摘要
正则化图像特征增强算法能够在保持图像较高分辨力的情况下抑制散斑和旁瓣。选择合适的正则化参数对于正则化算法所形成图像的质量至关重要。本文应用推广到非二次型正则化情形的广义交叉验证(GCV)、稳健广义交叉验证(RGCV)和Stein无偏风险估计(SURE)方法研究正则参数选择策略,推导了特征增强正则化方法中GCV,RGCV和SURE函数的直接计算公式,并提出了修正特征项后快速求取正则解的算法以及一般的不动点迭代算法,从而实现了正则参数的自适应选择。数值仿真和实测数据处理结果均说明所提方法的有效性。
The feature-enhanced regularization-based radar image formation technique can effectively obtain high resolution image with speckle and sidelobe artifacts suppressed. Hype-parameter selection is vital for the quality of the regularizing image. The Generalized Cross Validation(GCV), Robust Generalized Cross Validation(RGCV) and Stein's Unbias Risk Estimator(SURE) methods are applied in the non-quadratic regularization and the close form expressions of GCV, RGCV and SURE function are deduced. A fast algorithm and a generalized fix-point iteration algorithm are proposed to solve the regularization problem when the regularizing item is amended. The algorithm can be used for adaptive selection of the hyper-parameter. Numeric simulation proves the effectiveness of the proposed method.
出处
《太赫兹科学与电子信息学报》
2015年第6期937-941,共5页
Journal of Terahertz Science and Electronic Information Technology
基金
国家自然科学基金资助项目(61271417)
关键词
正则参数
广义交叉验证
鲁棒广义交叉验证
Stein无偏风险估计
特征增强
regularization parameter
Generalized Cross Validation
Robust Generalized Cross Validation
Stein's Unbias Risk Estimator
feature enhancement