基于学习矢量量化神经网络的水稻白穗和正常穗的高光谱识别

Classification of Empty and Healthy Panicles in Rice Plants by Hyperspectral Reflectance Based on Learning Vector Quantization (LVQ) Neural Network

水稻病虫害的发生会导致大量白穗的出现,对白穗和正常穗的区分是采取植保措施和灾害评估的基础。通过研究获取了由水稻二化螟和穗瘟造成的白穗和正常穗的室内光谱,选取红边斜率、红边面积、绿峰幅值和绿峰面积等4个高光谱变量作为输入向量,利用学习矢量量化(LVQ)神经网络对水稻白穗和正常穗进行分类。利用测试样本对网络进行测试,结果显示对白穗和正常稻穗的分类精度高达100%。研究表明,基于LVQ神经网络对水稻白穗和正常穗进行辨别的方法是切实可行的,可以补充和替代肉眼观测。

The incidence of disease and insect stresses in rice brings about numerous empty panicles. Differentiation of empty rice panicles from healthy ones was the basis of plant protection strategies and damage assessment. The hyperspectral reflectance of healthy rice panicles and empty ones caused by Chilo suppressalis （Walker） and rice panicle blast were obtained in the laboratory. Red edge and green peak parameters derived from first-order derivative spectrum were treated as the input vectors of learning vector quantization （LVQ） neural network. Rice panicles were classified into two classes with 30 epochs by LVQ neural network using the training sample （n= 108）. The performance of LVQ neural network was examined with the testing sample （n= 74）, and empty and healthy rice panicles could be successfully differentiated without errors. The result demonstrated that the new method was feasible to differentiate empty rice panicles from healthy ones and supplement or substitute the conventional visual survey.