Aiming at the problems of low recognition rate and poor robustness of tea buds in complex backgrounds, an improved YOLOv8 tea bud detection algorithm is proposed. By introducing the Swin Transformer self attention mechanism, a CTS feature extraction module is constructed to enhance the global feature extraction capability of the model; Drawing on the idea of multi-scale fusion, constructing the ExFModule module enriches semantic feature information while enabling the network to adaptively select useful features and suppress useless ones; In terms of feature fusion, a BFPAN feature map stitching method is proposed to enable the model to pay more attention to small target features and improve its feature fusion ability. The experimental results show that the improved YOLOv8 algorithm achieves an average accuracy of 93.4% on the tea tender bud dataset, an increase of 4.4% compared to before improvement, and the detection speed remains basically unchanged. It can achieve fast and accurate tea tender bud recognition and detection, and provide technical support for the intelligent picking of tea tender buds.