结合结构相似度的自适应多尺度SAR图像变化检测

发布时间：2018-07-13 16:53

【摘要】：目的结合高斯核函数特有的性质,提出一种基于结构相似度的自适应多尺度SAR图像变化检测算法。方法本文提出的算法包括差异图像获取、高斯多尺度分解、基于结构相似性的最优尺度选择、特征矢量构造以及模糊C均值分类。首先,通过对多时相SAR图像进行对数比运算获取差异图像,然后,利用基于图像的结构相似度估计高斯多尺度变换的最优尺度,继而在该最优尺度参数下逐像素构建变化检测特征矢量,最后通过模糊c均值聚类方法实现变化像素与未变化像素的分离,生成最终的变化检测结果图。结果在两组真实的SAR图像数据上测试本文算法,正确检测率分别达到0.9952和0.9623,Kappa系数分别为0.8200和0.8540,相比传统算法有了较大的提高。结论本文算法充分利用了尺度信息,对噪声的鲁棒性有所提高。实测SAR数据的实验结果表明,本文算法可以智能获取最优分解尺度,显著提高了SAR图像变化检测性能。
[Abstract]:Aim to propose an adaptive multi-scale Gao Si image change detection algorithm based on structural similarity. Methods the proposed algorithms include differential image acquisition, Gao Si multi-scale decomposition, optimal scale selection based on structural similarity, feature vector construction and fuzzy C-means classification. Firstly, the difference image is obtained by logarithmic ratio operation of multitemporal SAR images, and then the optimal scale of Gao Si multiscale transform is estimated by using the structure similarity based on the image. Then the feature vector of change detection is constructed by pixel under the optimal scale parameter. Finally, the separation of changing pixel and unchanged pixel is realized by fuzzy c-means clustering method, and the final change detection result diagram is generated. Results the algorithm is tested on two groups of real SAR image data. The correct detection rate is 0.9952 and 0.9623 Kappa coefficient is 0.8200 and 0.8540, respectively, which is higher than the traditional algorithm. Conclusion the algorithm makes full use of the scale information and improves the robustness of noise. The experimental results of measured SAR data show that the proposed algorithm can intelligently obtain the optimal decomposition scale and improve the performance of SAR image change detection.
【作者单位】：国防科学技术大学电子科学与工程学院;
【基金】：国家自然科学基金项目(61201338)
【分类号】：TN957.52

【参考文献】