基于P-N学习的高分遥感影像道路半自动提取方法
发布时间：2018-07-24 14:39
【摘要】：基于模板匹配的道路跟踪是半自动提取道路的主要方法。然而场景中地物干扰和道路宽度的变化降低了模板匹配的稳定性;另外,道路跟踪失败后缺乏重检测机制,使得道路提取过程中人机交互频繁。针对以上问题,提出了一种基于P-N(positive-negative)学习的高分遥感影像道路半自动提取方法。该方法由道路跟踪、检测和学习构成,关键是采用了P-N学习的策略迭代的训练分类器,通过纠正违反结构约束的样本分类结果来提高分类器性能。实验使用了不同场景下的城区高分遥感影像,与经典的模板匹配和在线学习的道路跟踪方法进行了比较。实验结果表明该方法在道路提取的精度和稳定性方面均有提升。
[Abstract]:Road tracking based on template matching is the main method of semi-automatic road extraction. However, the variation of ground objects and road width in the scene reduces the stability of template matching. In addition, the lack of re-detection mechanism after road tracking failure makes the human-computer interaction frequent in the road extraction process. Aiming at the above problems, a semi-automatic road extraction method for high score remote sensing images based on P-N (positive-negative) learning is proposed. The method is composed of road tracking, detection and learning. The key is to use the P-N learning strategy iterative training classifier to improve the performance of the classifier by correcting the classification results of samples that violate the structure constraints. The high score remote sensing images of different scenes are used in the experiment, and compared with the classic template matching and online learning road tracking methods. The experimental results show that this method can improve the accuracy and stability of road extraction.
【作者单位】： 武汉大学测绘遥感信息工程国家重点实验室;重庆市勘测院;武汉大学遥感信息工程学院;
【基金】：国家973计划(2012CB719906) 高分辨率对地观测系统重大专项~~
【分类号】：P237
，
本文编号：2141694
[Abstract]:Road tracking based on template matching is the main method of semi-automatic road extraction. However, the variation of ground objects and road width in the scene reduces the stability of template matching. In addition, the lack of re-detection mechanism after road tracking failure makes the human-computer interaction frequent in the road extraction process. Aiming at the above problems, a semi-automatic road extraction method for high score remote sensing images based on P-N (positive-negative) learning is proposed. The method is composed of road tracking, detection and learning. The key is to use the P-N learning strategy iterative training classifier to improve the performance of the classifier by correcting the classification results of samples that violate the structure constraints. The high score remote sensing images of different scenes are used in the experiment, and compared with the classic template matching and online learning road tracking methods. The experimental results show that this method can improve the accuracy and stability of road extraction.
【作者单位】： 武汉大学测绘遥感信息工程国家重点实验室;重庆市勘测院;武汉大学遥感信息工程学院;
【基金】：国家973计划(2012CB719906) 高分辨率对地观测系统重大专项~~
【分类号】：P237
，
本文编号：2141694
本文链接：https://www.wllwen.com/kejilunwen/dizhicehuilunwen/2141694.html
