Acta Scientiarum Naturalium Universitatis Pekinensis ›› 2018, Vol. 54 ›› Issue (1): 105-114.DOI: 10.13209/j.0479-8023.2017.073

Previous Articles     Next Articles

A Method for Extraction of Newly-Built Buildings in Road Region Using Morphological Attribute Profiles and One-Class Random Forest

SHI Zhongkui1, LI Peijun1,†, LUO Lun2, YANG Ke2   

  1. 1. Institute of Remote Sensing and Geographical Information System, School of Earth and Space Sciences, Peking University,Beijing 100871
    2. China Transport Telecommunications & Information Center, Beijing 100011
  • Received:2016-10-08 Revised:2016-12-26 Online:2018-01-20 Published:2018-01-20
  • Contact: LI Peijun, E-mail: pjli(at)


史忠奎1, 李培军1,†, 罗伦2, 阳柯2   

  1. 1. 北京大学地球与空间科学学院遥感与地理信息系统研究所, 北京100871
    2. 中国交通通信信息中心, 北京 100011
  • 通讯作者: 李培军, E-mail: pjli(at)
  • 基金资助:


The authors present a method for extraction of newly-built buildings in road-region using morphological attribute profiles and one-class random forest. The morphological attribute profiles are first obtained from bitemporal high-resolution remote sensing images. The morphological attribute profiles obtained and spectral features are then combined to extract newly-built buildings along road-regions using an improved one-class random forest. Bitemporal images of the Daoxiang Lake area in Beijing are used as experimental data to validate the proposed method, by quantitatively comparing with two conventional change detection methods, i.e., direct bitemporal classification and post-classification comparison methods based on support vector machine. The experimental results show that the accuracy of newly-built building extraction from the proposed method (i.e. using combined spectral features and attribute profiles) is significantly higher than that using only the spectral features, with an increase of 15.11% in Kappa. In addition, the Kappa of the proposed method is 1.78% and 25.15% higher than that of the direct bitemporal classification and that of the post-classification comparison. Therefore, the experimental results validate the effectiveness of the proposed method. Advantages of the one-class random forest include capabilities to effectively deal with high-dimensional data and measure the importance of different features used in one-class classification.

Key words: high-resolution remote sensing image, road-region, building change detection, morphological attribute profiles, one-class random forest


提出一种利用形态学属性剖面和单类随机森林分类的道路路域新增建筑物提取方法。用该方法计算路域范围内两时相高分辨率遥感影像的形态学属性剖面, 将得到的形态学属性剖面与光谱特征叠加, 采用改进的单类随机森林分类方法直接提取新增建筑物。以北京市稻香湖地区两时相高分辨率影像作为实验数据, 对比分析该方法与经典两时相直接分类及分类后比较方法的新增建筑物提取精度。结果表明, 综合利用形态学属性剖面和光谱特征提取得到的新增建筑物提取精度比仅使用光谱特征的提取精度显著提高, 其中Kappa系数提高15.11%。此外, 该方法提取结果的Kappa系数比两时相直接分类方法提高1.78%, 比分类后比较方法提高25.15%, 验证了所提方法的有效性。所采用的单类随机森林方法能够有效地处理高维数据, 并可以度量不同特征对分类结果的重要性。

关键词: 高分辨率遥感影像, 道路路域, 建筑物变化检测, 形态学属性剖面, 单类随机森林

CLC Number: