基于多层前馈神经网络的二维自适应消噪方法

A Novel Method of 2-D Adaptive Noise Cancellation Based on Multilayer Feedforward Neural Networks

提出了一种基于多层前馈神经网络对带噪声样本进行二维去噪声建模的新方法,分析了神经网络的结构和原理并改进了反向传播训练算法,从理论上进行了论证。仿真结果表明此方法能较好地实现噪声消除,相对于传统的线性噪声对消器和基于Hopfield网络的消噪方法,具有更优越的性能,保持了原始信号的完整性,达到了有效抑制噪声的目的。

A novel method to model 2-D nonlinear system using noise-contaminated samples based on muhilayer feedforward neural network is proposed. The structure and principal of neural network are investigated with improved back propagation algorithm which is theoretically proved. Simulation results demonstrate that the method has a better performance in noise cancellation than classical linear noise canceller and Hopfield noise cancellation, the integrality of original signal with preferable effect is kept, and that the noise is restrained efficiently.