Abstract: A significant compromise can often be found between the generalization of models and the insufficient original data of tea leaves in the field of deep learning, leading to a substantial decline in the detection of tender tea shoots. In this study, a Tea DCGAN (tea deep convolution generative adversarial networks) was proposed with its data augmentation. Initially, a 64×64×64 layer was added into both the generator and discriminator of the DCGAN (deep convolution generative adversarial networks), in order to enhance the perception and learning of low-dimensional features. Additionally, the LeakyReLU (leaky rectified linear unit) function in the DCGAN was replaced with the more linearly controllable ELU (exponential linear units) function, thereby improving the stability and accuracy of model training. Subsequently, a Fake Tea data augmentation framework was constructed using the Tea DCGAN network. The distribution of real tender tea shoots in the existing dataset was analyzed to determine the underlying patterns. According to these patterns, the sample images generated by the Tea DCGAN network were distributed into real tea tree images. A deep learning dataset was formed automatically. Finally, several experiments were carried out on data augmentation, including adversarial generative network ablation tests, and rare tea variety control tests. A comparison was also performed on various data augmentation at different scales. The ablation test results indicated that Tea DCGAN performed best, in terms of the FID (Frechet Inception Distance) metric. Especially after 100 000 training epochs, the FID values for the Zijuan and Longjing 43 tea varieties dropped from 322.10 to 265.63, and from 396.38 to 323.09, indicating the significantly high enhancement in the quality of the generated images. Fake Tea framework was outperformed over the various experiments of the detection model with multiple data augmentation. Specifically, the Faster R-CNN model was achieved in the mAP of 42.71% and 38.46% on the datasets with 25 Longjing 43 and 25 Zijuan tea pictures, respectively. The performance of the models was improved, as the dataset size increased. But the Fake Tea was consistently maintained on the highest mAP value across all dataset sizes. Notably, when the original dataset consisted of 200 images, the mAP value reached 89.41% suitable for intelligent tea harvesting. Therefore, there were the high effectiveness and superiority of Tea DCGAN and the Fake Tea data augmentation in the tea leaf image generation and object detection. The Tea DCGAN and Fake Tea data augmentation effectively enhanced the data acquisition to avoid the scarcity of samples. The high accuracy of detection was achieved in the tender tea shoot under various scenarios with limited samples.