The mechanization of famous tea harvesting is an essential way to develop China’s tea industry. This paper centers on the detection and positioning technologies in famous tea harvesting, systematically reviewing research progress in these domains. In tea detection, traditional methods rely on color space selection and image segmentation, exhibiting limitations such as insufficient accuracy and poor generalization capability. Conversely, deep learning algorithms demonstrate superior detection accuracy and robustness. Current research focuses on enhancing detection accuracy, inference speed, and multi-variety recognition. In picking positioning, depth information measurement technology utilizing RGB-D cameras provides foundational support. Positioning methods have evolved from traditional visual processing techniques to deep learning and point cloud approaches, seeking to overcome challenges including occlusion and irregular growth patterns. Notwithstanding notable technological advancements, existing methods confront three primary limitations: difficulties in adapting to diverse growth stage characteristics, reliance on large-scale annotated datasets, and inadequate occlusion handling. Future research ought to concentrate on three directions: developing highly universal tea bud detection models, refining model training techniques for small-sample scenarios, and improving tea-picking point positioning accuracy under occluded conditions. This review aims to furnish critical references for advancing high-end intelligent tea-picking machinery, thereby facilitating the tea industry’s mechanization and intelligentization.

Tea shoot segmentation is crucial for the automation of high-quality tea plucking. However, accurate segmentation of tea shoots in unstructured and complex environments presents significant challenges due to the small size of the targets and the similarity in color between the shoots and their background. To address these challenges and achieve accurate recognition of tea shoots in complex settings, an advanced tea shoot segmentation network model is proposed based on You Only Look Once version 8 segmentation (YOLOv8-seg) network model. Firstly, to enhance the model’s segmentation capability for small targets, this study designed a feature fusion network that incorporates shallow, large-scale features extracted by the backbone network. Subsequently, the features extracted at different scales by the backbone network are fused to obtain both global and local features, thereby enhancing the overall information representation capability of the features. Furthermore, the Efficient Channel Attention mechanism was integrated into the feature fusion process and combined with a reparameterization technique to refine and improve the efficiency of the fusion process. Finally, Wise-IoU with a dynamic non-monotonic aggregation mechanism was employed to assign varying gradient gains to anchor boxes of differing qualities. Experimental results demonstrate that the improved network model increases the AP50 of box and mask by 4.33% and 4.55%, respectively, while maintaining a smaller parameter count and reduced computational demand. Compared to other classical segmentation algorithms models, the proposed model excels in tea shoot segmentation. Overall, the advancements proposed in this study effectively segment tea shoots in complex environments, offering significant theoretical and practical contributions to the automated plucking of high-quality tea.

In China, tea products made from fresh leaves characterized by one leaf with one bud (1L1B) are classified as “Famous Tea”, which has better taste and higher economic value, but suffers from a labor shortage. Aiming at picking automation, existing studies focus on visual detection of 1L1B, but algorithm validation is limited to a specific variety of tea sprouting in a certain harvest season at a certain location, which limits the engineering application of developed tea picking robots working in various natural tea fields. To address this gap, a deep learning model DMT (detecting multispecies of tea) based on YOLOX-S was proposed in this paper. The DMT network takes YOLOX-S as a baseline and adds ECA-Net to the CSP Darknet and FPN of YOLOX-S. The average precision (AP), precision, and recall of DMT are 94.23%, 93.39%, and 88.02%, respectively, for detecting 1L1B sprouting in spring; 93.92%, 93.56%, and 87.88%, respectively, for detecting 1L1Bsprouting in autumn. These experimental results are better than those of the five current object detection models. After fine-tuning the DMT network with another dataset composed of multiple tea varieties, the DMT network can detect 1L1B for different varieties of tea in multiple picking seasons. The results can promote the engineering application of picking automation of fresh tea leaves.