近红外光谱的苹果内部品质在线检测模型优化

利用近红外光谱技术在线检测水果内部品质的关键是获取精度高稳健性好的定量分析模型。研究开发了短波近红外光谱苹果品质在线检测系统,试验时苹果样本传输速度为5个/s,以漫反射方式采集,有效光谱范围为500~1100 nm。经光谱强度标准化校正后,有比较的采用遗传算法、连续投影算法和蚁群优化算法等提取特征变量,分别建立偏最小二乘模型,同时分析了这三种方法提取光谱特征变量的搜索机制。特征变量提取方法建立的预测模型所用变量显著减少,预测效果均优于全光谱模型,且能提高运算速度,增强模型的稳健性;其中又以蚁群优化算法的模型预测能力最佳,预测集相关系数R为0.9358,预测均方根误差RMSEP为0.2619。研究结果表明,近红外光谱结合特征变量提取方法可以建立高效的苹果可溶性固形物含量在线检测模型,在产业化应用方面具有很大潜力。

Model Optimization for the On-line Inspection of Internal Apple Quality by Shortwave Near-infrared Spectroscopy

The critical part in the application of near-infrared (NIR) spectroscopy for the on-line inspection of internal fruit quality is to build quantitative analysis models with good robustness and high accuracy. A system based on shortwave NIR spectroscopy for on-line inspection of apple quality was developed. The spectra were collected in diffusion reflectance mode within the wavelength range of 500~1100 nm, and the conveyor belt speed was fixed to five samples per second. After the band intensity was normalized, genetic, successive projection, and ant colony optimization (ACO) algorithms were employed to select characteristic variables, following which the respective corresponding partial least square (PLS) models were constructed, and the spectral variable search mechanisms of these three methods were analyzed. Compared with the full spectral model, the predictive models built on the variable selection methods all exhibited better predictive performance with fewer variables, improved computational speed, and enhanced robustness. The best predictive performance was found in the model built using ACO-PLS, where the correlation coefficient of the prediction set was 0.9358 and the root mean square error of prediction was 0.2619 oBx. The results of this study demonstrate that NIR combined with variable selection methods can build an efficient model for the on-line determination of the apple soluble solid content, and has great potential for industrial application.