Classification of Jasminum Sambac Flowering Stages Using Deep Learning

In horticulture, flower is classified based on two types namely, structural and fragrance profiling. Jasminum sambac, one of the most cultivated species in south India, is categorized into five different stages starting from the bud to the senescence stage based on structural profiling. The structure is unique for each stage, and by understanding the petal arrangements and the stem size it is easy and useful for robots to harvest a particular stage of flower. The proposed work attempts to classify the flower based on structural profiling. In this work, the dataset was collected using an Intel RealSense D435i depth camera at Annur, Coimbatore. The dataset contains 3617 images. YOLOv5, the object detection algorithm was used as it suits for the small object detection because of its combination of layers, training techniques. The model was trained and validated in the ratio of 7:3 of the total dataset. The result shows that the model has an average precision and recall of 0.83 and 0.73. The model has achieved an average accuracy of 0.73. Thus, the trained model can be implemented for the future work of harvesting using robotic arms.