草莓品质光谱检测方法比较研究

Comparative Study of Strawberry Quality Spectrometry

为快速、无损检测草莓色度及糖度,采用近红外光谱和高光谱成像技术对草莓的色度和糖度进行光谱分析,采用偏最小二乘法构建预测模型,并对这两种光谱检测方式进行简单的对比分析。近红外光谱试验采用近红外光谱仪结合SpectraSuite光谱采集软件对草莓进行近红外光谱信息提取,高光谱成型技术采用高光谱成像系统结合高光谱图像采集软件Hyper Spectral Image以及图像处理软件HSI Analyzer采集草莓图像,并利用ENVI软件从图像中提取高光谱信息,最后利用The Unscrambler9.7软件对草莓的色度和糖度进行这两种光谱的建模和预测。结果表明:对于红色度值,通过近红外光谱建模得到的判定系数R^2=0.9913,校正均方根误差RMSEC=0.1313,预测均方根误差RMSEP=0.1307,通过高光谱建模得到的R^2=0.9894,RMSEC=0.1559,RMSEP=0.1528;对于可溶性固形物含量,通过近红外光谱建模得到的R^2=0.9917,RMSEC=0.1092,RMSEP=0.1028,通过高光谱建模得到的R^2=0.9849,RMSEC=0.1489,RMSEP=0.1397。通过分析发现绝大多数样本的残差值都在±1.0之间,检验集样本真实值和预测值之间的有很强的相关性,通过两种光谱建立模型的各指标数据均能达到要求。近红外光谱技术相比于高光谱成像技术更加稳定和准确。

In order to detect the chromaticity and sugar content of strawberry rapidly and non-destructively, near-infrared spectroscopy and hyperspectral imaging technology were used to analyze the chromaticity and sugar content of strawberry, and the two spectral detection methods were simply compared and analyzed. Near-infrared spectroscopy was used to extract the near-infrared spectra of strawberry by near-infrared spectroscopy and SpectraSuite spectroscopy. Hyperspectral imaging was used to synthesize strawberry images by hyperspectral imaging system combined with hyperspectral image acquisition software Hyper Spectral Image and image processing software HSI Analyzer, And the hyperspectral information was extracted from the image by using ENVI software. Finally, the color and sugar content of strawberry were modeled and predicted by the Unscrambler9.7 software. Experimental results showed that the red degree value, R^2=0.9913, RMSEC=0.1313, RMSEP=0.1307, R^2=0.9894, RMSEC=0.1559, RMSEP=0.1528 which is obtained by near-infrared spectroscopy and hyperspectral modeling. For the content of soluble solids, R^2=0.9917, RMSEC =0.1092, RMSEP =0.1028, R^2=0.9849, RMSEC =0.1489, RMSEP =0.1397, which is obtained by near-infrared spectroscopy and hyperspectral modeling. Through the analysis found that residual values in testing samples are between±1.0, through the two established model data can meet the requirements, and near-infrared spectroscopy has higher detection accuracy with lower deviations than the hyperspectral model.