陈皮年份的高光谱技术鉴别研究

Identification and Classification of Different Producing Years of Dried Tangerine Using Hyperspectral Technique with Chemometrics Models

市场上陈皮以次充好现象时有发生,而年份是衡量陈皮品质的重要指标。研究用高光谱技术结合化学计量学算法,在380~1 023及874~1 734nm两波段对不同放置方式的陈皮进行年份鉴别。为了寻找更合适的波段和模拟实际生产检测中陈皮放置的随机性,采集了四个年份共180个样本在380~1 023及874~1 734nm的正、反面高光谱图像(720幅)。用主成分分析法(principal component analysis,PCA)对陈皮光谱信息进行定性分析,发现不同年份陈皮基于正反面光谱有明显的聚类;而后以回归系数法(regression coefficient,RC)选取陈皮年份相关的特征波段以减少变量;用偏最小二乘判别分析(partial least squares discriminant analysis,PLS-DA)基于全波段、特征波段对三种放置方式(正、反、正反混合)的样本建立模型,最后对特征波段建立线性PLS-DA模型和非线性ELM模型并进行比较分析。研究表明:在380~1 023nm的预测效果大多高于874~1 734nm,基于非线性ELM的判别结果均高于线性PLS-DA模型,准确率最高可达到建模集100.00%,预测集98.33%,陈皮正、反、正反混合三种放置方式预测准确率多数可高于85%,故采用高光谱技术可实现对不同放置方式的陈皮年份进行无损鉴别,为进一步开发便携仪器或在线生产设备提供方法和理论依据。

The shoddy phenomena often occur in the market, and the producing year is an important index to measure the quality of dried tangerine. Thus, this research applied hyperspectral technique in the spectral windows of 380～1 023 and 874～1 734 nm combined with chemometric methods to identify different producing years of dried tangerine. Due to the actual detection of dried tangerine, spectra of front and back of dried tangerine were acquired. Hyperspectral imagesat both sides of a total of 180 samples of four years were collected within 380～1 023 and 874～1 734 nm （720 pictures）. Then principal component analysis （PCA） was carried out on the spectral data, which had a qualitativeanalysis about the dried tangerine. Regression coefficient （RC） was chosen to select the sensitive variables. Partial least squares-discrimiant analysis （PLS-DA） were used to compare the performances of full-spectra and sensitive variables. Finally, the linear PLS-DA model and nonlinear ELM model were established based on the sensitive bands. The results demonstrated that most of the predictive effects in 874～1 734 nm were higher than that of 380～1 023 nm. ELM models were outperformed PLS-DA among all the developed models, the highest accuracy was achieved 100% in the model set, and 98.33% in the prediction set. No matter what kind of placement, the prediction accuracy rate was higher than 85%. Hence, hyperspectral technique with chemometrics models can realize nondestructive identification of various producing years of dried tangerine, which provides a theoretical reference and basis for developing instruments of recognition the dried tangerine in further research.