光谱学与光谱分析, 2018, 38 (2): 535, 网络出版: 2018-03-14
高光谱图像识别霉变花生的光谱特征分析与指数模型构建
Spectral Analysis and Index Models to Identify Moldy Peanuts Using Hyperspectral Images
摘要
霉变花生极有可能含强致癌物质-黄曲霉素, 快速识别并分离霉变花生可从源头上阻止其进入食物链, 并降低人类摄入黄曲霉素的风险。 利用可见光-近红外高光谱数据, 通过光谱分析确定能有效识别霉变花生的光谱特征或指数模型。 共获取霉变花生样本253个, 健康花生247个, 并取其霉变(或健康)部位的均值光谱。 在对光谱进行连续统去除后, 首先对其求取了不同步长的一阶微分, 并在可分性较优的光谱区域计算了Area500~650指数; 其次, 用连续小波变换提取了光谱的形状和位置信息, 并利用Indexcwt指数识别霉变花生样本。 结果显示, 指数Area500~650的J-M距离为195, Indexcwt模型的J-M距离为199, 表明霉变和健康花生在构建的指数模型Area500~650和Indexcwt的特征空间可分性均较优。
Abstract
Moldy peanuts are likely to contain a strong carcinogen-aflatoxin. Identifying and separating the moldy peanuts quickly can prevent aflatoxin entering the food chain at the source, and reduce the risk of human ingesting aflatoxin. The study is to determine spectral features or index models to identify moldy peanuts efficiently by spectral analysis in Visible and Near-Infrared (VIR) hyperspectral images. Totally 253 moldy peanuts samples and 247 healthy samples were selected to obtain hyperspectral images, and a mean spectrum was calculated from each peanut kernel to represent the moldy or healthy sample. The continuous continuum removal was carried out on the spectra pixel-by-pixel. The modified first-order differential with different step-length was conducted, and the index of Area500~650 was calculated among dominantly separable spectral region of 500~650 nm. Then, the continuous Wavelet transform was applied to extract the spectral information of shapes and locations. Also, the index of Indexcwt was proposed to extract mold information. Results showed that the J-M distance was 195 in Area500-650 and 199 in Indexcwt, which indicates that the performance of both Area500~650 and Indexcwt is good enough to separate the moldy peanuts from the healthy.
乔小军, 蒋金豹, 李辉, 亓晓彤, 袁德帅. 高光谱图像识别霉变花生的光谱特征分析与指数模型构建[J]. 光谱学与光谱分析, 2018, 38(2): 535. QIAO Xiao-jun, JIANG Jin-bao, LI Hui, QI Xiao-tong, YUAN De-shuai. Spectral Analysis and Index Models to Identify Moldy Peanuts Using Hyperspectral Images[J]. Spectroscopy and Spectral Analysis, 2018, 38(2): 535.