Quantification of petal patterns and colours of genus Sinningia (GESNERIACEAE)

• Published in: Biosystems engineering 2020-09, Vol.197, p.324-335
• Main Authors: Hung, Tzu-Ting, Hsu, Hao-Chun, Kuo, Yan-Fu
• Format: Article
• Language: English
• Subjects: Flower colour; Fully convolutional network (FCN); Petal pattern; Sinningia
• ISSN: 1537-5110; 1537-5129
• DOI: 10.1016/j.biosystemseng.2020.07.008

Plants belonging to genus Sinningia are popular ornamental plants with diverse petal patterns and fascinating petal colours. The patterns and colours of petals are crucial horticultural properties that determine a plant's commercial value. Generally, the patterns and colours of petals are evaluated by experienced horticulturalists. However, manual evaluation can be subjective and the decision criterion could fluctuate due to fatigue. This work proposed a method to automatically quantify the petal patterns and colours by investigating 11 species of Sinningia. The images of the ventral petal of flower specimens were captured using flatbed scanners. The regions of interest (ROI) were defined and segmented from ventral images. Subsequently, a fully convolutional network was applied to the ROI for automatically segmenting variegated parts (i.e., spots, strips, and plaques) from the background. The patterns and colours of petals were then analysed. The proposed method can be used for the automation of petal pattern rating, which generally is performed using naked-eye observations. •A method was proposed to quantify petal colours and patterns.•A fully convolutional network (FCN) was trained to segment patterns of petals.•The patterns were categorised into 3 categories: spots, strips, and plaques.•The method is tested on the ventral petals of 10 species of genus Sinningia.•This approach is objective and precise compared with the method in DUS test.