Research On Daqu Feature Extraction Algorithms Based On Hyperspectral Technology

In order to promote the development of koji-making technology in the solid-state brewing of Luzhou-flavor liquor from traditional manual production to automatic production,the rapid detection of Daqu fermentation quality is particularly important.At present,most liquor-making enterprises use manual inspection of Daqu fermentation quality,with high potential manpower,time and economic costs,and no objective quantitative evaluation index,which cannot meet the requirements of Daqu fermentation state monitoring in automatic production of Qufang.In this paper,Daqu in solid-state fermentation of Luzhou-flavor liquor was taken as the research object.The data acquisition location is the koji-making workshop of Yibin Liuchixiang Limited Company.Based on hyperspectral technology,the feature extraction algorithm of Daqu sensory and physical and chemical indexes is studied.The main research contents are as follows:(1)The principle of hyperspectral imaging is studied in the band range of 935-1720 nm.According to the appearance,internal composition and fermentation mechanism of Daqu,the hardware selection of the hyperspectral data acquisition system was carried out,including hyperspectral camera,electronically controlled mobile platform,light source and so on.At the same time,choose rhe reasonable lighting mode,set the best software acquisition parameters,complete the construction of Daqu hyperspectral data acquisition system,and collect high-quality,undistorted Daqu hyperspectral data.(2)To study the spectral feature extraction algorithm of Daqu.According to the spectral characteristics,the black-and-white correction of Daqu hyperspectral data was made firstly,then three common spectral pretreatment methods were applied to the corrected hyperspectral data.The spectral parameters of Daqu after pretreatment were correlated with the moisture content of Daqu measured by physical and chemical methods through regression algorithm,and the prediction model of moisture content of Daqu was established.(3)According to the image characteristics of Daqu,the correlation between the texture characteristic parameters and moisture content of Daqu was explored.It was found that the contrast(CON)parameters of grayscale symbiotic matrix were correlated with the moisture content to some extent.The optimal band was searched and associated with the moisture content for modeling to verify the modeling effect of other wavebands and ROI areas.(4)The sensory index detection method of Daqu hyperspectral data was studied.Covariance matrix determinant method and spectral correlation coefficient method were used to calculate the combination ranking among the bands of Daqu hyperspectral data.The combination of bands with the highest score was selected.The RGB three channels was matched with the combination of bands by visual interpretation,and color images were synthesized.After pre-processing the pseudo-color image of Daqu,the sensory indexes such as size,color and crack of Daqu were detected by dynamic threshold segmentation,RGB conversion Lab color space,CNN convolution neural network and other methods,and the quality evaluation table of Daqu was established by combining the moisture content of daqu.