基于可见-近红外光谱分析的圆白菜与杂草识别研究  

Research on Discrimination of Cabbage and Weeds Based on Visible and Near-Infrared Spectrum Analysis

作　　者：祖琴[1,2,3] 赵春江[1,3] 邓巍[1,3] 王秀[1,3] 

机构地区：[1]北京农业信息技术研究中心,北京100097 [2]贵州大学电气工程学院,贵州贵阳550025 [3]北京农业智能装备技术研究中心,北京100097

出　　处：《光谱学与光谱分析》2013年第5期1202-1205,共4页Spectroscopy and Spectral Analysis

基　　金：国家(863计划)项目(2012AA101904);"十二五"科技支撑计划项目(2011BAD20B07);国家(948计划)项目(2011-G32)资助

摘　　要：杂草的自动识别是实现作物草害精准施药的基础。利用ASD光谱仪采集两个品种的圆白菜及稗草、狗尾草、马唐、牛筋草和小藜等五种杂草在350～2 500nm波段内的冠层光谱反射率。根据光谱曲线特征,在不同波段内对数据进行不同程度的压缩,以提高运算效率;利用不同参数设置的Savitzky-Golay(SG)卷积平滑求导和多元散射校正方法(MSC)的不同顺序组合对光谱去噪,然后结合主成分分析法(PCA)提取主成分,建立模型,最后利用簇类的独立软模式(SIMCA)分类法对各种植物进行分类,并比较分类结果。试验结果显示利用MSC与3阶5次21点SG相结合的方法对光谱数据预处理后,运用PCA提取前10个主成分作为分类模型的输入变量,取得了100%的分类正确率,能够快速无损地识别圆白菜与几种常见杂草。The automatic identification of weeds forms the basis for precision spraying of crops infest. The canopy spectral re- flectance within the 350-2 500 nm band of two strains of cabbages and five kinds of weeds such as barnyard grass, setaria, crabgrass, goosegrass and pigweed was acquired by ASD spectrometer. According to the spectral curve characteristics, the data in different bands were compressed with different levels to improve the operation efficiency. Firstly, the spectrum was denoised in accordance with the different order of multiple scattering correction （MS（2） method and Savitzky-Golay（SG）convolution smoothing method set by different parameters, then the model was built by combining the principal component analysis （PCA） method to extract principal components, finally all kinds of plants were classified by using the soft independent modeling of class analogy （SIMCA） taxonomy and the classification results were compared. The tests results indicate that after the pretreatment of the spectral data with the method of the combination of MS（2 and SG set with 3rd order, 5th degree polynomial, 21 smoothing points, and the top 10 principal components extraction using PCA as a classification model input variable, 100% correct classifi- cation rate was achieved, and it is able to identify cabbage and several kinds of common weeds quickly and nondestructively.

关 键 词：杂草识别 可见-近红外 主成分分析 多元散射校正 

分 类 号：O657.3[理学—分析化学]