解码动物转向行为的ICA-小波特征提取方法-《郑州大学学报(工学版)》

文章信息/Info

Title:: ICA-wavelet Feature Extraction Method for Decoding of Pigeon Turning

Author(s):: Dong Fangfang Shang Zhigang Liu Xinyu Wanhong; School of Electrical Engineering, Zhengzhou University, Zhengzhou, Henan 450001

Keywords:: local field potentials; independent component analysis; wavelet decomposition; time-frequency analysis

摘要:: 针对提取局部场电位(LFP)用于运动意图解码的特征时,存在LFP信噪比低、编码时间窗难以确定等问题,提出了一种结合独立成分分析(ICA)与小波分解的特征提取方法,用于动物转向行为的神经信息解码.首先结合动物运动行为视频与LFP信号时频分析方法,确定编码时间窗的范围;然后用ICA对时间窗内的LFP进行去噪处理,提高LFP信噪比;接着利用小波分解进一步确定LFP编码频带,并通过滑窗方法计算频带内的时序能量,构建编码特征;最后采用k近邻方法对编码特征进行分类,验证其解码性能.实验结果表明,利用提出的特征提取方法,经过1 000次交叉互验证,分类正确率达到(92.35±5.87)％,能够准确稳定地解码动物的转向行为.

Abstract:: In order to overcome the problems as low signal-to-noise ratio (SNR) of LFP and difficulty in identifying encoding time window when extract the features of motion intention,a method that combines independent component analysis (ICA) with Wavelet was presented to extract the features of turning.Firstly,the motion videos of animals were analyzed and the time-frequency diagrams of LFP were plotted to determine the time window of signal which encoded the motion information.Then,ICA was used to increase the SNR of LFP.Thirdly,the encode bands of LFP were extracted by wavelet method as well as the encode features were extracted by sliding time window method.Lastly,k-nearest neighbor method was used to classify the encode features.And via 1 000 times cross validation the precision was(92.35 ±5.87)％,the results showed that it could decode reliably the motion intention of animals.