PLS-DA优化模型的马铃薯黑心病可见近红外透射光谱检测

PLS-Discriminant Analysis on Potato Blackheart Disease Based onVIS-NIR Transmission Spectroscopy

马铃薯黑心病是一种马铃薯主要内部缺陷,严重损害薯条、薯片、全粉等加工制品的质量和产率。目前对马铃薯的分级主要侧重于外部品质检测,针对内部缺陷检测的研究很少。旨在开发一种马铃薯黑心病的快速无损检测技术,为此搭建了马铃薯可见近红外透射光谱分析平台,分析健康与黑心病马铃薯的透射光谱特性并优化光谱判别模型参数。基于现有马铃薯分级线和复享PG2000高速光谱仪,采用左右透射方式(光源与光纤探头位于分级线果盘左右两侧),采集470个马铃薯(其中健康薯234个、黑心薯236个)的透射光谱图,建立偏最小二乘判别模型(PLS-DA),并利用主成分分析(PCA)与光谱形态特征相结合的方法选择特征波长,优化模型。分析发现,健康薯与黑心薯的可见近红外透射光谱在吸光度值和光谱形态特征方面均存在明显区别。黑心薯的平均光谱吸光度值高于健康薯(650~900 nm范围内),但黑心薯的平均光谱曲线较为平缓,无明显吸收峰,而健康薯平均光谱曲线在665,732和839 nm附近有明显吸收峰,并且健康薯与黑心薯的平均光谱差值在705 nm处达到最大值。基于PLS-DA法建立了马铃薯黑心病判别模型,对黑心病的判别效果显著,分类器特性曲线(ROC)下面积(AUC)值为0.9942,黑心薯识别总正确率能够达到97.16%,RMSE CV和RMSE P分别为0.28和0.26。此外,成功利用PCA与光谱形态特征相结合的方法对模型进行简化,最终得到由6个波长(658,705,716,800,816和839 nm)组成的特征波长组合,简化后的模型总正确率能够达到96.73%,接近全波段模型判别水平。研究表明,左右透射的方式能够准确识别黑心马铃薯,实现对马铃薯内部缺陷的快速无损检测。对我国马铃薯产业的发展起到一定的促进作用,为马铃薯内部缺陷在线检测技术的提高提供了重要的理论基础和实践依据。

Potato blackheart disease is an internal defect,which decreases the quality and yield of potato processed products such as fries,chips and whole powder.At present,the classification of potatoes mainly focuses on their external quality,rather than internal defects.The purpose of this research was to develop a fast non-destructive detection technology that could be used to detect potato blackheart disease.A visible and near infrared(VIS-NIR)transmission spectroscopy platform was built for potato detection.The spectral transmission characteristics of healthy and blackheart potatoes were analyzed,and the spectral discrimination model parameters were further optimized.Based on the potato grading line and the PG2000 high-speed spectrometer,the transmission spectra of 470 potatoes,including 234 healthy potatoes and 236 blackheart potatoes,were collected using left-to-right transmission method,of which the light source and the optical fiber probe were located on the left and right sides of the fruit plate of grading line respectively.A partial least squares discriminant analysis(PLS-DA)model was established.Furthermore,the principal component analysis(PCA)and spectral morphological features were combined to select essential wavelengths for model optimization.According to the VIS-NIR transmission spectra,there were significant differences between healthy and blackheart potatoes in absorbance values and spectral morphological characteristics.The average spectral absorbance values of blackheart potatoes in the range of 650~900 nm were higher than that of healthy potatoes.The average spectrum curve of blackheart potatoes was relatively smooth without obvious absorption peaks.However,obvious absorption peaks around 665,732 and 839 nm appeared in that of healthy potatoes.The average spectral difference of blackheart and healthy potatoes reached the maximum at 705 nm.Based on the PLS-DA method,a potato blackheart disease discrimination model was established,which had a significant effect on detecting blackheart disease.The area under the receiver operating characteristic curve(AUC),total discrimination accuracy,RMSE CV and RMSE P of the model were 0.9942,97.16%,0.28 and 0.26,respectively.Moreover,a useful wavelength combination consisting of 6 wavelengths(658,705,716,800,816 and 839 nm)was obtained.The total accuracy of the simplified model could reach 96.73%,which was similar to that of the full-band model.It is shown that the left-to-right transmission method can accurately and rapidly identify blackheart potatoes.The study provides an important theoretical,and practical basis for improving the online detection technology of internal potato defects.