Title: Intelligent tea picking model integrating YOLOV5 and Fast R-CNN algorithm

Authors: Yafei Li; Xuanzhang Zhu

Addresses: School of Information Engineering, Hunan University of Science and Engineering, Yongzhou, 425199, China ' Information and Network Center, Hunan University of Science and Engineering, Yongzhou, 425199, China

Abstract: Due to the limitations of traditional manual tea picking, an intelligent tea picking model is proposed to enhance efficiency. The model pre-processes images using Gaussian filtering and two colour spaces for tender leaf and background segmentation, optimised by Otsu's algorithm. An improved watershed algorithm segments the tea leaves, while the Zhang refinement algorithm and Shi-Tomasi corner detection determine picking points. Combining YOLOv5 and Fast R-CNN, with ResNet-50 and CBAM for feature extraction, ensures accurate tea recognition. A binocular vision system provides 3D coordinates, and a robotic arm performs precise picking. Results show the YOLOv5s model achieved over 0.8 in accuracy, recall, and average precision, with 97.2% segmentation accuracy, and CBAM enhanced model performance. This model offers a robust solution for intelligent, automated tea picking, supporting the mechanisation of tea production.

Keywords: tea; YOLOV5; Fast R-CNN; RGB; image; binocular vision.

DOI: 10.1504/IJSCC.2025.149363

International Journal of Systems, Control and Communications, 2025 Vol.16 No.4, pp.374 - 391

Received: 14 Jan 2025
Accepted: 09 Apr 2025
Published online: 27 Oct 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article