摘要:
针对GF-1卫星影像数据的特点,分别采用归一化差分水体指数( nomalized difference water index,NDWI)阈值法、支持向量机( support vector machine,SVM)和面向对象等方法对鄱阳湖区的GF-1影像进行水体信息提取实验,并根据提取结果分析和比较各种方法的优势与不足。选取2块不同尺度和不同复杂度的代表性区域,以人工解译的水体信息为真值,进行漏提率、误提率和提取精度的统计。结果表明:3种方法在2个区域的提取精度都较高,其中,SVM法的提取精度最高(2个区域的提取精度分别为99.4742%,98.0993%),面向对象法的提取精度次之(99.3164%,97.8779%),NDWI阈值法的提取精度相对最低(99.1456%,97.5900%)。
Abstract:
In this paper, the authors conducted an applicability study of water - based information extraction method according to the data features of GF-1 image. Firstly, water index ( normalized difference water index, NDWI) threshold method, support vector machine ( SVM ) method and object - oriented method were used respectively to conduct water information extraction experiments on the Poyang Lake area based on GF-1 image so as to analyze and compare the advantages and disadvantages of these methods. Secondly, statistic analysis of the rate of leakage and error as well as extraction accuracy was made by selecting two representative areas with different scales and complexities, with the manual interpretation of these two water areas as reference. The results show that the above three kinds of methods all have a high accuracy in both areas, with the extraction accuracy of the simple area ( area 1 ) slightly higher than that of the complex area ( area 2 );A comparative study of these three methods shows that SVM method has the highest accuracy ( 99 . 474 2%, 98 . 099 3%) , followed by the object-oriented method (99. 316 4%, 97. 877 9%), and then by NDWI threshold method(99. 145 6%, 97. 590 0%).
下载: