摘要
针对现有声目标识别技术鲁棒性较差的实际情况,提出了一种监督性流形学习算法—保局判别投影(LPDP)算法。算法在流形学习保局投影(LPP)算法的基础上,引入了控制类间和类内距离的改进最大边缘标准(MMMC),使得这种特征提取算法既具有线性流形学习算法样本外点学习的优点,又能够有效解决小样本问题,并能在后续的分类中取得良好效果。通过在公开数据库和战场实际数据的特征提取实验,结果表明算法的识别率和稳定性均优于现有其他算法。
This paper presents a new supervised manifold learning algorithm called locality preserving discriminant projections to solve the problem of poor robustness in acoustic target recognition.This algorithm is based on locality preserving projections（LPP）,and the method called modified maximum margin criterion（MMMC） which is adopted to automatically explore the optimal linear transformation for translation and resealing.So the proposed algorithm not only can solve the small sample size problem but also has the ability of out-of-sample learning and can achieve good results in classification.The proposed algorithm is tested on public databases and actual battlefield data.Experimental results show that the proposed algorithm is more precise and stable than the other methods.
出处
《电路与系统学报》
CSCD
北大核心
2011年第5期100-106,共7页
Journal of Circuits and Systems
基金
国家自然科学基金(60872113)
关键词
声目标识别
保局投影
改进最大边缘标准
acoustic target recognition
locality preserving projections
modified maximum margin criterion
