Recognition and Positioning Method of Famous Tea Buds Based on Improved YOLOv8n Model

Abstract

To address the issue of poor accuracy and low efficiency in tea bud picking recognition with the obstruction of tea buds and leaves, a detection algorithm utilizing an improved YOLOv8n network model is proposed in this study. The improved YOLOv8n model incorporates an adaptive feature reconstruction recognition algorithm and an adaptive sparse activation convolution algorithm into the backbone network, based on the YOLOv8n model. The experimental results show that the precision, recall rate, F1 score, and mean average precision of the improved YOLOv8n model are 90.23%, 84.54%, 87.29%, and 88.62%, respectively, which are higher than those of other models. The improved YOLOv8n model achieves a reasoning speed of 97 frames/s. Consequently, the improved YOLOv8n model is suitable for detecting tea buds and exhibits robustness in harvesting scenarios, offering a theoretical foundation for high-quality image processing in intelligent tea picking systems.