计算机工程与设计
計算機工程與設計
계산궤공정여설계
COMPUTER ENGINEERING AND DESIGN
2010年
1期
157-160
,共4页
粒子群优化算法%优化算法%隐马尔可夫模型%隐马尔可夫模型优化%手写数字识别
粒子群優化算法%優化算法%隱馬爾可伕模型%隱馬爾可伕模型優化%手寫數字識彆
입자군우화산법%우화산법%은마이가부모형%은마이가부모형우화%수사수자식별
particle swarm optimization%optimization algorithm%hidden Markov model%hidden Markov model optimize%handwrite digits recognition
针对隐马尔可夫模型传统训练算法易收敛于局部极值的问题,提出一种带极值扰动的自适应调整惯性权重和加速系数的粒子群算法,将改进后的粒子群优化算法引入到隐马尔可夫模型的训练中,分别对隐马尔可夫模型的状态数与参数进优化.通过对手写数字识别的实验说明,提出的基于改进粒子群优化算法的隐马尔可夫模型训练算法与传统隐马尔可夫模型训练算法Baum-Welch算法相比,能有效地跳出局部极值,从而使训练后的隐马尔可夫模型具有较高的识别能力.
針對隱馬爾可伕模型傳統訓練算法易收斂于跼部極值的問題,提齣一種帶極值擾動的自適應調整慣性權重和加速繫數的粒子群算法,將改進後的粒子群優化算法引入到隱馬爾可伕模型的訓練中,分彆對隱馬爾可伕模型的狀態數與參數進優化.通過對手寫數字識彆的實驗說明,提齣的基于改進粒子群優化算法的隱馬爾可伕模型訓練算法與傳統隱馬爾可伕模型訓練算法Baum-Welch算法相比,能有效地跳齣跼部極值,從而使訓練後的隱馬爾可伕模型具有較高的識彆能力.
침대은마이가부모형전통훈련산법역수렴우국부겁치적문제,제출일충대겁치우동적자괄응조정관성권중화가속계수적입자군산법,장개진후적입자군우화산법인입도은마이가부모형적훈련중,분별대은마이가부모형적상태수여삼수진우화.통과대수사수자식별적실험설명,제출적기우개진입자군우화산법적은마이가부모형훈련산법여전통은마이가부모형훈련산법Baum-Welch산법상비,능유효지도출국부겁치,종이사훈련후적은마이가부모형구유교고적식별능력.
To solve the problem that easy to converge to local optimal solutions of hidden Markov model (HMM) training,a selfadaptive particle swarm optimization algorithm with disturbed extremum is presented and it is used in the training of HMM to optimize the state number and parameters of HMM.Comparing the proposed approach with Baum-Welch algorithm HMM training method,the hand-write digits recognition experimental results show that the proposed method is superior to the Bantu-Welch training method and make the trained HMM has better recognition ability.
