水稻是人类的主要粮食作物，其发芽率是评定水稻质量的重要指标之一。以南粳46 为研究对象，利用高光谱成像技术预测剥壳后的稻种(以下简称糙米)发芽率。在400~1000 nm 波长范围内，采集960 粒饱满、无霉变糙米的高光谱图像，提取感兴趣区域的平均光谱曲线，利用主成分分析(PCA)提取特征波长，再结合偏最小二乘法(PLS)、反向传播神经网络(BPNN)、径向基神经网络(RBFNN)和广义回归网络(GRNN)4 种建模方法分别对糙米5 个区域特征波长的光谱数据建立预测模型并加以比较。4 种建模方法对糙米A 区域(含胚芽)的平均预测效果最好( R_p=0.970)，其中，GRNN 模型对该区域预测精度最高( R_p =0.982, f_RMSEP =0.978)。研究结果表明利用高光谱成像技术并结合PCA和GRNN 检测糙米发芽率是可行的。

Rice is the main food crop for human beings, whose germination rate is one of the most important indexes to evaluate rice quality. The germination rate of brown rice named Nan Jing 46 is predicted by using hyperspectral imaging system. Hyperspectral images of 960 samples which are full and not moldy are captured, and the spectral region is from 400 nm to 1000 nm. The mean spectra are extracted from the region of interest of each image and principal component analysis (PCA) is applied to select characteristic wavelengths from the full- spectrum. The prediction models are established based on spectrum data of characteristic wavelengths of different rice parts using 4 prediction methods, including partial least squares (PLS), radial basis function neural network (RBFNN), general regression neural network (GRNN) and back-propagating neural network (BPNN). After repeated tests, the top area of brown rice (containing the germ) is chosen as the characteristic part, which has the best prediction performance ( R_p =0.970). The order of the prediction accuracy from low to high is PLS, BPNN, RBGNN, GRNN. Among these methods, GRNN has the highest prediction accuracy ( R_p=0.982, f_RMSEP =0.978). The results indicate that it is feasible to detect the germination rate of brown rice by the hyperspectral imaging system.