融合主题语义特征的高分辨率遥感影像建筑物自动化识别

Exploration and Practice on Building Recognition Method of High Resolution Remote Sensing Image Based on Topic Model

建筑物提取是高分辨率遥感影像解译领域的一项重要工作,由于遥感技术的宏观性、实时性和周期性特点,基于遥感影像的建筑物识别对于城市管理与国土规划等领域具有非常大的应用潜力。本文介绍了一种基于影像主题语义特征建模的高分辨率遥感影像建筑物提取方法并通过具体的影像场景加以实践应用,该方法首先对影像进行对象级的分割,然后利用LDA模型对分割影像提取语义特征,最后使用SVM分类器综合光谱-空间-语义特征开展建筑物提取。实验表明,该方法能实现较高的建筑物提取精度。

Building extraction is an important task in the field of high-resolution remote sensing image interpretation. Due to the macroscopic, real-time and periodic characteristics of remote sensing technology, building identification based on remote sensing image has great application potential for urban management and land planning. This paper introduces a kind of building extraction method of high resolution remote sensing image based on semantic feature modeling and applies it to the specific image scene. This method firstly carries out object-level segmentation of images, then extracts semantic features from the segmented images using LDA model, and finally take advantages of SVM classifier to carry out building extraction based on comprehensive spectral-spatial-semantic features. Experiments show that this method can achieve high extraction accuracy of buildings.