青藏高原数字照片植被覆盖度自动算法与应用

覆盖度是植被评价的重要指标,也是遥感反演的关键参数。估算植被覆盖度的常用方法是目测法,但这种方法受观测人员的主观影响。近年来有研究人员利用冬小麦垂直数字照片的HLS颜色空间的色度特征,设计了自动提取覆盖度算法,具有较高的计算精度。但青藏高原植被颜色丰富多样,下垫面背景色彩差异很大,这种常规覆盖度自动提取算法存在困难。论文分析了青藏高原地表垂直数字照片的图像颜色特征,发现超绿色算法对绿色植被比较敏感,能够增强植被和背景的灰度差异,有效抑制土壤背景干扰。并采用K均值聚类算法,设计了青藏高原植被覆盖度的自动提取软件。通过将覆盖度自动提取结果和人工监督分类进行比较,两者误差在5%以内。此外,通过分析分类后的结果图像,提出了进一步改进的方法。

Useing Digital Photography to Measure Vegetation Coverage in Qinghai-Tibet Plateau

As an important indicator of vegetation,vegetation coverage,which is generally estimated by visual measurement,is always used as an important index in vegetation evaluation and a key parameter for remote sensing inversion.But this method strongly depends on individual variables,and without the reproducibility of the results.That is to say,different observers almost certainly record different measurements with the same quantity.The advent of digital photography and automated image processing promises a revolution in the way vegetation coverage is measured.Recently,an automatically extracting algorithm for vegetation coverage was studied based on color features and some threshold values,and have a high accuracy when it is used to calculate the winter vegetation coverage with vertical digital photographs.However,this automatic method isn＇t suitable for Tibetan Plateau due to its various types of vegetation and background.Though analyzing on the color characters of vertical photographs,we found a named excess green method,which is sensitive to green vegetation and enhances the contrast between plant and soil background.A k-means clustering algorithm was designed to divide photographs into plant and background,and calculate vegetation coverage automatically.Compared with the result from the manually supervised classification method,the root mean square error was less than 5%,but it spend less time than supervised classification and had a higher accuracy than that of visual methods.Moreover,some approaches to improve classification accuracy were discussed by analyzing the error source of automatic classification.