论文部分内容阅读
胡杨、柽柳是干旱荒漠区生境的指示种,其树冠提取是荒漠生境遥感定量监测的基础。以塔里木河下游胡杨、柽柳为研究对象,基于QuickBird数据,使用光谱单数据源SVM、光谱结合纹理SVM、面向对象分类和最大似然分类法提取树冠。结果表明:1光谱结合纹理SVM比光谱单源SVM分类精度高9.65%,冠幅估测精度高7.18%,表明高分辨影像上纹理是提高分类精度的重要因素;2面向对象分类法精度最高,分类总体精度86.47%,较光谱单源SVM提高15.67%,较光谱结合纹理SVM提高6.02%,较最大似然法提高22.58%,其冠幅估测精度达87.45%。它兼顾面向对象影像分割与支持向量机方法优点,有效利用分割对象光谱、纹理和空间等信息,较好地解决了其他方法“同物异谱、异物同谱”造成提取树冠破碎的问题,使树冠提取具有较好的稳定性和较高精度。
Populus euphratica and Tamarisk are the indicator species of habitats in arid desert regions. The extraction of canopy is the basis of remote sensing quantitative monitoring of desert habitat. The Populus tomentosa and Tamarix chinensis in the lower reaches of the Tarim River were selected as the research object. Based on the QuickBird data, the canopy was extracted using spectral single data source SVM, spectral combination texture SVM, object-oriented classification and maximum likelihood classification. The results show that: (1) Spectral texture-combined SVM is 9.65% higher than single-source SVM classification and the estimation accuracy of crown is 7.18% higher, indicating that high-resolution texture is an important factor to improve the classification accuracy.2 Object-oriented classification has the highest accuracy, The overall accuracy of the classification was 86.47%, which was 15.67% higher than that of the single-source SVM and 6.02% higher than the SVM of the spectral combination, which was 22.58% higher than the maximum likelihood method. The accuracy of the crown estimation was 87.45%. It takes into account the advantages of object-oriented image segmentation and support vector machine method, and effectively uses the information of the spectrum, texture and space of the segmentation object to solve the problem that the extraction of the crown can be broken by other methods , So that the crown extraction has good stability and high accuracy.