Convolutional Neural Network Analysis of BG Regale, the Sansevieria rorida Radiation Mutant

Abstract

Phenotypic screening in plant breeding is essential for identifying superior mutant lines for ornamental plant variety development. However, the process is tedious and time-consuming involving both qualitative and quantitative methods. To fast-track phenotypic screening, artificial intelligence (AI) presents a promising tool with higher efficiency and minimal standard error in identifying mutant lines for registration. Thus, in this study, an algorithm for classifying novel ornamental plant varieties is proposed. After training a convolutional neural network (CNN) model with the images of radiation mutant BG regale and control, Sansevieria rorida, comparisons of model loss and accuracy were done with Adam, SGD, RMSprop, and Adadelta optimization techniques. Among the optimization techniques, RMSProp achieved the lowest model loss (3.68%), the highest accuracy (98.44%), precision (98.44%), recall (98.39%), and F1-score (98.39 %). In this “proof-of-concept” study on the Philippines’ PHP 10 M (equivalent to USD 178,700; PHP 1 = USD 55.96) BG Regale, one of the world’s most expensive ornamental plants, we introduce the idea of AI-aided facilitation of large-scale phenotype selection during mutation breeding and varietal screening for SEPOPs (or super-expensive Philippine ornamental plants priced at more than PHP 100,000 per leaf).