Aiming at the problem that the solid-state fermentation of liquor depended on the artificial experience to judge the quality of Daqu without quantitative judgment standard, a set of Daqu quality detection system based on machine vision was studied. The dynamic threshold segmentation method was used in this system, RGB transform Lab color space, CNN convolutional neural network and other methods were used to extract three kinds of visual information characteristics of Daqu, the geometric parameters, color and crack. Therefore, the corresponding relationship between the visual information characteristics and the quality of Daqu was established, and the quality of Daqu was judged according to the established relationship. The experimental results showed that the measuring accuracy of the system was 1mm. At the same time, the system could accurately identify and extract the visual characteristics of milky white hyphae, Monascus and Daqu surface fractures of relative sections. Through 1 000 experiments, the accuracy of the identification of Daqu visual information could reach 99.0%, which couldmeet the requirements of the related wine production.