TeaBudLiteNet: a lightweight network for detecting tea leaf buds

Accurate and real-time detection of tea leaf buds is a fundamental requirement for intelligent tea harvesting and smart agriculture. However, achieving high detection accuracy for small targets under complex tea plantation environments remains challenging, particularly for deployment on resource-constrained devices. Existing object detection models often suffer from excessive computational complexity or insufficient performance when applied to such scenarios. Therefore, it is necessary to develop a lightweight detection framework that balances detection accuracy and computational efficiency.

To address these challenges, this study proposes a lightweight object detection model named TeaBudLiteNet. The model introduces a novel C2f_PConv module, which integrates the computational efficiency of PConv with the non-linear feature representation capability of the C2f module, achieving an effective trade-off between accuracy and efficiency. In addition, the SimAM_Slice attention mechanism is incorporated to enhance feature weighting across different scales, thereby improving small target detection. The Focaler-IoU_Inner regression loss function is further employed to dynamically optimize sample importance and accelerate model convergence, enhancing generalization and adaptability. Experimental results demonstrate that TeaBudLiteNet outperforms mainstream detection models in terms of accuracy, model size and inference speed. Specifically, the model achieves a precision of 90.47%, representing an improvement of 2.08% over the baseline. The parameter count is reduced to 1 912 947, achieving more than a 90% reduction in model size compared to conventional approaches, at the same time as maintaining a high inference speed of 227 frames per second.

The proposed TeaBudLiteNet effectively reduces model complexity at the same time as preserving high detection accuracy and real-time performance. Its lightweight architecture and superior efficiency make it highly suitable for deployment on resource-limited smart agricultural devices. By providing an efficient and robust solution for tea leaf bud detection in complex environments, this study demonstrates significant potential for automating tea harvesting and contributes to the advancement of intelligent agricultural systems. © 2026 Society of Chemical Industry.