Tea is an important cash crop. Because the tea shoots in the tea garden are very similar to the old leaves, and the environment of tea buds in tea gardens is complicated, the difficulty of picking tea buds is increased. Aiming at the problems that tea buds are tough to identify and the picking points are challenging to locate in the complex environment in reality, this paper proposed a detection model based on Yolov7 for tea bud recognition and key point location. In this model, some conventional convolutional modules are changed to Diverse Branch Block (DBB), which can enrich the feature space and obtain more feature information of tea buds. The GE attention convolution module is incorporated to the neck network to boost the model's use of contextual features, so as to accurately locate the key point of the tea bud. The ELAN module is replaced by P-ELAN module, which reduces the redundant calculation and the size of the model. The outcomes of the experiment demonstrate that the mean Average Precision (mAP) of tea buds and their key points is increased by 3.81% in contrast to the original Yolov7.