高光谱无损识别野生和种植黑枸杞

Identification of Wild Black and Cultivated Goji Berries by Hyperspectral Image

高光谱图像技术在农产品检测及识别方面有广阔的应用前景。野生黑枸杞经济效益显著,经常被种植黑枸杞冒充。提出一种利用高光谱图像对野生黑枸杞无损快速识别的方法。主要内容和结果如下:(1)共采集256份(野生、种植各128份)黑枸杞在900~1700 nm范围的高光谱反射光谱,每份平均光谱作为此样品的光谱;(2)采用标准正态变换(SNV)对采集的光谱预处理;基于Kennard-Stone法,按照校正集和预测集比例为2∶1对样品划分,用连续投影算法(SPA)对光谱进行降维处理,提取特征波长30个;分别将全光谱和SPA提取的30个特征波长作为模型输入,建立支持向量机(SVM)、极限学习机(ELM)和随机森林(RF)识别模型。(3)结果表明,在识别野生黑枸杞模型中,基于全光谱和SPA建立的SVM,ELM和RF模型校正集识别率均高于98.8%,基于全光谱和SPA建立的SVM,ELM和RF模型预测集识别率均高于97.7%。基于全光谱(FS)建立的三种识别模型略优于基于SPA建立的三种识别模型。但从简化模型方面,SPA提取的特征波常数仅为全光谱的11.8%,大大降低了模型运算量。三种模型中,基于随机森林模型无损识别野生黑枸杞效果最好,均达到100%。研究表明,利用高光谱图像技术结合分类模型可快速识别野生黑枸杞。

Hyperspectral image technology has a broad application in the detection and identification of agricultural products.Wild black Goji berries have remarkable economic benefits,and are often impersonated by growing black Goji berries.A nondestructive and fast identification method for wild black Goji berries using hyperspectral image technology is proposed.Obtained results were as follows:(1)a total of 256 samples of black Goji berries(Wild,Growing,128 each)in the range of 900~1700 nm were observed,and each average spectra were used as simple spectra.(2)spectral is preprocessed with standardized normal variate transform(SNV)based on the Kennard-Stone(K-S)method,the calibration set and prediction set samples ratio were observed in 2∶1(pairs).However,the spectra were found reduced in dimension by the successive projections algorithm method(SPA),and the 30 characteristic wavelengths extracted by the full spectra(FS).Then the 30 characteristic wavelengths and the full spectra are used as model inputs,the support vector machine(SVM),extreme learning machine(ELM),and random forest(RF)recognition models were established.(3)In the identification of wild black Goji berries models,the results showed that the calibration identification rate of SVM,ELM,and RF model with reference to FS and SPA were higher than 98.8%,and the prediction set samples rate of SVM,ELM,and RF model were also higher than 97.7%.The identification model of FS was slightly better than the identification model of SPA.However,the characteristic wave constant extracted by SPA is 11.8% less compared to FS,which eventually reduces the calculated model.RF identification model was reported better compared to SVM,and ELM,RF identification rate is 100%.The study has shown that the use of hyperspectral image technology combined with classification models can quickly identify wild black Goji berries.