Search

To evaluate the quality of Daqu, the change of moisture content during the fermentation process is an important indicator. To improve the accuracy of measuring the moisture content, we studied a measuring method that used a near‐infrared hyperspectral imaging technique. In the process of using multigranularity cascade forest algorithm to do deep learning on the detection task, two improvements were made to improve detection accuracy. One was to establish a multigranularity cascade forest model with various scanning window combinations to solve the problem of distribution offset in the prediction results. The other was to do continuum removal processing to obtain enhanced spectral data from the sample spectral data. The results showed that the multigranularity cascade forest was markedly better than a support vector machine and a backpropagation neural network. The coefficient of determination of the prediction set (R2) and root mean square error of the Prediction set reached 0.9977 and 0.0021, respectively. The study results indicated that the hyperspectral imaging technique could achieve highly accurate detection and distribution visualization of the moisture content of Daqu during the fermentation process. Practical applications: Daqu, a flavoring agent and fermentation starter, plays a vital role in the solid‐state liquor brewing process. During the fermentation process, moisture content is an important indicator for quality evaluation of Daqu. Therefore, it is highly important to detect the moisture content in Daqu quickly and accurately. The hyperspectral imaging technique, as an emerging detection technology, has advantages that traditional detection methods do not have, so that the disadvantages of traditional detection can be overcome. In our study, the hyperspectral imaging technique could achieve high‐precision detection of the moisture content of Daqu during the fermentation process, indicating that hyperspectral technology could be used to detect Daqu moisture content. Moreover, this method has great potential for real‐time indicators detection in winery for future work. [ABSTRACT FROM AUTHOR]