經由線性與非線性的計算方式進行數位影像合成。以影像相關係數與紋理特徵的差異作為比較影像相似程度的根據,再使用affinity propagation分群方法進行分析。分析結果顯示,二十種COC編碼合成影像的分群結果為兩大群,且行為實驗所得兩種組合分屬兩群,因此雖然有多種色彩拮抗編碼組合,經過神經處理機制之後,視覺訊號卻只有兩大類,符合行為實驗與神經電生理實驗所得到的結果。 The objective of this research is to simulate and analyze the neural mechanisms of honeybee color vision using image processing technique, and to explain the relationship between the intercellular recording data and the color discrimination function. There are over 19 COC patterns in the former but only 2 patterns in the later. In order to acquire images matching to the spectral sensitivities of the honeybee's photoreceptors, a CCD with ultraviolet sensitivity was employed with specific UV filter, blue light filter, and green light filter. The images of flowers and spiders which are usually seen by honeybees in the nature were acquired and analyzed. Image fusion with those spectral images according to the neural mechanisms of honeybee color vision was employed in this study. There are 27 patterns of color-coding neurons, including hypothesized patterns and real recorded neurons. Because the multi spectral neural mechanisms are more important, 7 patterns (narrow-band neurons) are ignored while the remained 20 patterns (board-band and color opponent neurons) were used for images fusion with linear or nonlinear COC patterns. After computing the correlation and the difference of texture features between two images in the image sets, these images were clustered by 'affinity propagation' clustering algorithm. The experimental results reveal that there are essentially 2 clusters in the 20 COC patterns. This results is consistent with the finding from the behavior experiments in the existing literature. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/40722 |
全文授權: | 有償授權 |
顯示於系所單位: | 生物機電工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-97-1.pdf 目前未授權公開取用 | 8.28 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。