面向对象的多尺度山地城市土地覆盖信息提取技术研究——以重庆永川区胜利路街道为例

Survey on land cover information extracting of mountain city based on object-oriented and multi-scale technology: taking Shenglilu Community of Yongchuan District for an example

基于eCognition平台,利用面向对象的多尺度分割技术和基于特征的分类方法,对空间分辨率为2.46 m的CBERS 02B HR影像进行了土地覆盖信息提取。采用多尺度分割技术,根据RMAS法计算出最佳分割尺度,基于分割后的对象,首先创建了知识库,然后利用邻近分类的方法进行了土地覆盖信息的分类,提取了影像中的植被、水体、房屋、道路等信息,并利用最佳分类结果精度与分类稳定性两种方法进行了成果精度评价。结果表明:面向对象的多尺度的土地覆盖提取方法除能较好地利用影像的光谱信息外,更能充分利用影像的纹理、形状、大小等空间信息,能较高精度地提取山地城市复杂的土地覆被信息,其中适宜植被的分割尺度为25,房屋的最佳分割尺度为20,水体与道路最佳分割尺度为15,分类精度平均达95.63%。该方法具有较强的可行性及推广性。

To extract land cover information from CBERS 02B HR image whose spatial resolution is 2. 46 meters,the authors used object-oriented and multi-scale segmentation technology based on eCognition. Using multi-scale segmentation technique,they calculated the optimal segmentation scale on account of RMAS method. According to the results of segmentation,they created a knowledge base firstly,and then used the nearest neighborhood classification to extract land cover information,such as vegetation,water,houses,roads and other information. Finally,they used accuracy and stability to evaluate the classification results. It shows that this method can not only make better use of the spectrum of image,but also the texture,shape,size and other spatial information can be used effectively,especially in that of complex mountain city. In conclusion,the best segmentation scales for vegetation,house,water and roads are respectively 25,20 and 15,and the average classification accuracy is up to 95. 63%.