摘要
干涉合成孔径雷达(InSAR)技术凭借其地表形变监测精度高、覆盖范围广的优势,已成为地表形变监测的重要手段。然而,形变目标通常存在显著的尺度和形态差异,且在形变密集区边界呈现模糊和不规则特征。现有检测模型在细粒度和边缘信息表征方面存在不足,导致小尺度形变区漏检及密集区边界定位不准确。为解决这一问题,提出了一种融合InSAR与目标检测算法的形变区域快速识别方法。该方法首先利用SBAS-InSAR技术获取形变速率图,然后在YOLOv11n的特征提取与特征融合层中引入多尺度空洞卷积、动态上采样和上下文引导下采样等模块,构建了高精度InSAR形变检测网络,实现了对不同尺度与复杂形态形变区域的高效精准识别。在甘肃白龙江流域的实验中,本次研究基于改进YOLOv11n模型的InSAR形变区域快速检测方法在形变区域检测中取得了较高的召回率(89.5%)、准确率(91.7%)和平均精度(AP_(50)=91.3%)。此外,在公开贵州省InSAR数据集上的泛化实验进一步验证了该方法在不同区域的鲁棒性和稳定性。本文提出的方法在保证计算效率的前提下,实现了广域InSAR形变区域的高精度识别,能够满足广域地表形变监测的实际需求。
Interferometric synthetic aperture radar(InSAR)technology has lately emerged as an important tool for monitoring the surface deformation owing to its high precision and wide spatial coverage.However,the targets of deformation often exhibit significant variations in scale and morphology,and involve blurred and irregular boundaries in densely deformed areas.Currently available models of detection are limited in terms of representing fine-grained and edge-related features,and this causes them to omit small-scale zones of deformation and inaccurately localize the boundary in dense regions.To address these issues,this study proposes a rapid method to identify regions of deformation that integrates InSAR data with an object detection algorithm.Specifically,the proposed method first applies the SBAS-InSAR technique to derive maps of the deformation velocity.It then introduces multi-scale atrous convolutions,dynamic upsampling modules,and context-guided downsampling modules to the feature extraction and fusion layers of the YOLOv11n architecture to construct a high-precision network to detect deformations in InSAR data.This network enables the efficient and accurate recognition of the regions of deformation at varying scales,and in the presence of complex morphologies.In experiments conducted on data from the Bailong River Basin of Gansu Province,China,the proposed method achieved high values of recall(89.5%),precision(91.7%),and mean average precision(AP_(50)=91.3%).Furthermore,the results of generalization experiments on a publicly available InSAR dataset from Guizhou Province validated the robustness and stability of the proposed approach across regions.Overall,the proposed method can detect regions of deformation in InSAR data over a wide area while maintaining a high computational efficiency.It thus satisfies the practical requirements for the large-scale monitoring of surface deformation.
作者
陈怀圆
何毅
张清
杨进昆
金龙
CHEN Huaiyuan;HE Yi;ZHANG Qing;YANG Jinkun;JIN Long(Faculty of Geomatics,Lanzhou Jiaotong University,Lanzhou 730070,China;National-Local Joint Engineering Research Center of Technologies and Applications for National Geographic State Monitoring,Lanzhou 730070,China;Key Laboratory of Science and Technology in Surveying&Mapping,Gansu Province,Lanzhou 730070,China)
出处
《成都理工大学学报(自然科学版)》
北大核心
2025年第6期1151-1166,共16页
Journal of Chengdu University of Technology: Science & Technology Edition
基金
国家自然科学基金项目(42471471)
兰州市青年科技人才创新计划项目(2024-QN-12)
天水市秦州区科技支撑计划项目(2025-SHFZG-8017)。
