基于改进脉冲耦合神经网络的数据降噪方法研究

Research on Data Noise Reduction Method Based on Modified PCNN

为了解决数据中存在噪声点降低了数据的质量,影响了对数据进一步分析的可靠性等问题,提出了一种基于改进脉冲耦合神经网络(pulse coupled neural network,PCNN)的数据降噪方法。该方法在保留基本PCNN模型一些特性的基础上将其简化,省去了部分参数的设置,并改进突触链接强度为自适应取值,添加了记录神经元点火次数的点火频次矩阵。根据神经元点火次数辨识并滤除噪声点,使得该方法能够简单有效地对数据进行降噪处理,改善了数据质量。实验结果表明了该方法不仅能够有效滤除低维数据中的噪声点,而且对高维数据中的噪声点去除效果较好,且均较好保持了原有数据的特征信息。

To solve the issues that the quality of data is reduced and the reliability of the further analysis for data is affected due to the noise points in data, a data noise reduction method based on modified PCNN is presented. It is modified that original PCNN model for the method, and the basic characteristics of PCNN model is retained. The method saves some parameters setting, the synaptic connection strength is improved as adaptive value and an ignition frequency matrix is added. It can record fired time of neurons. Noise points are identified and filtered according to the fired time of neurons, so that the method can denoise simply and effectively for data, and data quality is improved. The experimental results show that the method not only can filter out the noise points effectively in the low dimensional data, but also removal effect for the noise points in high dimensional data is also well, and both have the characteristic information of the original data.