论文部分内容阅读
提出了一种改进型隐马尔可夫模型／神经网络混合分类器，该分类器将隐马尔可夫模型的时间校正能力与神经网络的静态区分能力结合在一起。它首先利用循环无跳转ＨＭＭ模型时每一测试特征序列进行全状态分割．将Ｔ帧特征序列按时间演化顺序校正成Ｎ帧平均状态序列。然后将其作为ＲＢＦ网络的输入矢量进行分类。实验结果表明，该分类器比单纯的神经网络或隐马尔可夫模型分类器具有更好的分类效果。
An improved Hidden Markov Model / neural network hybrid classifier is proposed. The classifier combines the time-correcting ability of Hidden Markov Model with the static discriminating ability of neural network. It first uses state-by-state partitioning of each test signature sequence using a loop-free jump HMM model. The T-frame feature sequence is corrected to the N-state average state sequence in time evolution order. It is then categorized as the input vector to the RBF network. Experimental results show that this classifier has a better classification effect than simple neural network or hidden Markov model classifier.