量化岩桐屬植物之花瓣顏色和花紋

Quantifying Color and Variegated Patterns of Petals in Sinningia

洪子庭(Tzu-Ting Hung)

指導教授 ： 郭彥甫

國立臺灣大學/生物資源暨農學院/生物產業機電工程學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201903795

摘要

岩桐屬的花朵因具有高多樣性之紋理與顏色，成為受歡迎的觀賞植物，而孕育出理想的紋理和顏色也可以提升花朵之價值。花朵的顏色及紋理特徵通常是仰賴有經驗之園藝家進行量化，然而，人為的量化方式不僅耗時且會受主觀認知影響。本研究提出基於影像處理的方法自動量化岩桐屬花辦的顏色和紋理特徵。在本研究中，會先將花瓣自花朵分離，並透過掃描機取得花瓣之影像。接著用影像處理之方式定義花瓣腹側之感興趣區域（ROI）以進行切割，並利用全卷積網路和大津演算法（Otsu）將ROI中之紋理特徵（例如：斑點和條紋）與背景分離。最後將分離出的紋理特徵與ROI於L*a*b*色彩空間的數值進行量化，結果證明了自動化量化花瓣顏色和紋理特徵的可行性。

關鍵字

L*a*b*色彩空間 ； 顏和紋理特徵 ； 全卷積神經網路 ； 岩桐屬 ； 自動化

並列摘要

Sinningias are popular ornamental plants with a fascinating colour and pattern diversity in their flower petals. Breeding desirable floral traits increases the values of the plants. Conventionally, the colour and patterns of petals are evaluated by experienced horticulturalists. However, manual evaluation is subjective and may not reflect on the genetics of floral traits. This work proposed a serial of image-based methods to automatically quantify the colour and patterns of Sinningia petals. In the study, the flower specimens were dissected and flattened. The images of the ventral regions were next acquired using flatbed scanners. The regions of interest (ROI) were then defined and segmented from the ventral regions using image preprocessing. Subsequently, fully convolutional networks and Otsu’s method were applied to the ROI for segmenting the variegated parts (i.e., spots and strips) from the background. The colour values in the CIE L*a*b* space and patterns of the petals were then quantified. The results demonstrated the feasibility of automating the quantification of color and patterns of petals

並列關鍵字

CIE L*a*b*color space ； Color and textural patterns ； Fully convolutional networks ； Sinningia ； automation.

參考文獻

Badrinarayanan, V., A. Handa, and R. Cipolla. 2015. Segnet: A deep convolutional encoder-decoder architecture for robust semantic pixel-wise labelling. arXiv preprint arXiv:1505.07293.

Chen, J., Y. Fan, T. Wang, C. Zhang, Z. Qiu, and Y. He. 2018. Automatic segmentation and counting of aphid nymphs on leaves using convolutional neural networks. Agronomy, 8(8), 129.

De Keyser, E., P. Lootens, E. Van Bockstaele, and J. De Riek. 2013. Image analysis for QTL mapping of flower colour and leaf characteristics in pot azalea (Rhododendron simsii hybrids). Euphytica, 189(3), 445-460.

Dice, L. R. 1945. Measures of the amount of ecologic association between species. Ecology, 26(3), 297-302.

Dillencourt, M. B., Samet, H., and Tamminen, M. 1992. A general approach to connected-component labeling for arbitrary image representations. Journal of the ACM (JACM), 39(2), 253-280.