基于深度学习的遥感影像断层自动检测方法  

Automatic Fault Detection in Remote Sensing Image Using Deep Learning

作　　者：杨易鑫 周子勇[1] YANG Yixin;ZHOU Ziyong(Geosciences College,China University of Petroleum-Beijing,Beijing 102249,China)

机构地区：[1]中国石油大学(北京)地球科学学院,北京102249

出　　处：《遥感信息》2023年第2期33-39,共7页Remote Sensing Information

摘　　要：在遥感影像断层识别中,传统的基于目视判读的方法准确度较高,需要解译人员具备一定的专业基础,且解译过程比较繁琐。此外,在遥感影像断层自动检测方面,基于传统的图像处理方法主要根据影像纹理等浅层特征进行分类判别,受环境影响较大。针对这些问题,文章提出了一种基于深度学习的遥感影像断层自动检测方法,并设计了基于UNet网络的FUnet(fault detection UNet)网络。利用新疆某地区的10级和11级影像,结合1∶50万地质图,制作了断层样本14012张,最后基于模型训练和测试,通过目视对比和定量评价,分析得出10级影像断层识别准确度好于11级影像,而且不同区域识别的准确度也不一样。文章结果初步证实了深度学习方法用于遥感影像断层自动检测的可行性。For remote sensing image fault recognition,traditional visual interpretation can recognize faults with high accuracy,but this is a very professional work and the interpretation process is time consuming.On the other hand,the currently commonly used methods for automatic detection of faults utilize only some shallow features of remote sensing images,such as image textures,which are greatly affected by the environmental factors.In view of these difficulties,this paper proposes an automatic detection method of remote sensing image faults based on deep learning,and designs a FUnet(fault detection UNet)network based on UNet network.In this paper,14012 fault samples are produced using the 10th and 11th grade images in a certain area of Xinjiang.Finally,the visual comparison and quantitative evaluation based on confusion matrix are adopted to analyze the effectiveness of the model based on the testing set.The results show that the accuracy of the 10th level image is higher than that of the 11th level image,and the accuracy of different regions is different.The results of the work preliminarily validate the feasibility of deep learning method for remote sensing image fault detection.

关 键 词：遥感影像处理 深度学习 语义分割 断层检测 UNet 

分 类 号：P237[天文地球—摄影测量与遥感]