基于无人机低空遥感的农作物快速分类方法

无人机以其高时效、高分辨率、低成本、低风险及可重复使用的优势,给遥感技术在各领域的应用提供了新的平台。为了提高无人机遥感中农田信息获取的时效性和精度,该文分析了无人机低空航飞获得的高空间分辨率农作物遥感影像特征,以冬小麦为研究对象,基于农作物波谱特征和NDVI变化阈值,提出了一种农作物快速分类提取方法,并与其他几种常用的遥感分类方法进行比较,探讨了其普适性。结果表明,该方法从无人机高分辨率影像中提取不同种类的农作物分类信息具有较高的正确率和普适性,兼具快速和低成本的特点,在海量农作物无人机航拍数据的信息提取上具有较广的应用。

Rapid crops classification based on UAV low-altitude remote sensing

Abstract: Unmanned Aerial Vehicle (UAV) provides a new platform for the application of remote sensing with its advantages of high efficiency, high spatiotemporal resolution, low cost and risk. This paper designed an experiment to obtain the remote sensing data of winter wheat and corn by the ADC Air vegetation canopy camera carried on UAV platform in Shunyi district of Beijing from April 3, 2011 to November 13, 2011. In order to acquire remote sensing data of high quality, the UAV was arranged to fly every 7~10 days during the whole growing period of winter wheat and corn, and the total flight times amounted to 33. Based on these data the spectral characteristics of winter wheat were analyzed, and the NDVI statistical characteristic value of wheat, light soil and shadow soil was also computed. According to these work, this paper proposed an automatic classification algorithm to classify different crop objects in UAV remote sensing images. Specifically, the reflectance of green band and infrared band was compared to classify three kinds of objects roughly, and then NDVI was calculated for further classification. In this experiment, the NDVI threshold 0.7 was chosen to separate winter wheat from light soil, and 0.4 to separate light soil form shadow soil. As for coin, the NDVI threshold 0.5 was chosen to separate coin from light soil and 0.3 to separate light soil from shadow soil. This automatic classification algorithm attained the accuracy of 96.18% in winter wheat identification, and 90.14% in corn identification, which means the algorithm can get almost same accuracy as maximum likelihood classification, while it needs less time and artificial participation. The classification results show that, compared to other commonly used methods of the remote sensing image classification (maximum likelihood method, SVM method, ISODATA method etc.), this automatic classification method has higher accuracy and universality but lower time cost. This method would have an extensive application prospect in extracting the information of crop from mass data of UAV system.