【摘要】 针对经典马尔可夫模型没有考虑模型应用中状态、观测量间的上下文相关性以及状态转移概率动态性、可变性,提出一种模糊深隐马尔可夫模型。该模型通过增加观测值间的相关性、解决概率转移问题中的不确定性和改进参数优化算法,使之能够较好地应用于强噪声、训练数据缺损等情形的模式识别中。理论证明,显式模糊深隐马尔可夫模型在同等模型复杂度下具有模型优化程度高、区分度好、误识率低、鲁棒性高的特性。更多还原

【Abstract】 To avoid the disadvantage of hidden Markov model which doesn’t consider the contextual relevant relationship of states and observations and the changeability of transfer probability, an improved model called fuzzy buried Markov model is put forward in this paper. Adding relationship among the different observations, resolving the problem of transfer probability uncertainty and ameliorating the parameter optimization arithmetic make the novel model be suitable to apply in pattern identification with much noise and losing of some training data. Compared with other model of the same complexity, fuzzy buried Markov model shows more good character such as optimum performance, better division degree, lower error rate and lustihood, which can be proved by graphical theory.更多还原