Black tea withering moisture detection method based on convolution neural network confidence

Abstract

AbstractDeep learning method was applied to rapidly and nondestructively predict the moisture content in withered leaves. In this study, a withering moisture detection method based on confidence of convolution neural network (CNN) was proposed. The method used data augmentation to preprocess the original image. The prediction results obtained by the CNN model were compared with the results of traditional partial least squares (PLS) and support vector machine regression (SVR) models. The results clarified that the quantitative prediction model of the moisture content in withering leaves based on the confidence of convolutional neural network has the best prediction performance. The performance parameters of the optimal prediction model: correlation coefficient (Rp), root‐mean‐square error of external verification set (RMSEP) and relative standard deviation (RPD) are 0.9957, 0.0059, and 9.5781, respectively. Compared with traditional linear PLS and nonlinear SVR algorithms, deep learning method can better characterize the correlation between images and moisture. The moisture‐related information in the image can be extracted to a greater degree by the convolution kernel of the convolutional neural network. The model has better generalization, which can rapidly and nondestructively predict the moisture content in withered leaves.Practical applicationsCNN is increasingly used in food technology. This study solves the problem that the withered leaves moisture content cannot be quantitatively predicted based on the confidence of the proposed CNN. Compared with traditional machine vision methods, our proposed CNN model can retain more original information in addition to the color and texture features of withered leaves. And it can quickly and accurately judge the moisture content without destroying the tissue components of the withered leaves. This study is of great significance to the intelligence of black tea processing equipment. Simultaneously, the proposed model based on deep learning provides a new idea for the intelligent detection of black tea withering process.