论文部分内容阅读
对玉米种子高光谱图像的光谱维信息进行分析,探索利用高光谱图像技术鉴定玉米杂交种纯度的可行性。实验中利用高光谱成像系统采集玉米品种农华101的母本和杂交种的高光谱图像,波长范围871～1699nm;在每个玉米样本上提取感兴趣区域的平均光谱信息,利用处理后的数据建立农华101母本和杂交种的鉴定模型。讨论了样品的摆放方式(种子胚正对光源和背对光源,种子在样品台上的位置)和实验环境对鉴定模型性能的影响。鉴定模型对不同摆放方式和实验环境下获得的同种样品的光谱的正确识别率和正确拒识率均达到90%以上,模型稳健性良好。利用Qs方法选择特征波段[1],发现在1 230nm附近(1 195～1 246nm)农华101的母本和杂交种差异最大。实验中利用特征波段内的数据进行建模和测试,正确识别率和正确拒识率达到90%以上,与利用全波段(925～1597nm)获得的识别效果相当。分析结果表明,利用高光谱图像技术鉴定玉米杂交种纯度是可行的。
The spectral information of corn hyperspectral image was analyzed, and the feasibility of using hyperspectral image to identify the purity of corn hybrid was explored. In this experiment, hyperspectral images of maize and hybrids of maize variety Nonghua101 were collected by using hyperspectral imaging system in the wavelength range of 871 ~ 1699nm. The average spectral information of the region of interest was extracted from each maize sample. Using the processed data Establishing identification model of female parent and hybrid of Nonghua101. The effects of sample placement (the seed embryo is opposite to the light source and back to the light source, the position of the seed on the sample stage) and the experimental environment on the characterization of the model are discussed. The correct recognition rate and the correct rejection rate of the identification model for the spectra of the same samples obtained under different display modes and experimental environments are both above 90%, and the robustness of the model is good. Using the Qs method to select the characteristic band [1], it was found that there was the most significant difference in the female parent and hybrid between the Nonghua 101 near 1 230 nm (1 195 ~ 1 246 nm). In the experiment, the data in the characteristic bands were used for modeling and testing. The correct recognition rate and the correct rejection rate reached more than 90%, which is equivalent to the recognition effect obtained by using the full band (925 ~ 1597nm). The results show that it is feasible to use hyperspectral image to identify the purity of maize hybrids.