Objective: This study focuses on realizing the automatic classification of liquor flowers and then improving the real-time and stability of liquor picking. Methods: The machine vision combined with convolutional neural network was used to replace human eyes for liquor picking. Comparing with many image classification methods, the superiority of the improved algorithm was verified. Results: The results showed that the classification accuracy of the model based on the improved Vgg16 convolutional neural network plus transferring-learning method was up to 96.69%. Conclusion: This method can be used in the real-time classification of Baijiu hops stably.