“基于片层-面向类”的竹林信息提取算法与应用分析

Algorithm and Application Analysis of Film-Based & Class-Oriented for Bamboo Forest Information Extraction

随着竹资源经济效益的不断挖掘,及时、准确地掌握竹林的分布状况与动态信息成为亟须解决的问题,而传统的实地调查并不能很好地满足该需求。3S技术的兴起为掌握竹林分布信息提供了新手段,但受多种因素的制约,长期以来竹林信息提取效果不佳。该文在前期研究基础上,进一步以ALOS高分辨率遥感影像为数据源,利用提出的"基于片层—面向类"算法提取南方山地丘陵区顺昌县的竹林信息,以验证该方法的有效性。除利用传统的基于像元的分类方法外,引入近年来时兴的面向对象法,以作比较;同时,对"非片层区"进行进一步处理,以提高竹林信息提取的精度。用实测点验证表明:基于像元与面向对象的竹林信息提取精度分别为84.42%、86.46%,Kappa系数分别为0.677 3、0.723 4;从目视效果看,基于像元的分类结果存在椒盐状,而面向对象的结果图连续性更佳。两种方法结果虽相差不大,但面向对象方法可能更适用于诸如ALOS这样的高分辨率遥感影像。此外,两种方法共同表明,利用"基于片层-面向类"的信息提取思路,能有效地提取竹林信息,并可作为山地丘陵区域影像解译的重要技术参考。

As the continued excavation for economic benefits of bamboo resources,it has been a problem which is urgent need to solve to know the distribution and dynamic information of bamboo forest timely and accurately.However,the traditional field survey cannot meet this demand.The development of Geomatics technology has provided new approaches to seize the bamboo forest distribution information,but restricted by many factors,the extraction of bamboo information is not effective for quite a long time.On the basis of former studies,the paper took the high-resolution of ALOS image as the data source and focuses on bamboo forest information extraction in Shunchang County,which is hilly area in Southern China,with the algorithm of Film-based Class-oriented which has been put forward by us,in order to test the effectiveness of the method.In addition to using the traditional classification method of pixel-based,it introduced the object-oriented method that has been fashional in recent years,in order to compare them.What＇s more,it processed the non-film area further for higher bamboo forest information extraction accuracies.Verifying with the surveying points showed that the extraction accuracies with pixel-based and object-oriented were 84.42% and 86.46% respectively with Kappa of 0.677 3 and 0.723 4.Four results could be indicated.Firstly,both these two methods showed that the algorithm of Film-based Class-oriented was well suitable for the bamboo forest information extraction,and could supply a significant technical reference to the image interpretation of hilly area.Secondly,from the accuracy,the method of object-oriented was a little better than the pixel-based one.Nevertheless,from the visual effect of view,there were spiced salt shapes in the figure of pixel-based,while that of object-oriented had better continuity.Although the results of these two methods were little different,the object-oriented one may be more suitable for the high-resolution images such as ALOS image.Thirdly,it is hard to interpret and extract the bamboo forest information with the help of spectral features from remote sensed images single,so we have to bring in the texture information to raise the extraction accuracy.In addition to these,the terrain factors including elevation,slope and aspect are needed to regard carefully.We could infer that the bamboo forest information extraction accuracy will be raised more if the terrain combination conditions were taken to account.