期刊文献+

基于Mask-RCNN的建筑物目标检测算法 被引量:30

Building target detection algorithm based on Mask-RCNN
原文传递
导出
摘要 针对在航空影像中,城区80%的人工目标物为建筑物和道路,建筑物是遥感影像中主要地物的类别,所以建筑物的检测会直接影响到地物提取的自动化水平这一问题。该文提出了一种基于Mask-RCNN的建筑物目标检测方法,是基于卷积神经网络思想,在深度学习框架下通过多线程迭代训练,将无人机影像作为训练样本,在卷积神经网络中得到目标特征再通过区域建议网络(RPN)与ROIAlign操作将特征输入不同的全连接分支。最后得到具优化的权重参数的目标检测模型。在不同场景图像中,该模型可以检测出建筑物目标。实验结果达到了预期要求,提高了航空影像中建筑物检测的准确性。 Aiming at the problem that in aerial imagery,80%of the artificial targets in urban areas are buildings and roads,and buildings are the main types of features in remote sensing images,so the detection of buildings will directly affect the automation level of feature extraction.This paper proposed a building object detection method based on Mask-RCNN?It was based on the idea of convolutional neural network.Through the multi-thread iterative training in the deep learning framework,the unmanned aerial vehicle(UAV)image was used as the training sample.The target features were obtained in the neural network and then the features were input into different fully connected branches through the region proposal network(RPN)network and the ROIAlign operation.Finally,the target detection model with optimized weight parameters was obtained.In different scene images,the model could detect building targets.The experimental results met the expected requirements and improved the accuracy of building detection in aerial imagery.
作者 李大军 何维龙 郭丙轩 李茂森 陈敏强 LI Dajun;HE Weilong;GUO Bingxuan;LI Maosen;CHEN Minqiang(Faculty of Geomatics,East China University of Technology,Nanchang 330013,China;State Key Laboratory of Information Engineering in Surveying,Mapping and Remote Sensing,Wuhan University,Wuhan 430079,China;Gansu Forestry Polytechnic,Tianshui,Gansu 741020,China)
出处 《测绘科学》 CSCD 北大核心 2019年第10期172-180,共9页 Science of Surveying and Mapping
基金 国家重点研发计划项目(2016YFB0502200) 国家自然科学基金项目(41127901) 测绘遥感信息工程国家重点实验室专项科研经费资助项目
关键词 建筑物目标检测 卷积神经网络 Mask-RCNN ResNet101网络 TensorFlow building target detection convolutional neural network Mask-RCNN ResNet101 network TensorFlow
  • 相关文献

参考文献5

二级参考文献27

共引文献356

同被引文献282

引证文献30

二级引证文献112

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部