基于高光谱图像的玉米种子特征提取与识别

玉米种子的形态特征是玉米品种识别的重要因素之一.采用高光谱成像系统获取9个品种共432粒玉米种子的高光谱反射图像,对图像进行校正和预处理,提取每个样本在563.6~911.4 nm共55个波段范围内的形状特征.分别利用单波段、多波段和全波段下的玉米种子形状特征结合偏最小二乘判别法进行模型分类.结果显示,全波段范围内训练集和测试集的平均正确识别率达到98.31%和93.98%,均优于多波段和单波段的正确识别率.研究表明,该方法能充分利用高光谱图像中可见光和近红外区域的有效特征信息,较准确地鉴别玉米品种,为玉米品种的自动识别领域提供了一种新方法.

Morphological Characteristics of Maize Seed Extraction and Identification Based on the Hyperspectral Image

Morphological characteristic of maize seed is an important factor in identifying maize varieties. Hyperspectral images of 432 maize seeds including nine varieties were acquired using hyperspectral imaging system. The images were corrected and pre-processed, and then shape features of each sample were extracted in the range of 563.6~911.4 nm including 55 wavelengths. The classification models were developed using the shape features of maize seeds from single-wavelength, multi-wavelengths and full wavelengths coupled with partial least squares discriminant analysis (PLSDA), respectively. Simulation results indicate that the average correct identification rate of training set and testing set with full wavelengths is 98.31% and 93.98%, which are better than single-wavelength and multi-wavelengths. Therefore, that is the accurate mean for identifying maize varieties using the feature information of visible and near-infrared region from hyperspectral images.