Abstract: The quality of rice is closely related to the variety, so variety identification is of great significance for the implementation of the “high-quality grain project”. Hyperspectral reflectance data of 600 rice particles from 6 varieties with similar appearance were collected, and the spectral data were pre-processed by multiple scattering correction (MSC), second derivative (2ND) and standard normal transformation (SNV). The texture features of the grayscale image corresponding to the characteristic wavelength were extracted by the gray level co-occurrence matrix (GLCM). The full-band, feature band, texture feature and spectrum-texture feature fusion data were used to establish a variety identification model based on support vector machine algorithm (SVM). The results showed that the spectral-texture fusion features had the highest classification accuracy rate, reaching to 94.12%. After the model parameters were optimized  by crow search algorithm (CSA), the accuracy rate was 96.57%. Therefore, the SVM combined with the CSA under the spectrum-texture feature combination could make full use of the spectral and texture information of the hyper-spectral image, and realize the rapid non-destructive identification of rice varieties.

Key words: hyper-spectral imaging technology, variety identification, feature fusion, parameter optimization

摘要： 大米品质与品种密切相关，因此品种鉴别对实施“优质粮食工程”具有重要意义。采集外观相似的6个品种共600粒大米的高光谱反射率数据，经过多元散射校正（MSC）、二阶导数（2ND）和标准正态变换（SNV）对光谱数据进行预处理。利用连续投影算法（SPA）和主成分分析（PCA）对光谱数据降维。以灰度共生矩阵（GLCM）提取特征波长对应灰度图像的纹理特征。应用全波段、特征波段、纹理特征以及光谱-纹理特征融合数据分别建立基于支持向量机算法（SVM）的品种鉴别模型。结果表明，光谱-纹理融合特征的分类准确率最高，达到94.12%。利用乌鸦搜索算法（CSA）对模型参数进行优化后，准确率达96.57%。因此，光谱-纹理特征组合下的支持向量机结合乌鸦搜索算法能充分利用高光谱图像的光谱和纹理信息，实现对大米品种的快速无损鉴别。

关键词: 高光谱成像技术, 品种鉴别, 特征融合, 参数优化