基于图像熵信息的玉米种子纯度高光谱图像识别

种子纯度是种子质量的一个重要标志,为寻求快速有效的种子纯度识别方法,该文利用高光谱图像技术研究了玉米种子的分类识别问题。首先对17类玉米品种共1632粒种子的高光谱图像提取400～1000nm波长范围内233个波段的熵信息作为分类特征;然后利用偏最小二乘(PLS)投影算法对玉米高光谱图像进行最优波段选择,共获得65个最优波段特征;最后结合偏最小二乘判别分析法(PLSDA)实现了玉米种子的准确识别分类。分类结果表明,在最优波段数仅为全波段27.90%的情况下,其训练精度可以达到99.19%、测试精度为98.90%,可实现多类别样本条件下的玉米种子纯度识别。

Maize seed classification based on image entropy using hyperspectral imaging technology

The seed purity is an important index of the seed quality. Hyperspectral imaging technology was investigated to classify the maize seeds in this study. First, the image entropy of hyperspectral images were extracted as classification features for 17 varieties including 1632 samples, which the spectral region covered 400-1 000 nm and contained 233 wavelengths. Then, sixty-five optimal wavelengths were selected using partial least squares (PLS) projection algorithm. At last, partial least squares discriminant analysis (PLSDA) was used to develop the classification models for the maize seed purity. The results indicated that the training accuracy of 99.19% and the testing accuracy of 98.90% were achieved by the models with the optimal wavelengths (only 27.90% of full wavelengths), which can implement the purity classification of multi-class maize seeds.