紫花苜蓿是我国北方重要的栽培牧草, 由于其种子硬实率高达20%～80%, 在播种时无法快速辨别硬实种子而影响其种用价值, 对实际生产造成了直接影响。 为研究出快速无损分选苜蓿硬实种子和非硬实种子的方法, 选取不同产地、 不同收获年份、 不同品种的紫花苜蓿种子样品121份, 同时经人工配比获得不同硬实比率的紫花苜蓿种子样品共31份, 利用近红外光谱技术, 采用偏最小二乘法, 对其硬实率进行了分析, 建立紫花苜蓿种子硬实率的近红外光谱分析模型, 为紫花苜蓿硬实种子的快速检测和识别技术进行探索性研究。 结果表明, 最终获得的光谱模型建模相关系数（R2cal）为0.981 6, 交叉检验标准误差（RMSECV）为5.32, 相对标准误差（RPD）为3.58, 模型的预测能力相对较好, 能够应用于紫花苜蓿种子硬实率的估测, 为种子硬实率的无损检测提供了一种新方法。 模型的建立为研究出快速无损确定种子硬实特性方法提供技术指导, 促进种子硬实特性的深入研究。

Alfalfa (Medicago sativa L.) is the most commonly grown forage crop due to its better quality characteristics and high adaptability in China. However, there was 20%~80% hard seeds in alfalfa which could not be identified easily from non hard seeds which would cause the loss of seed utilization value and plant production. This experiment was designed for 121 samples of alfalfa. Seeds were collected according to different regions, harvested year and varieties. 31 samples were artificial matched as hard rates ranging from 20% to 80% to establish a model for hard seed rate by near infrared spectroscopy (NIRS) with Partial Least Square (PLS). The objective of this study was to establish a model and to estimate the efficiency of NIRS for determining hard rate of alfalfa seeds. The results showed that the correlation coefficient (R2cal) of calibration model was 0.981 6, root mean square error of cross validation (RMSECV) was 5.32, and the ratio of prediction to deviation (RPD) was 3.58. The forecast model in this experiment presented the satisfied precision. The proposed method using NIRS technology is feasible for identification and classification of hard seed in alfalfa. A new method, as nondestructive testing of hard seed rate, was provided to theoretical basis for fast nondestructive detection of hard seed rates in alfalfa.