基于高光谱成像技术的面条中马铃薯全粉含量检测

Detection of Potato Powder Addition in Noodles Based on Hyperspectral Imaging

为了快速检测面条中马铃薯全粉含量,研究近红外高光谱成像技术定量检测面条中马铃薯全粉含量的可能性,自制了马铃薯全粉质量分数在0~35%内随机均匀分布的120个面条样品,在900~2500 nm范围采集高光谱图像,随机选取80个样品作为校正集,分别采用原始光谱和经过6种预处理方法预处理后的光谱建立了偏最小二乘回归、主成分回归、支持向量机回归模型。结果表明经标准化预处理后用偏最小二乘回归建模效果最好,校正集决定系数(R2C)为0.8653,交叉验证集决定系数(R2CV)为0.6914。用回归系数法在经过标准化预处理后的光谱数据中提取了与全粉含量相关的特征波长,建立了马铃薯全粉含量偏最小二乘回归简化模型,校正集决定系数(R2C)为0.8685,交叉验证集决定系数(R2CV)为0.8021,基于特征波长建立的模型效果优于全波段模型,模型效果得到了一定的提高。以剩余40个未参与校正模型建立的样品作为预测集,基于特征波长建立了标准化偏最小二乘回归简化预测模型,预测集决定系数(R2P)为0.8546,模型具有较好的预测能力。结果表明利用近红外高光谱成像技术可检测面条中马铃薯全粉含量,可为马铃薯全粉面条的快速无损检测建立新的方法。

In order to quickly and non-destructively detect the content of whole potato powder in potato noodles,the hyperspectral imaging technology was used to quantitatively detect the content of whole potato powder in noodles.Totally 120 noodle samples with a total potato flour content of 0~35%were self-made,and hyperspectral images of the noodles were collected in the 900~2500 nm spectral range.Totally 80 samples were randomly selected as the calibration set,and the original spectra and the spectra preprocessed by moving average,smoothing SG,baseline,normalize,standard normalized variate,and multiplicative scattering correction were used to establish the partial least squares regression model,principal component regression model and support vector machine regression model.The results showed that the partial least squares regression modeling effect was the best after the standardized preprocessing method.The coefficient of determination of the calibration set(R2C)was 0.8653,the coefficient of determination of the cross validation set(R2CV)was 0.6914.The characteristic wavelength was extracted from the spectral data preprocessed by normalize by regression coefficient method,and a simplified model of potato powder content PLSR was established.The coefficient of determination of the calibration set(R2C)was 0.8685.The validation set determination coefficient(R2CV)was 0.8021.The results showed that the model based on the characteristic wavelength was better than the full-band model.Using the remaining 40 samples as the prediction set,the Normalize-PLSR simplified prediction model was established based on the characteristic wavelength.The coefficient of determination of the prediction set(R2P)was 0.8456.The model had good prediction ability.The results showed that it was feasible to use hyperspectral imaging technology to detect the total potato flour content in noodles.