论文部分内容阅读
为克服神经网络受噪声和冗余特征的影响而出现过拟合,提出一种自适应级联神经网络(ACNN)及学习算法。ACNN从少量特征开始学习,在学习过程中根据特征对分类的有效性增加新特征,用映射递归算法调节权值,逐步确定网络结构,使其含有最少数目的输入和隐层神经元。此方法应用于区分两种思维状态下的脑电信号(EEG),经训练的网络对测试段的分类正确率为83.1%,与文献[1]中采用BP网络的结果相比,显示了ACNN较好的分类能力。
In order to overcome the influence of noise and redundancy in neural network, an adaptive cascade neural network (ACNN) and learning algorithm are proposed. ACNN starts learning from a few features, adds new features to the validity of the classification according to features during learning, adjusts weights by using mapping recursive algorithm, and gradually determines the network structure so that it contains a minimum number of input and hidden layer neurons. This method is applied to distinguish between EEG and EEG in two states of mind. The accuracy of the trained network in the classification of the test segment is 83.1%. Compared with the results of BP network in [1], ACNN Better classification ability.