成熟度是影响种子活力的重要因素之一, 是种子质量的重要指标。 种子分级时将成熟种子和未成熟种子区分开来可提高种子批活力, 使种子批活力具有一致性。 采用400~1 000 nm波段范围的高光谱成像技术研究成熟和未成熟玉米种子, 找出二者区分度最高的特征波段图像, 通过图像处理方法进行种子分类。 选用主成分分析（PCA）法分析高光谱图像, 分析差异最明显的PC2主成分图像的各波段权重系数并提取出特征波段（501 nm）。 从70粒成熟度较低玉米种子样本高光谱图像上选取成熟和未成熟两类感兴趣区域, 采用偏最小二乘回归（PLSR）法分析两类感兴趣区域的平均光谱, 选取与成熟度相关的敏感波段（518 nm）。 采用波段比运算并结合KW检验, 分析两类感兴趣区域的平均光谱, 找出差异最大的最优波段比（640 nm/525 nm）。 以864粒玉米种子为研究对象, 提取特征波段对应的单波段图像和最优波段比对应的波段比图像, 采用图像处理技术分析图像并判别。 结果表明: 采用单波段灰度图像分割容易将玉米种子冠部的浅色部分误识别为种子成熟度较低的区域, 识别准确率低； 而采用640 nm/525 nm的波段比图像可以减轻这种不利影响, 平均正确识别率为93.9%。 该方法可以有效识别未成熟的玉米种子, 为进一步开发在线分级装备提供了依据。

The seed maturity, which is one of the important factors that affect seed vigor, is an important quality index. During seed sorting, separating mature seeds from immature seeds can improve the vigor of seed lot and keep vigor consistency. Hyperspectral imaging that covered the range of 400~1 000 nm was used to find out the sensitive bands reflecting corn seed maturity, and corresponding images were employed to classify the immature corn seeds. Principal component analysis (PCA) algorithm was adopted to analyze the hyperspectral image. PC2 of PCA had the greatest difference between immature and mature areas on the seeds, therefore, the weighted coefficients of PC2 was selected to extract sensitive wavebands (501 nm). Regions of interest (ROI) from mature and immature area of 70 immature kernels was selected for mean spectra calculation. Partial least square regression (PLSR) algorithm was employed to analyze the spectra of ROI and extract wavelength related to maturity (518 nm). Band ratio algorithm and Kruskal-Wallis test were used to select the best band ratio that had the biggest difference between mature and immature areas (640 nm/525 nm). 864 kernels of corn seed were analyzed by gray images of the selected wavelengths as well as band ratio images. Results showed that the light color regions of the seed crown were misidentified as immature region when the images of selected single band wavelengths were used, while the band ratio image of 640 nm/525 nm could be identified correctly. The immature seeds can be separated from the mature seeds according to the area ratio of segmented immature region to the whole kernel. The correct recognition rate was 93.9%. Using the grey images of selected band ratio can differentiate immature corn seeds from mature seeds effectively, which provide a theoretical reference for the development of seed sorting device in further work.