苗期作物和杂草的光谱分析与识别

田间杂草信息是指导变量喷洒除草剂的依据，利用光谱特征识别杂草的方法在实时性方面具有明显的优势。本文利用傅里叶变换红外(FTIR) 光谱法测量并分析了小麦、小藜和荠菜等几种杂草在700～1 100 nm波长范围内的反射率，再运用SPSS统计软件进行判别分析。先把原始数据进行压缩和标准化处理，然后运用逐步判别分析法寻求特征波长点，最后以选定的特征波长点为变量建立判别模型进行判别分析。统计分析的结果表明: 运用选定的特征波长点建立判别模型识别小麦和杂草的正确识别率达到了97%;在680～750 nm“红边”附近的特征波长点较为显著;在一定范围内，正确识别率随着特征波长点个数的增加而增加。本研究选定特征波长点，选择适当的滤光片，并配合黑白摄像机对小麦和杂草进行了多光谱图像采集和分析。

Spectrum Analysis of Crop and Weeds at Seedling

The infestation information on field weeds is the basis of variable spraying her bicides. It was found that the method using the spectral characteristics of plan t is superior in real-time respect. The Fourier transform infrared spectrum tec hnique was applied to measure the reflectance of wheat and weeds in the range fr om 700 to 1 100 nm. The discrimination analysis was done using the SPSS softwa re. Firstly, the source spectrum data were compressed and normalized. Secondly, the characteristic wavelengths were selected by using stepwise method. Thirdly, the discrimination model was set up to use the selected wavelengths as the varia bles for detecting wheat and weeds. It was shown by the result of discrimination analysis that the correct classification rate of wheat and weeds detection with the selected wavelength points achieved 97%. In addition, the selected waveleng th points were marked in the "red edge" of reflectance within some range, and th e rate of correct classification increased with the increase in the numbers of t he selected wavelength points. According to the selected wavelength poin ts, the proper filters were chosen to perform the multi-spectral images captured an d processed with the machine vision system.