基于高光谱成像技术的酿酒高粱品种分类

针对不同品种酿酒高粱分类难、分类精度有待提高等问题,提出了一种结合光谱和图像信息的高光谱成像技术从而对酿酒高粱进行分类的方法。通过采集11类共550个高粱样本的高光谱数据,运用连续投影算法从多元散射校正预处理后光谱中筛选出48个特征波长,再提取图像的灰度共生矩阵作为图像特征,利用纹理特征、全光谱、特征光谱及其结合图像特征分别建立支持向量机、偏最小二乘判别和极限学习机分类模型,最后再采集220个未参与建模样本对所建模型进行外部验证。结果表明,基于特征光谱结合纹理特征建立的支持向量机模型效果最佳,训练集和测试集的识别率分别为96%和95.3%,验证集的识别率达到91.8%,高于单一光谱数据建模效果,说明光谱和图像信息结合可以提高酿酒高粱的分类识别率。该方法为高粱品种的高精度分类和不同酿酒原料的快速无损检测提供了可行的方法。

Classification of liquor sorghum varieties based on hyperspectral imaging technology

In order to solve the problem in classifying liquor sorghum,and to improve the classification accuracy,a method using hyperspectral imaging technology combined with spectral and image information was proposed.The hyperspectral data of 550 sorghum samples from 11 classes were collected,48 characteristic wavelengths were selected using the successive projection algorithm from the pre-processed spectrum of multiple scattering correction,and then the gray level co-occurrence matrix were extracted as image feature,the support vector machine(SVM),partial least squares discriminant analysis and extreme learning machine classification models were established using the texture feature,full spectrum,feature spectrum and their combined image feature.Finally,220 non-participating modeling samples were collected for external verification of the built models.The results showed that SVM model based on the feature spectrum combined with texture feature had the best effect,and the recognition rate of the training set and testing set was 96%and 95.3%,respectively.The recognition rate of the verification set was 91.8%,which was higher than the modeling effect of single spectral data,the effects indicated that the combination of spectral and image information could improve classification recognition rate of liquor sorghum.A feasible method was developed for high-precision classification of sorghum varieties and rapid non-destructive testing of different brewing materials.