Land use/cover classification of a complex agricultural landscape using single-dated very high spatial resolution satellite-sensed imagery
Land use/cover classification of a complex agricultural landscape using single-dated very high spatial resolution satellite-sensed imagery
复制标题
DOI：
10.5589/m11-010
复制
发表时间：
2010-12
影响因子：
2.6
通讯作者：
S. Okubo;Parikesit;Dendi Muhamad;K. Harashina;K. Takeuchi;M. Umezaki
中科院分区：
文献类型：
--
作者：
S. Okubo;Parikesit;Dendi Muhamad;K. Harashina;K. Takeuchi;M. Umezaki
监测潮湿热带农业景观的土地使用/覆盖变化对实现可持续农村发展至关重要。然而，镶嵌景观的时空复杂性使得很难获得准确的土地利用/覆盖图使用单日期和中等分辨率的遥感图像。在这项研究中，我们试图分类土地利用/覆盖利用纹理措施，以提高面向对象的分类的基础上单日期的QuickBird图像。该程序包括两个步骤：（1）使用全色波段，基于灰度共生矩阵（GLCM），评估要分割的图像对象的空间大小，该空间大小足以获得通过Haralick的纹理测量计算的土地利用/覆盖中的可辨别的纹理特征，以及（2）使用纹理和光谱信息，开发具有分类和回归树模型（CART）的分级分类规则集。当在3 × 3到31 × 31像素的窗口大小中计算土地覆盖类型之间的可分离性时，最好的区分来自最大的窗口。GLCM纹理措施，特别是熵，提高了分类精度划定稻田水稻生长阶段无关。分类规则集来自CART建模是直观的理解：整个图像分为绿色和非绿色纹理均匀或异质的土地利用/覆盖类，这似乎描述了农业景观的各种土地利用/覆盖的特征的基本性质。
Monitoring land use/cover changes in humid tropical agricultural landscapes is crucial to establishing sustainable rural developments. However, the characteristic spatiotemporal complexity of mosaic landscapes makes it difficult to obtain accurate land use/cover maps using single-dated and moderate-resolution remotely sensed images. In this study, we attempted to classify land use/cover by utilizing texture measures to improve object-oriented classification based on a single-dated QuickBird image. The procedure consists of two steps: (1) assessing the spatial size of image objects to be segmented that is adequate for obtaining discriminable textural features among land uses/covers calculated by Haralick's texture measures based on a grey-level co-occurrence matrix (GLCM) using the panchromatic band, and (2) developing a hierarchical classification rule set with a classification and regression tree model (CART) using the textural and spectral information. The best discrimination was derived from the largest windows when separability among land-cover types was calculated in window sizes from 3 × 3 to 31 × 31 pixels. GLCM texture measures, especially entropy, improved classification accuracy in delineating paddy fields irrespective of the stage of rice growth. The classification rule set derived from the CART modelling was intuitively understandable: a whole image was divided into green and nongreen texturally homogeneous or heterogeneous land use/cover classes, which seems to describe the fundamental nature of the characteristics of various land uses/covers of agricultural landscapes.