基于成像高光谱的苹果树叶片病害区域提取方法研究

【目的】对带病斑苹果树叶片的高光谱图像进行病斑提取，为作物病虫害的遥感监测提供支持。【方法】对带有病斑的苹果树叶片成像高光谱图像，从传统基于光谱特征和面向对象特征2个方向入手进行病斑提取。为减少高光谱图像波段之间的冗余，首先对高光谱图像采用PCA变换进行降维处理，利用降维之后的前11个波段，分别采用波谱角分类和面向对象分类的方法提取苹果树叶片病害区域。【结果】由于同物异谱和异物同谱现象的存在，波谱角分类算法在提取病斑时，对叶柄和叶脉产生了错误的分类，而且以像元为分类单位的波谱角分类，在分类结果图中存在椒盐噪声，而面向对象分类则避免了这一现象的发生。【结论】采用面向对象分类方法提取苹果叶片病斑的结果优于基于光谱特征的波谱角分类方法，其总体精度和Kappa系数分别为98.44%和0.97。

Research for extracting method of apple leaf ill spots based on hyperspectral image

【Objective】 The research is to extract apple leaf ill spots through hyperspectral image for the sake of providing support for remote monitoring of crop pests and diseases.【Method】 The research uses both the traditional spectral features and object-oriented features methods to extract leaf spots from hyperspectral image of apple leaf.In order to reduce the redundancy between bands,PCA transform is applied to reduce dimension of hyperspectral image.Then,the first 11 bands after PCA transform are used to extract spots with the two methods.【Result】 As the same object with different spectra and different objects with the same spectrum exist,the stalk and vein are classified into the wrong category using SAM.Furthermore,salt and pepper noise exists in classification results of SAM.However,object-oriented classification avoids the occurrence of this phenomenon.【Conclusion】 The result of extracting spot with object-oriented classification is better than SAM.The overall accuracy with object-oriented classification achieves 98.44% and the Kappa coefficient reaches 0.97.