基于改进ART2算法的模式识别  

Based on the Improved ART2 Algorithm of Pattern Recognition

作　　者：尚小晶[1] 周宇航[1] 

机构地区：[1]吉林建筑大学城建学院,长春130111

出　　处：《吉林建筑大学学报》2017年第2期97-101,共5页Journal of Jilin Jianzhu University

基　　金：吉林省教育厅"十二五"科学技术研究项目(2014596)

摘　　要：手势动作的识别研究以成为康复医学的热点问题.本文主要针对7种难以区分的手势动作进行识别,首先采用小波变换将有效的肌电信号与噪声信号分离,然后对肌电信号特征提取,将绝对值积分、方差、3阶AR模型系数和功率谱比值这几个特征参数联合,充分反映不同动作的相关信息,并将其作为分类器的输入.ART2神经网络作为较成熟的分类器容易陷入局部极小值,本文提出模拟退火算法对其改进,对向下曲腕、向后曲腕、向侧曲腕,手掌伸展、握拳、反掌,食指伸展7种手势动作识别分类.实验效果显著,比传统的网络识别率有明显提高.Gestures recognition study is become a hot issue of rehabilitation medicine. This paper mainly aims at the gestures of the seven kinds of difficult to distinguish between identification. First of all, using wavelet transform will effective electrocardiography signal and noise signal separation, then the electrocardiography signal feature extrac- tion, the integral absolute value, variance, 3 order auto regressive（AR） model coefficients and this a few characteris- tic parameters of joint power spectrum ratio, fully reflect the information of different actions, as a classifier input. As a more mature ART2 neural network classifier easy to fall into local minimum value. This paper simulated annealing algorithm is put forward for its improvement, the curved downward wrist, wrist, back to the side of the wrist, palm

关 键 词：手势动作 肌电信号 ART2神经网络 模拟退火算法 

分 类 号：TP2[自动化与计算机技术—检测技术与自动化装置]