一种高分辨率遥感影像建筑物自动检测方法  

An Automatic Building Detection Method from High Resolution Remote Sensing Images

作　　者：张通 潘励[1] ZHANG Tong;PAN Li(School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,China)

机构地区：[1]武汉大学遥感信息工程学院,湖北武汉430079

出　　处：《测绘地理信息》2020年第2期101-105,共5页Journal of Geomatics

基　　金：国家自然科学基金(41322010)。

摘　　要：高分辨率遥感影像具有丰富的空间信息、地物几何结构和纹理信息,有助于对地物目标进行认知和解译。而建筑物目标在人类活动区域内占据重要地位,对高分辨率遥感影像中的建筑物进行自动检测具有重大意义。提出了一种基于全卷积神经网络的建筑物自动检测方法,并制作了建筑物样本数据集,利用基于区域的全卷积神经网络和特征检测网络进行建筑物检测模型的参数训练,对待检测影像进行预处理之后利用模型进行建筑物检测,得到影像中的建筑物目标的具体位置和类别置信度。实验证明,提出的检测方法具有更好的效果和更快的速度。检测召回率达到92%,检测准确率达到98%,证明了该方法针对建筑物检测具有较高的精度和较强的稳定性。High-resolution remote sensing image has rich spatial information,geometric structure and texture information,which can help to understand and interpret the target of the images.The goal of building occupies an important position in the human activity area,and the automatic detection of buildings in high-resolution remote sensing images is of great significance.We propose an automatic building detection method based on the fully convolutional neural network.We use the fully convolutional neural network based on regional(region-based fully convolutional network,R-FCN)and feature detection network(residual network,ResNet)for building samples parameter training.After doing some pre-processing to the test images,we utilize the training template to obtain the location and confidence of the building target.Experimental results showed that the proposed method has excellent performance.The detection recall rate reaches 92% while the detection accuracy reaches 98%,which proves that the model has high precision and strong stability in the field of building detection.

关 键 词：高分辨率遥感影像 建筑物检测 全卷积神经网络 深度学习 

分 类 号：P237[天文地球—摄影测量与遥感]