摘要
提出一种基于岭回归协助稀疏表示的红外小目标检测方法。该方法分别采用二维高斯模型和正态分布随机矩阵生成红外小目标样本和背景样本,继而建立超完备字典。红外小目标检测包括两个阶段,在第一阶段利用岭回归表示快速计算所有测试样本的岭回归重建误差;在第二阶段,根据岭回归重建误差自适应选择候选目标,并计算其稀疏表示重建误差实现目标检测。对提出的方法进行了实验验证,结果表明:提出的方法具有较快的速度和较强的鲁棒性。
An infrared small target detection method is proposed based on sparse representation assisted by ridge regression. The proposed method constructs over-complete dictionary with target samples and background samples which are produced by two-dimensional Gaussian model and normal random matrix respectively. Infrared small target detection consists of two stages. In the first stage, the construction errors of detect samples are calculated fast using ridge regression. In the second stage, candidate samples are selected adaptively by ridge regression reconstruction errors, and infrared small targets are detected with the reconstruction errors of the candidate samples selected which are computed with sparse representation. The experimental results on several infrared images show that the proposed method is faster and more robust than the existing methods.
出处
《光电工程》
CAS
CSCD
北大核心
2014年第2期81-87,共7页
Opto-Electronic Engineering
基金
国家自然基金资助项目(61175035
61379105)
关键词
红外小目标检测
稀疏表示
岭回归
重建误差
infrared small target detection
sparse representation
ridge regression
reconstruction error
