Abstract: This study determined the developmental patterns, influencing factors and fertilization ratios of different developmental stages of ornamental and edible lilies through experiments. The phenological period and flower trait data were collected, and an automated recognition model was established through machine vision to establish a foundation for precise cultivation of ornamental and edible lilies in facilities. The results showed that: (1) During the seedling and bud stages of lilies, the consumption of bulb nutrients promoted plant development, while bulb development stagnated. During the flowering and decline stages, plant development stagnated, and bulbs began to develop. (2) The key indicators that are highly correlated with the swelling of lily bulbs are the weight of bulbs during the decline period, the diameter of bulbs during the decline period, the stem diameter during the seedling period, the number of leaves during the seedling period, the stem diameter during the bud period, the weight of bulbs during the flowering period, the plant height during the bud period, the number of leaves during the bud period, the longitudinal diameter of bulbs during the decline period, and the diameter of bulbs during the flowering period. (3) The application of 20-20-20 water-soluble fertilizer during the seedling and bud stages can increase stem thickness during the seedling stage, number of leaves during the seedling stage, stem thickness during the bud stage, number of leaves during the bud stage, and plant height during the bud stage. The application of 20-20-20 or 15-20-25 water-soluble fertilizer during the flowering and decline stages can increase the diameter and weight of the bulbs. (4) The training results of the phenological period recognition model show that vgg11 has the smallest train loss value of 0.168 and the highest accuracy value of 0.933, indicating that the model is the best. Through error analysis, it was shown that except for the measurement error of the width of the slender and curved filament and stigma, the error of other indicators was relatively small, which can be further optimized and applied to field flower trait phenotype investigation.