摘要
松材线虫病在秦岭以南传播广、蔓延快,利用遥感技术提取病害松可为疫情防控提供数据支撑。以无人机获取的高分辨率、高光谱数据为基础,捕捉区分病害松与健康松的特征波段,提出一种结合植被指数与小波神经网络的林地分层分类新方法,用于提取病害松树冠图斑。病害松识别结果与实地核查数据对比表明,分层分类结果与实地核查情况基本一致,病害松识别总精度达到89.87%,优于单一使用支持向量机、小波神经网络的分类结果。林地分层分类新方法与传统分类方法相比,分类精度更高,所需的样本更少。无人机高光谱数据信息丰富,病害松、健康松在可见光、近红外范围内均有光谱敏感特征,后续应进行病害松时序光谱监测,进一步研究松材线虫病害影响因子。
Since pine wood nematode disease is widespread and fast spreading in the south of Qinling Mountains,remote sensing technology can provide data support for the relevant epidemic prevention and control.Based on the high-resolution hyperspectral data obtained by unmanned aerial vehicle(UAV),the characteristic bands that distinguished diseased type from healthy mode was captured in this paper,then combined the vegetation index and wavelet neural network to extract the crown pattern spots.The comparison between the identification results of disease pine and the field verification data showed that,the hierarchical classification results were basically consistent with the field verification,and the total accuracy of disease pine identification reached 89.87%.It is superior to the classification results using only support vector machine and wavelet neural network.Compared with the traditional classification method,the new hierarchical classification method of forest land had higher classification accuracy with fewer samples.The UAV hyperspectral data is rich in information.The diseased pine and healthy pine have hyperspectral sensitive characteristics in the visible light and near infrared range.The sequential spectral monitoring of diseased pine should be carried out to further study the impact factors of pine wood nematode disease.
作者
张雅洁
周晓敏
张德成
ZHANG Yajie;ZHOU Xiaomin;ZHANG Decheng(The First Geodetic Surveying Brigade of the Ministry of Natural Resources,Xi’an Shaanxi China)
出处
《北京测绘》
2023年第10期1338-1341,共4页
Beijing Surveying and Mapping
基金
陕西测绘地理信息局科技创新项目(SCK2021-06)。
关键词
无人机
高光谱
小波神经网络
分层分类
松材线虫病
unmanned aerial vehicle(UAV)
hyperspectral
wavelet neural network(WNN)
hierarchical classification
pine wood nematode disease
