Detection and classification of tea buds based on deep learning

Abstract

The detection and classification of tea is the premise to realize the automation and intelligence of the famous and high-quality tea picking. The tea buds and tender leaves, as the raw materials of famous and high-quality tea, have similar colors to older leaves, so tea buds can only be picked manually at present. To solve the problem of detection and classification of different grades of tea in mechanical picking for famous and high-quality tea, this paper proposes a detection and classification approach of a two-level fusion network with a variable universe. This approach combines the rapid detection ability of YOLOv3 and the high-accuracy classification ability of DenseNet201 to realize the accurate detection of tea buds. Furthermore, the influence of the shooting angle of the camera on the detection result is compared under the two conventional shooting styles, and the corresponding dataset is established for famous and high-quality tea. The experimental results show that the detection accuracy of the proposed approach is 95.71% for the side-shot tea buds, which is 10.60% higher than the detection accuracy of the top-shot tea buds. This research has certain theoretical and practical significance for intelligent and accurate picking of famous and high-quality tea.