基于储层网络算法的带噪图像识别技术研究与应用

Research and Application on Reservoir Computing Networks for Noisy Image Recognition Technology

储层网络算法(RCN)是一种特殊的递归神经网络,其输入和递归关系是随机产生的,只对输出结果进行加权训练。RCN具有处理时域信息、易于训练和抗噪能力强的特点。通过优化RCN的参数,该算法在语音识别领域有较好的抗噪性。本研究的目的是将该算法拓展到图像处理领域,并证明通过参数调优,RCN在图像处理领域有着同样稳定的抗噪性。基于优化参数后的RCN手写数字识别器在以MNIST数据分析库为基准的干净测试样本上进行测试,错误率降低到0.81%,其降噪器能够有效地过滤掉各种类型的噪声。

Reservoir Computing Networks(RCN)are a special type of recursive neural networks,which the input and the recursive connections are randomly generated.Only the output weights are trained.Besides the ability to process temporal information,the key points of RCN are easy training and robustness against noise.Through tuning the parameters of RCN,evaluation in the domain of noise robust speech recognition proves that this method is effective.The aim of this work is to extend that study to the field of image processing,by showing that the proposed parameter tuning procedure is equally valid in the field of image processing.In particular,we investigate the potential of RCN in achieving competitive performance on the MNIST dataset by following the aforementioned parameter optimizing strategy.The experiments demonstrate that the proposed RCN-based handwritten digit recognizer achieves an error rate of 0.81 percent on the clean test data of the MNIST benchmark and the proposed RCN-based denoiser can effectively filter the various types of noise.