Classification of Garden Chrysanthemum Flowering Period Using Digital Imagery from Unmanned Aerial Vehicle (UAV)

Abstract

Monitoring the flowering period is essential for evaluating garden chrysanthemum cultivars and their landscaping use. However, traditional field observation methods are labor-intensive. This study proposes a classification method based on color information from canopy digital images. In this study, an unmanned aerial vehicle (UAV) with a red-green-blue (RGB) sensor was utilized to capture orthophotos of garden chrysanthemums. A mask region-convolutional neural network (Mask R-CNN) was employed to remove field backgrounds and categorize growth stages into vegetative, bud, and flowering periods. Images were then converted to the hue-saturation-value (HSV) color space to calculate eight color indices: R_ratio, Y_ratio, G_ratio, Pink_ratio, Purple_ratio, W_ratio, D_ratio, and Fsum_ratio, representing various color proportions. A color ratio decision tree and random forest model were developed to further subdivide the flowering period into initial, peak, and late periods. The results showed that the random forest model performed better with F1-scores of 0.9040 and 0.8697 on two validation datasets, requiring less manual involvement. This method provides a rapid and detailed assessment of flowering periods, aiding in the evaluation of new chrysanthemum cultivars.