土地覆盖变化检测方法比较——以内蒙古草原区为例

随着对地观测技术的不断发展,遥感影像分辨率逐渐提高,促进了基于遥感影像的变化检测从传统像元级的检测向面向对象的检测转变。为了探究面向对象的变化检测方法在土地覆盖变化检测中的有效性和适用性,对面向对象的变化检测方法与常规的变化检测方法进行对比评价。以内蒙古鄂尔多斯和包头地区为试验区,选取2002年及2011年的Landsat TM/ETM+影像为数据源,比较了图像代数运算、图像变换、图像空间结构特征和面向对象的多种变化检测方法,对研究区两期土地覆盖进行了变化检测研究。结果表明:面向对象的变化检测方法在总体精度、kappa系数上都有明显的优越性,总体精度均在87.42%以上,尤其以面向对象的变化矢量分析方法精度最高,达91.56%。此外,主成分差异法也有较好的检测效果,总体精度为87.83%。对总体精度较高的3种方法在不同土地覆盖变化类型中检测效果的研究表明:对于研究区几种主要土地覆盖变化类型,面向对象的变化矢量分析法均有较理想的检测效果,平均精度为85%左右,且始终优于面向对象的光谱向量相似法,以居民地及旱地相关的变化类型最为明显;主成分差异法对不同土地覆盖变化类型检测效果差异很大,对其中4种变化类型的精度甚至达到了93%以上,但对于检测草地与裸地间转化精度很低,甚至只有8.69%;在与工矿用地有关的土地覆盖变化类型中,面向对象的变化矢量分析法的精度明显高于主成分差异法,而在与居民地有关的变化类型中,主成分差异法表现出一定优势。

Comparative analysis of land cover change detection in an Inner Mongolia grassland area

Accurate and timely data on land cover change are not only important for global change study, but also provide significant foundation for decision-making, management and monitoring in resources sustainable application. As the modern remote sensing systems have provided a huge amount of data for land cover change study, remote sensing technology has become the most economical and effective way to acquire land cover change information. With the rapid development of earth observation technology, image resolution has been improved gradually, and remote sensing change detection algorithms have been remarkably developed. Remote sensing change detection methods are changing from traditional pix-level detection to object-oriented detection. In order to explore the validity and applicability of object-oriented change detection methods, we compared and evaluated object-oriented change detection methods and traditional change detection methods using Landsat TM/ETM+ images in the grassland area of Baotou and Ordos in Inner Mongolia. The results showed that object-oriented change detection methods had significant advantages both in overall accuracy and kappa coefficient. The overall accuracies obtained by object-oriented change detection methods were above 87.42%. The object-oriented change vector analysis got the highest accuracy, with overall accuracy of 91.56%. Besides, principal component differencing also had good detection result with overall accuracy of 87.83%. The three methods with the highest overall accuracy were further compared over different land cover change types. The results showed that the object-oriented change vector analysis was better than object-oriented spectral vector similar method, and the difference was most obvious for those change types related to construction land and dry land. The object-oriented change vector analysis provided good detection for all land cover change types in the study area with the average accuracy of 85%. However, there was a big difference in the detection results of principal component differencing between different land cover change types. Its accuracy reached as 93% for four land cover change types. But for detection of the transformation between bare land and grassland, the accuracy was low to 8.69%. While detecting the change types related to industrial and mining sites, object-oriented change vector analysis was more accurate than principal component differencing. But principal component differencing showed its superiority in detecting the change types related to construction land.