基于旋转扩展和稀疏表示的鲁棒遥感图像目标识别  

Robust Remote Sensing Image Target Recognition Based on Extending Training Set by Rotation and Sparse Representation

作　　者：殷飞[1] 焦李成[1] 

机构地区：[1]西安电子科技大学智能信息处理研究所智能感知与图像理解教育部重点实验室,西安710071

出　　处：《模式识别与人工智能》2012年第1期89-95,共7页Pattern Recognition and Artificial Intelligence

基　　金：国家自然科学基金项目(No.60702062);国家863计划项目(No.2009AA12Z210);国家973计划项目(No.2006CB705707)资助

摘　　要：针对含有残缺图像的遥感图像目标识别问题,提出一种基于旋转扩展和稀疏表示的目标识别方法.首先对训练集进行旋转扩展,使得测试图像能近似用训练集稀疏表示,然后通过求解一个l1范数最小化问题得到测试图像相对于训练集的一个稀疏表示,进而根据不同类对应的稀疏表示对测试图像的近似程度进行识别.与代表性的方法进行比较,实验结果与分析表明,该方法识别率优于已有方法,对残缺图像的识别有较好的鲁棒性,且在小样本、低采样率情况下也能保持较好的识别性能.A remote sensing target recognition method based on extending training set by rotation and sparse representation is proposed for remote sensing image recognition which includes incomplete images.Firstly,the training set is extended by rotating every image to guarantee the test image can be expressed approximately as a sparse linear combination of the extended training set.Then,based on the sparse representation computed by l1 norm minimization,the test image is classified according to the degree of approximation to test image of the sparse representation corresponding to different classes.The experimental results show that the proposed algorithm surpasses the existing methods in recognition rate,is robust to incomplete images and still gets good results under the condition of small scale samples and low sampling rate images.

关 键 词：遥感目标识别 残缺图像 稀疏表示 旋转扩展 l1范数最小化 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]