光谱学与光谱分析, 2010, 30 (2): 426, 网络出版: 2010-07-23
园艺作物病害的多光谱组合分类
Horticultural Plant Diseases Multispectral Classification Using Combined Classified Methods
多光谱成像技术 光谱分类 分类器 园艺作物病害 Multispectral imaging technique Spectra classification Classifier Horticultural plant diseases
摘要
选取设施园艺作物黄瓜的主要病害为研究对象, 利用窄带滤光片型多光谱成像系统, 获取患病黄瓜叶面的14个可见光通道和近红外通道、 全色通道的多光谱图像。 采用多光谱图像分类技术中的距离分类器、 相关系数分类器和BP人工神经网络分类器, 将不同病害类型对应的16个波段的反射率看作线性波谱, 对210个多光谱数据样本进行识别分类, 目的是探讨一个能有效识别黄瓜植株常见病害的多光谱组合分类器。 实验结果表明, 将人工神经网络和距离分类器有效组合, 不仅分类性能明显优于单个分类器的性能, 而且能够充分发挥各个分类器的特性, 对园艺作物病害进行快速、 准确、 实时的无损检测。
Abstract
The research on multispectral data disposal is getting more and more attention with the development of multispectral technique, capturing data ability and application of multispectral technique in agriculture practice. In the present paper, a cultivated plant cucumber’ familiar disease (Trichothecium roseum, Sphaerotheca fuliginea, Cladosporium cucumerinum, Corynespora cassiicola, Pseudoperonospora cubensis) is the research objects. The cucumber leaves multispectral images of 14 visible light channels, near infrared channel and panchromatic channel were captured using narrow-band multispectral imaging system under standard observation and illumination environment, and 210 multispectral data samples which are the 16 bands spectral reflectance of different cucumber disease were obtained. The 210 samples were classified by distance, relativity and BP neural network to discuss effective combination of classified methods for making a diagnosis. The result shows that the classified effective combination of distance and BP neural network classified methods has superior performance than each method, and the advantage of each method is fully used. And the flow of recognizing horticultural plant diseases using combined classified methods is presented.
冯洁, 李宏宁, 杨卫平, 侯德东, 廖宁放. 园艺作物病害的多光谱组合分类[J]. 光谱学与光谱分析, 2010, 30(2): 426. FENG Jie, LI Hong-ning, YANG Wei-ping, HOU De-dong, LIAO Ning-fang. Horticultural Plant Diseases Multispectral Classification Using Combined Classified Methods[J]. Spectroscopy and Spectral Analysis, 2010, 30(2): 426.