实现快速、 精确地鉴别玉米单倍体籽粒对玉米单倍体育种技术十分重要。 近红外光谱分析技术可在线分析、 监测, 且无损、 分析速度快、 操作简便、 测试成本低, 对实现自动化的大规模鉴定并分拣玉米单倍体非常有帮助。 通过美国JDSU的近红外光谱仪进行玉米近红外光谱的数据采集, 交叉采集玉米单倍体、 多倍体数据。 数据处理时, 将数据分为训练集和测试集两部分。 依次对数据做预处理以消除噪声影响, 做核变换将其投射到更高维度空间中增强可分性并进行特征提取, 最后建立分类模型鉴别分析。 分别统计采用不同的特征提取算法并建立模型鉴别测试的正确识别率。 实验结果表明, 采用核局部保持投影(KLPP)的特征提取算法的正确识别率更高、 稳定性更好, 在两组测试集上的正确识别率的均值分别达到95.71%和96.43%。 通过分析可以得出, 玉米种子的近红外光谱数据经过非线性变换(为高斯核变换)投影到更高维度的空间后, 表现出更易于分类的分布特点, 保持数据的局部特性也更利于后续的分类。 这为玉米单倍体鉴定进一步研究提供了新的方向。

Haploid identification plays a key role in the field of maize-haploid breeding. To achieve mass and automated identification, Near-infrared Spectroscopy (NIRS) Analysis Technology is widely used. Its advantages include online monitoring, rapid analysis, easy operation, lossless process, cost-effectiveness, etc. At the beginning of the experiment, NIRS data of haploid and polyploidy maize seeds are cross collected via JDSU’s near-infrared spectrometer. To enhance validity, this experiment encompasses a testing set of data besides a training set. After pre-processing, experiment data is subsequently mapped in a higher-dimensional space to enhance its divisibility, and haploid feature is extracted. Then the experiment establishes identification models to predict whether maize seeds are haploid. It needs to point out that the experiment applies different feature extraction algorithms, thus different identification models are established accordingly. The experiment results show that the feature extraction algorithm of Kernel Locality Preserving Projection (KLPP) guarantees accurate recognition in a more stable way. Recognition rate of testing set and training set reaches up to 95.71% and 96.43%. The above experiment proves that NIRS data of maize seeds can be classified more effectively and accurately through non-linear transformation (Gaussian kernel transform in this experiment) and high-dimensional spatial mapping. The above process also maintains partial characteristics of NIRS data. Therefore, this paper may provide some new idea and method for Maize Haploid Identification technology.