支持向量机
偏最小二乘回归
人工智能
平滑的
近红外光谱
计算机科学
模式识别(心理学)
发酵
数学
机器学习
化学
统计
食品科学
量子力学
物理
作者
Xianhui Chang,Xingyi Huang,Weidong Xu,Xiaoyu Tian,Chengquan Wang,Li Wang,Shanshan Yu
摘要
Abstract As a critical control point in the processing of Chinese steamed bread, fermentation is mainly judged by traditional methods. The purpose of this work was to develop an intelligent method of dough fermentation monitoring by near‐infrared (NIR) spectroscopy technology. First, Savitzky–Golay smoothing filter was utilized as the best method to preprocess the original NIR spectra for eliminating spectral noise by comparison, and the frame size was 15. Second, the unrepresentative intervals were eliminated from preprocessed NIR spectra using synergy interval partial least squares (Si‐PLS) preliminarily. Based on the selected intervals, competitive adaptive reweighted sampling (CARS) was adopted to further select variables. Then, 13 significant variables were screened from full wavelength variables by utilizing CARS‐Si‐PLS. Finally, the K‐nearest neighbor (KNN) and support vector machine (SVM) monitoring models of dough fermentation state were established. The results showed that the accuracy of training set and prediction set of KNN model were 89 and 86%, respectively, and that of SVM model were 94 and 92%, respectively. By comparison, the performance of two models, it was found that SVM was superior to KNN. This work verified that NIR spectroscopy couple with multivariate calibration models can replace assessor judgment to realize rapid and online monitoring of dough fermentation. Practical Applications The method based on NIR spectroscopy coupled with supervised learning algorithm can be used instead of assessor to judge the fermentation state of dough, so as to avoid subjective influence of assessor on the judgment result. In addition, it can realize the rapid online monitoring of dough fermentation state, which is beneficial to enhance the automation level of Chinese steamed bread processing.
科研通智能强力驱动
Strongly Powered by AbleSci AI