| 摘要: |
| 【目的】探讨高分辨率遥感影像土地利用信息提取方法的优劣，为研究土地利用/覆盖动态变化提供参考。【方法】以结合纹理特征的支持向量机 (Support vector machine，SVM)分类和多尺度分割的面向对象分类为主要技术，对陕西佛坪长角坝乡遥感影像的土地利用信息进行提取，并将分类结果与基于传统像元的最大似然法分类结果进行比较分析。【结果】面向对象分类法的总精度达到90.67%，较结合纹理特征的SVM法提高了8.34%，而与最大似然分类法相比提高了近20.32%，克服了其他分类方法存在的同谱异物现象及分类结果中地物破碎等缺点，取得了较好的分类结果。【结论】利用面向对象分类法不仅达到了提取土地利用信息的目的，而且精度高、速度快。 |
| 关键词: 遥感影像 土地利用信息 支持向量机 纹理特征 多尺度分割 精度评价 |
| DOI： |
| 分类号: |
| 基金项目:国家林业局“948”项目（2009-4-45） |
|
| Study on extraction methods of land utilization information based on SPOT5 |
|
ZHANG Fafa1, LI Weizhong1, LU Liuye,et al2
|
|
1.College of Forestry ,Northwest A&F University;2.College of Natural Resource and Environment ,Northwest A&F University
|
| Abstract: |
| 【Objective】The study explored the effect of extracting approach for information of land utilization based on high resolution remote sensing image to provide evidence for studying land utilization and cover dynamic variation.【Method】This paper extracted the information of land utilization focused on Changjiaoba town,using SVM classification of texture feature and object-oriented classification of multi-resolution segmentation.The classification result was compared with maximum likelihood classification.Then the classification result was analyzed.【Result】The overall classification accuracy of object-oriented was 90.67%,which increased by 8.34% compared with SVM classification of texture feature and increased by 20.32% compared with maximum likelihood classification.This kind of classification not only can get over the disadvantages of other classifications,e.g.Spectral Similar and Ground-object Fragmentations,etc.but also acquire good effectiveness.【Conclusion】Using the classification of object-oriented can realize the purpose of extracting the land utilization information,and this method is accurate and fast. |
| Key words: remote sensing image information of land uatilization support vector machine texture feature multiresolution segmentation accuracy assessment |
