高层建筑的快速识别方法研究

Research on fast identification method of high-rise building

利用卫星图像对高层建筑进行分类和定位,相比市售数据具有成本更节省、更新更及时的优势,在无线网络规划的环境调查工作中具有重要的意义。本文通过不同方法在高层建筑识别上进行对比测试,发现在高层建筑的识别上,使用VGG16网络具有最好的召回率。在算法和预训练网络的挑选上,需要根据数据集的特点进行仔细甄别。Faster R-CNN配以VGG16骨干网络是效果相对较好的一种方法。不同算法架构和预训练网络在不同的数据集上表现相差较大。

Using satellite images to classify and locate high-rise buildings has the advantages of saving and updating more timely.Compared with outsourcing data,it is of great signifi cance in the environmental survey of wireless network planning.In this paper,by comparing different methods in the identifi cation of highrise buildings,it is found that VGG16 network has the best recall rate in the identifi cation of high-rise buildings.In the selection of algorithm and pre training network,we need to carefully screen according to the characteristics of our own data set.Faster-CNN with VGG16 backbone network is a relatively good method.The performance of different algorithmic architectures and pre training networks varies greatly in different data sets.