Establishment of a Daqu Grade Classification Model Based on Computer Vision and Machine Learning

Area of Science:

• Food Science and Technology
• Computer Vision
• Machine Learning

Background:

• Daqu grade significantly impacts Baijiu quality.
• Current Daqu grading methods are subjective, labor-intensive, and inefficient.

Purpose of the Study:

• To develop an automated, objective Daqu grade evaluation system.
• To enhance the efficiency and accuracy of Daqu classification for light-flavor Baijiu.

Main Methods:

• Image segmentation techniques (thresholding, morphological fusion, K-means clustering) for Daqu image extraction.
• Feature selection using Random Forest-Mean Decrease Accuracy (RF-MDA), RFE, LASSO, and ridge regression.
• Machine learning models (SVM, LR, RF, KNN, stacking) for Daqu grade classification.

Main Results:

• Morphological fusion achieved high image segmentation performance (96.67% accuracy).
• Random Forest (RF) models excelled in classifying Daqu-P, Daqu-F, and Daqu-S (96.67% accuracy).
• A combination of RF-MDA and a stacking model showed superior performance in distinguishing Daqu-P from Daqu-F (90.00% accuracy).

Conclusions:

• The proposed computer vision and machine learning model offers an efficient and objective solution for Daqu grade evaluation.
• This approach provides valuable theoretical and technical support for the Baijiu industry.
• Automated Daqu grading can overcome limitations of traditional methods, ensuring consistent quality.