基于OVR-SVM及fMRI的词语识别认知状态分类研究  

Research on Cognitive State Classification of Word Recognition Based on OVR-SVM and fMRI

作　　者：何享[1,2] 张笑非 HE Xiang;ZHANG Xiaofei(School of Foreign Language,Jiangsu University of Science and Technology,Zhenjiang 212003;School of Foreign Language,Nanjing University,Nanjing 210023;Faculty of Information,Beijing University of Technology,Beijing 100124;School of Computer,Jiangsu University of Science and Technology,Zhenjiang 212003)

机构地区：[1]江苏科技大学外国语学院,镇江212003 [2]南京大学外国语学院,南京210023 [3]北京工业大学信息学部,北京100124 [4]江苏科技大学计算机学院,镇江212003

出　　处：《计算机与数字工程》2021年第6期1158-1162,共5页Computer & Digital Engineering

基　　金：教育部人文社科基金项目“声音条件下阅读认知加工机制的研究”(编号:14YJC740030)资助。

摘　　要：词语识别是人脑的一种高级认知过程。论文基于OpenfMRI开源数据集,利用脑图Dosenbach进行特征提取,采用OVR-SVM作为分类器对词语识别及相关认知状态进行了分类器的训练,使用分层K折交叉验证和ROC曲线对分类器进行了性能评估。实验结果显示,采用线性核、多项式核、RBF核、以及Sigmoid核的几种OVR-SVM中,RBF核的分类性能在统计上都显著高于其他三种核对应的SVM。实验结果表明,OVR-SVM作为一种计算模型可以有助于对词语识别认知状态的识别。Word recognition is a high-level cognitive process of human brain.This paper is based on the fMRI dataset provid⁃ed by OpenfMRI project,brain atlas of Dosenbach is used for feature extraction.The OVR-SVM is used as the classifier to train the word recognition and related cognitive state,and the performance of the classifier is evaluated using hierarchical K-fold cross-vali⁃dation and ROC curve.The experimental results show that in several OVR-SVMs with linear kernel,polynomial kernel,RBF ker⁃nel,and sigmoid kernel,the classification performance of RBF kernel is statistically higher than that of the other three kernels.The experimental results indicate that OVR-SVM can be used as a computational model to effectively identify the cognitive state of word recognition.

关 键 词：词语识别 OVR支持向量机 功能磁共振成像 Dosenbach脑图 ROC曲线 

分 类 号：O235[理学—运筹学与控制论]