摘要
与高斯噪声相比,拖尾有更多的异常值,利用传统的神经网络不能有效的检测信号。该文提出一种基于中值滤波的鲁棒神经网络进行处理,首先利用中值滤波抑制异常值,进一步利用BP(Back Propagation)神经网络消除残留噪声,检测目标信号。基于误差分析的实验结果表明,与传统神经网络相比,所提出的方法不仅能更好地消除拖尾噪声,有效检测信号,而且能有效检测高斯噪声中的目标信号,具有很好的鲁棒性和自适应特性。
Compared with Gaussian noise, Heavy-tailed noise has more outliers, and the traditional neural network can not suppress outliers. A new neural network based on median filter is proposed. After suppressing the outliers in signal through median filter, the BP （Back Propagation） is used and remained noise is eliminated further. The experiment based on the error analyses shows that compared with the traditional neural network, the proposed method can suppress heavy-tailed noise and detect target signal more effectively. It can, perform well for both heavy-tailed noise and Gaussian noise background, which shows its robustness and adaptiveness.
出处
《电子与信息学报》
EI
CSCD
北大核心
2007年第8期1864-1867,共4页
Journal of Electronics & Information Technology
基金
全国优秀博士学位论文作者专项基金(200237)资助课题
关键词
信号检测
神经网络
BP算法
拖尾噪声
中值滤波
Signal detection
Neural network
BP algorithm
Heavy-tailed noise
Median filter
