基于无人机高光谱影像波段选择的薇甘菊分类

薇甘菊是危害最严重的外来入侵物种之一，其生长与传播极其迅速，对我国森林生态系统造成了严重破坏，相关管理部门需要一个有效的薇甘菊监测手段。传统人工调查方式需要投入大量的人力物力，成本高昂、效率低下；近年来快速发展的高光谱遥感技术为薇甘菊的监测提供了新思路。本文以无人机搭载的Nano-Hyperspec高光谱仪获取的广东省增城林场遥感影像数据为基础，对高光谱数据进行几何校正、影像降噪处理、辐射定标及坏带波段剔除等影像预处理；运用最佳指数因子法（OIF）、自适应波段法（ABS）、自动子空间划分（ASP）与自适应波段相结合的波段选择法（ASP+ABS）3种方法进行波段选择，获取信息量较大且波段间相关性较低的特征波段组成薇甘菊分类最佳波段组合，生成3幅遥感影像；最后采用支持向量机方法（SVM）对生成的3幅不同遥感影像进行分类，以分类结果的精度评价3种波段组合对薇甘菊高光谱特征的响应程度，选出更能反映薇甘菊的光谱特征的波段组合。试验结果表明，针对Nano-Hyperspec遥感影像数据，使用OIF波段选择法，研究区内薇甘菊的制图精度和用户精度分别为74.62%、66.52%；使用ABS波段选择法，研究区内薇甘菊的制图精度和用户精度分别为74.37%、67.43%；使用ASP+ABS波段选择法，研究区内薇甘菊的制图精度和用户精度分别达到95.98%、92.98%，分类精度最佳，相较OIF法中薇甘菊的制图精度和用户精度分别提高了21.35%、26.46%，相较ABS法中薇甘菊的制图精度和用户精度分别提高了17.15%、19.3%。可见，本文使用的子空间划分与自适应波段相结合的波段选择方法相较其他两种波段选择方法能更好地反映薇甘菊的光谱特征，可为薇甘菊监测提供有效的技术手段。

Classification study of Mikania micrantha kunth from UAV hyperspectral image band selection

Mikania micrantha kunth is one of the most harmful invasive species, which has caused serious damage to our country’s forest ecosystem for its’ rapidly growing and spread.The relevant management needs an effective method for monitoring Mikania micrantha kunth.The traditional methods of manual investigation requires a lot of manpower and material resources,which are considerably costly and inefficient.In recent years,the rapid development of hyperspectral remotes sensing technology provids a new method of monitoring the Mikania micrantha kunth.This paper bases on the remote sensing image data of Zengcheng forest farm in Guangdong Province,which is obtained by the Nano-Hyperspec hyperspectral imaging instrument carried by UAV. The hyperspectral data is pretreatmented by geometric correction,image denoising,radiometric calibration and bad band elimination. OIF, ABS, ASP+ABS methods for band selection are used, obtaining the characteristic wave band of the most informative and low correlation band to constitute the optimal band combination for Mikania micrantha kunth. Generating three remote sensing images, and using the support vector machine method (SVM) classify the three different remote sensing images.The degree of response of the three band combinations to the hyperspectral characteristics of Mikania micrantha kunth is evaluated with the accuracy of the classification results, and a band combination better reflected the spectral characteristics of Mikania micrantha kunth is selected. The experimental results show the cartographic accuracy and user accuracy are 74.62% and 66.52% by using the method of OIF, 74.37% and 67.43% by using the method of ABS, 95.98% and 92.98% by using the method of ASP+ABS,which has the best classification accuracy. Compared with the method of OIF, the method of ASP+ABS improvs by 21.35% and 26.46%.Compared with the method of ABS, the ASP+ABS improves by 17.15% and 19.3%.Compared with the other two methods, the band selection methods of ASP+ABS in this paper has the better performance of reflecting spectral characteristics of Mikania micrantha kunth, which can provide an effective technical method of monitoring the Mikania micrantha kunth.