摘要
滑坡灾害因其极大的破坏性而引起高度重视,如何快速、高精度地自动检测滑坡体成为主要研究问题。针对滑坡体检测数据不足、精度低、检测滑坡体不完全等问题,本文结合卷积神经网络(CNN)和Transformer的优点,以Transformer为主体,采用DETR网络实现滑坡体的自动检测。首先,对于数据集数据不足的问题,采用离线数据增强的方式实现滑坡体数据增广;然后,采用编码器-解码器结构的DETR网络结构对增广数据集进行多尺度训练和预测;最后,对试验结果进行定量评价。试验结果表明,采用DETR网络对滑坡体检测的平均准确率(AP)达0.997,可准确识别和检测滑坡体。此外,试验结果还验证了数据增强可有效提升DETR网络对滑坡体的检测精度。
Landslide disasters have attracted great attention because of their great destructiveness,and how to quickly and accurately detect landslides has become a major problem.Aiming at the problems of insufficient landslide detection dataset,low accuracy,and incomplete detection of landslide body,this paper combines the advantages of convolutional neural networks(CNN)and Transformer,and adopts the DETR network to realize the automatic detection of landslide body with Transformer as the main body.First of all,in order to solve the problem of insufficient data in the data set,the offline data enhancement method is used to achieve landslide data augmentation;Secondly,the DETR network structure using the encoder-decoder structure performs multi-scale training and prediction of the augmented dataset;Finally,the experimental results are quantitatively evaluated.Experimental results show that the average accuracy(AP)of landslide detection is 0.997,which can accurately identify and detect landslide bodies.In addition,the experimental results also verify that data enhancement can effectively improve the detection accuracy of landslide bodies in the DETR network.
作者
杜宇峰
黄亮
赵子龙
李国柱
DU Yufeng;HUANG Liang;ZHAO Zilong;LI Guozhu(Faculty of Land Resource Engineering,Kunming University of Science and Technology,Kunming 650093,China;Surveying and Mapping Geo-Informatics Technology Research Center on Plateau Mountains of Yunnan Higher Education,Kunming 650093,China;Yunnan Haiju Geographic Information Technology Co.,Ltd.,Kunming 650093,China)
出处
《测绘通报》
CSCD
北大核心
2023年第5期16-20,共5页
Bulletin of Surveying and Mapping
基金
云南省基础研究计划(202201AT070164)
国家自然科学基金(41961039)
云南省基础研究计划(202101AT070102)。
关键词
滑坡
目标检测
卷积神经网络
DETR
注意力机制
landslide
object detection
convolutional neural network
DETR
attention mechanism
