基于差异图构造与融合的SAR图像变化检测方法  

SAR image change detection method based on difference image construction and fusion

作　　者：林娇 火久元[1,2,3] LIN Jiao;HUO Jiuyuan(School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China;National Cryosphere Desert Data Center(NCDC),Lanzhou 730000,China;Lanzhou Ruizhiyuan Information Technology Co.LTD,Lanzhou 730070,China)

机构地区：[1]兰州交通大学电子与信息工程学院,兰州730070 [2]国家冰川冻土沙漠科学数据中心,兰州730000 [3]兰州瑞智元信息技术有限责任公司,兰州730070

出　　处：《激光杂志》2024年第4期128-134,共7页Laser Journal

基　　金：国家自然科学基金(No.62262038);甘肃省科技计划-中小企业创新基金(No.21CX6JA150);国家冰川冻土沙漠科学数据中心数据专题;兰州交通大学“百名青年优秀人才培养计划”基金。

摘　　要：针对合成孔径雷达(SAR)图像固有的相干斑点噪声而影响变化检测精度和准确性等问题,提出了一种基于差异图构造与融合的SAR图像变化检测方法。该方法通过L-SRAD混合滤波对SAR图像进行预处理,使用基于边缘预检测的小波融合算法实现对数双曲余弦比值差异图DCLR和邻域比值差异图DNR的融合,结合FCM算法和CWNN卷积神经网络对所得融合差异图进行变化检测。其中FCM算法将融合差异图预分类为三个聚类,选择合适的预分类结果作为训练样本训练CWNN模型,最后使用CWNN模型对预分类结果进行二次分类,得到最终的变化检测图。在Bern数据集上进行了对比实验,实验结果证明该方法具有较强的变化检测能力,变化检测准确率达到99.67%。In view of the inherent coherent speckle noise in Synthetic-aperture radar(SAR)images,which af-fects the accuracy and accuracy of change detection,this paper proposes a change detection method for SAR images based on difference map construction and fusion.This method preprocesses SAR images through L-SRAD hybrid filte-ring,uses wavelet fusion algorithm based on edge pre-detection to achieve the fusion of logarithmic hyperbolic cosine ratio difference map DCLR and neighborhood ratio difference map DNR,and combines FCM algorithm and CWNN Convo-lutional neural network to detect changes in the fusion difference map.The FCM algorithm pre-classifies the fused difference map into three clusters,selects appropriate pre-classification results as training samples to train the CWNN model,and finally uses the CWNN model to perform secondary classification on the pre-classification results to obtain the final change detection map.Comparative experiments were conducted on the Bern dataset,and the experimental results showed that this method has strong change detection ability,with a change detection accuracy of 99.67%.

关 键 词：SAR变化检测 L-SRAD滤波器 对数双曲余弦比 改进的小波融合 卷积小波神经网络 

分 类 号：TN29[电子电信—物理电子学]