水稻种子品种的单粒鉴别对于防止制种时的混杂、掺假现象，保证种子纯度具有重要意义。利用高光谱图像技术研究了水稻种子品种的单粒快速鉴别方法。采集了10 类水稻种子在400~1000 nm 范围内的高光谱反射图像并提取其光谱、纹理和形态特征；结合偏最小二乘判别分析模型比较了不同特征及其组合下的分类精度，并利用多次递进无信息变量消除算法结合偏最小二乘投影分析方法筛选最优波段。结果显示，在仅利用23 个最优波段情况下，融合均值、熵、能量和形态特征所建立的鉴别模型获得了令人满意的识别精度，其训练集、测试集精度分别为99.22%、96%。结果表明，高光谱特征融合可以在少量波段情况下有效地提高水稻种子品种单粒鉴别的精度，基本满足国家标准对种子纯度的检测要求。

Variety discrimination for single rice seed is important to prevent the mixing and adulteration during seed production and to ensure the seed purity. A fast discrimination method for single rice seed by using the hyperspectral imaging technology is investigated. Hyperspectral images of rice seeds from ten varieties are collected over the wavelength region of 400~1000 nm, and the spectral, texture and morphological features of rice seeds are extracted. The discrimination accuracy of different features and their combinations is compared by using the partial least squares discriminant analysis, and the multiple progressive uninformative variable elimination algorithm combined with the partial least squares projection analysis algorithm is used for optimal waveband selection. The results show that the satisfactory discrimination accuracy, which is 99.22% and 96% for the training set and test set respectively, is achieved when mean, entropy and power features for the 23 optimal wavebands and morphological features are integrated. It suggests that multiple hyperspectral feature integration can effectively improve discrimination accuracy for single rice seed at the case of a small amount of wavebands, which basically meets the requirements of the national standards on the seed purity identification.