Mode recognition method on multi-feature conditions

• Main Authors: CHEN XINGHAN, DAI HAIWEI, YU DI, ZHOU YONGJIANG, WU BINGSHUAI
• Format: Patent
• Language: Chinese; English
• Published: 22.12.2017
• Subjects: CALCULATING; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention provides a mode recognition method on multi-feature conditions. The method comprises: step 1, establishing a directed connected graph model of a target mode needing to be recognized; step 2, defining features, expressed by means of functions, of connected nodes and edges; step 3, obtaining a real-time observation signal and carrying out preprocessing to obtain a usable observation signal; step 4, calculating feature functions of all nodes and edges in the directed connected graph model by using the usable observation signal; step 5, calculating current state probabilities of all nodes and transition probabilities of all edges based on the feature functions; step 6, repeating the steps from step 2 to step 5; and step 7, recognizing an optimal matching mode by decoding. 本发明提供了种多特征条件下的模式识别方法，包括：步骤1、建立所需识别目标模式的有向连通图模型；步骤2、定义连通的节点与边的特征，所述特征用函数的形式表现；步骤3、获得实时观测信号，并预处理为可使用观测信号；步骤4、根据可使用观测信号计算有向连通图模型中所有节点以及边的特征函数；步骤5、根据特征函数计算当前各节点的状态概率与各边的转移概率；步骤6、反复进行步骤二至步骤五，并记录每轮的相关信息；步骤7、通过译码识别最佳匹配模式。