基于空间分割前向竞争网络的信号识别

本文提出运用空间分割竞争网络结合Kohonen学习规则实现有规则模拟信号分类识别方法.以信号谐波、信号面积和信号正负面积之比等参数为识别特征,利用软边界处理的Kohonen训练规则,将识别特征值分布特性通过软边界技术快捷地训练到寻址层隶属度矩阵中.该方法减少了训练样本、提高学习速度,克服了由于大样本训练导致空间块隶属度统计不足的缺点.实验证明该方法识别常见信号特征有效性高.

Signal Identification Based on Area Division Forward Competitive Neural Network

This paper proposes a new signal identification approach, which combines Area Division Competitive Neural Network (ADCNN) with Kohonen learning rule. The sum of ten harmonics, the waveform area of one period and the ratio of the positive area to negative area in one cycle are selected as identification characteristic parameters. The ADCNN is trained with Kohonen rule, and the distribution characteristic of signal identification feature is trained into the membership matrix of the network addressing layer using soft-boundary processing technology. This method reduces the training sample, increases the learning speed and solves the problem of insufficient membership of space-area block when the sample is too large. Experiment proves that the method can identify common analog signal features effectively.