基于高光谱成像技术快速检测酸肉发酵过程中酸含量的动态变化

Rapid detection of dynamic changes in acid content during the fermentation process of sour meat based on hyperspectral imaging technology

中文关键词:  高光谱成像  酸肉  乳酸  偏最小二乘

英文关键词:hyperspectral imaging technology  sour meat  lactic acid  partial least squares regression

中文摘要:

      目的 快速无损检测酸肉发酵过程中的乳酸、总酸。方法 基于高光谱成像技术, 针对408～1049 nm范围内的光谱反射图像, 采集不同发酵时期的酸肉光谱信息, 提取图像中感兴趣区域内的反射光谱信息后, 采用kennard-stone算法(KS)划分训练集和测试集。原始数据通过标准正态变量转换(standard normal variate transformation, SNV)和多元散射校正(multivariate scatter correction, MSC)进行数据预处理后, 采用偏最小二乘回归算法(partial least squares regression, PLSR)建立模型。采用连续投影算法(successive projection algorithm, SPA)、竞争性自适应重加权采样算法(competitive adaptive reweighted sampling, CARS)以及无信息变量消除法(uninformative variable elimination, UVE)对特征波长进行提取。并基于PLSR分别建立模型, 与全波长预测模型进行对比。结果 酸肉中乳酸的最优预测模型为SNV-CARS-PLSR, 训练集决定系数(R2)为0.9113,均方根误差(root mean square error of cross-validation, RMSECV)为0.7236, 测试集R2为0.9104, RMSECV为0.7342。总酸的MSC-CARS-PLSR模型的预测效果最佳, 训练集R2和RMSECV为0.9307和0.6782, 预测集R2和RMSECV为0.8682和0.6907。结论 利用高光谱成像技术构建的模型可潜在实现酸肉中乳酸和总酸的快速无损检测, 具有潜在的应用价值。

英文摘要:

      Objective To achieve rapid and non-destructive detection of lactic acid and total acidity during the fermentation process of sour meat. Methods Utilizing hyperspectral imaging technology, spectral reflectance images in the range of 408 to 1049 nm were collected to obtain spectral information of sour meat at different fermentation stages. After extracting the reflectance spectra from the regions of interest in the images, the kennard-stone algorithm (KS) was employed to divide the data into training and testing sets. The raw data underwent preprocessing through standard normal variate transformation (SNV) and multivariate scatter correction (MSC), followed by model establishment using partial least squares regression (PLSR). Feature wavelengths were extracted using the successive projection algorithm (SPA), competitive adaptive reweighted sampling (CARS), and uninformative variable elimination (UVE). Models were developed based on PLSR and compared with full-wavelength prediction models. Results The optimal prediction model for lactic acid in sour meat was SNV-CARS-PLSR, with a coefficient of determination (R2) of 0.9113 and a root mean square error of cross-validation (RMSECV) of 0.7236 for the training set, while the testing set yielded an R2 of 0.9104 and RMSECV of 0.7342. The MSC-CARS-PLSR model for total acidity demonstrated the best predictive performance, with training set R2 and RMSECV values of 0.9307 and 0.6782, respectively, and prediction set R2 and RMSECV values of 0.8682 and 0.6907. Conclusion The models constructed using hyperspectral imaging technology have the potential to enable rapid and non-destructive detection of lactic acid and total acidity in sour meat, indicating significant application value.