Identification of tea bud with improved DCGAN algorithm and GhostNet-RCLAM network.

As the raw material of high-grade tea, tea bud can only be picked artificially now. For the realization of intelligent picking, tea bud classification is the premise to increase the intelligent degree of the tea industry. To solve the problem of tea bud classification, deep learning technology is employed to identify different types of tea buds. Firstly, image acquisition is performed in Qing Zhen Hong feng Mountain Yun Tea Factory Co., Ltd in Guiyang, Guizhou Province, China. Then, images are generated by an improved DCGAN network and traditional technology, which can increase the scale and diversity of the training samples. Meanwhile, channel locking attention module is embedded into GhostNet network to improve the classification performance of the network. Additionally, the robustness of model proposed in this paper can be improved, with the pretrained on the Oxford-17 flower dataset. It is remarkable that the Focal Loss was introduced to enhance the feature transfer capability of the model during the pre-training process. According to the experimental results, the highest classification accuracy of our method reaches 96.8%. It is significantly higher than in the four comparison experiments. [ABSTRACT FROM AUTHOR]