穿戴式跌倒检测中特征向量的提取和降维研究
|
摘要
穿戴式跌倒检测中老年人特征属性过多会造成维数灾难,影响后续跌倒检测精度。针对此问题,首先采用时域分析法提取初始特征向量集,用提出的改进核主成分分析算法(IKPCA)对特征向量进行降维,从而获得优质的特征向量集,使得后续的分类具有更好的效果。IKPCA算法首先利用I-RELIEF算法对初始特征向量集进行特征选择,然后计算跌倒特征向量的信息度量和相似度度量;最后根据跌倒特征向量的相似度度量剔除无效的跌倒特征向量。IKPCA算法不但保持核主成分分析算法(KPCA)较好的降维能力,而且扩充了较好的分类能力。利用真实的数据集进行实验,对比分析表明,相比其他算法,IKPCA算法能够得到更优质的特征向量数据集。
In wearable fall detection of the elderly,too much characteristics will cause the curse of dimensionality,and affect the accuracy of subsequent fall detection. To solve this problem,this paper used time domain analysis method to extract feature vector. The proposed improved kernel principal component analysised( IKPCA) algorithm was used to reduce the feature vectors,so as to obtain high-quality feature vectors,which made the subsequent classification more effective. IKPCA algorithm firstly used the I-RELIEF algorithm to select the initial feature vectors,then calculated the information measure and similarity measure of the falling feature vectors. Finally,according to the similarity measurement of the falling feature vectors,it eliminated the invalid falling feature vectors. The IKPCA algorithm can not only keep better dimensionality reduction ability of the KPCA algorithm,but also expands better classification ability. Experiments ran on real data sets. The comparative analysis shows that,compared with other algorithms,the IKPCA algorithm can obtain higher-quality feature vector data set.
In wearable fall detection of the elderly,too much characteristics will cause the curse of dimensionality,and affect the accuracy of subsequent fall detection. To solve this problem,this paper used time domain analysis method to extract feature vector. The proposed improved kernel principal component analysised( IKPCA) algorithm was used to reduce the feature vectors,so as to obtain high-quality feature vectors,which made the subsequent classification more effective. IKPCA algorithm firstly used the I-RELIEF algorithm to select the initial feature vectors,then calculated the information measure and similarity measure of the falling feature vectors. Finally,according to the similarity measurement of the falling feature vectors,it eliminated the invalid falling feature vectors. The IKPCA algorithm can not only keep better dimensionality reduction ability of the KPCA algorithm,but also expands better classification ability. Experiments ran on real data sets. The comparative analysis shows that,compared with other algorithms,the IKPCA algorithm can obtain higher-quality feature vector data set.
引文
[1] Okeyo G,Chen Liming,Wang Hui,et al. Dynamic sensor data segmentation for real-time knowledge-driven activity recognition[J].Pervasive and Mobile Computing,2014,10(2):155-172.
[2] Storm F A,Heller B W,MazzàC. Step detection and activity recognition accuracy of seven physical activity monitors[J]. PLo S One,2015,10(3):e0118723.
[3] Kai Kunze,Lukowicz P. Sensor placement variations in wearable activity recognition[J]. IEEE Pervasive Computing,2014,13(4):32-41.
[4] Xie Feng,Song A,Ciesielski V. Genetic programming based activity recognition on a smartphone sensory data benchmark[C]//Proc of IEEE Congress on Evolutionary Computation. Piscataway,NJ:IEEE Press,2014:2917-2924.
[5] Fahad L G,Tahir S F,Rajarajan M. Activity recognition in smart homes using clustering based classification[C]//Proc of the 22nd International Conference on Pattern Recognition. Piscataway,NJ:IEEE Press,2014:1348-1353.
[6] Iso T,Yamazaki K. Gait analyzer based on a cell phone with a single three-axis accelerometer[C]//Proc of the 8th Conference on HumanComputer Interaction with Mobile Devices and Services. New York:ACM Press,2006:141-144.
[7] Xie Shengkun. Principal component analysis based feature extraction methods applied to biomedical and communication network data[D].Philosophy:University of Guelph,2010.
[8]钱叶魁,陈鸣,叶立新,等.基于多尺度主成分分析的全网络异常检测方法[J].软件学报,2012,23(2):361-377.(Qian Yekui,Chen Ming,Ye Lixin,et al. Network-wide anomaly detection method based on multiscale principal component analysis[J]. Journal of Software,2012,23(2):361-377.)
[9] Xu Yong,Zhang D,Song Fengxi,et al. A method for speeding up feature extraction based on KPCA[J]. Neurocomputing,2007,70(4-6):1056-1061.
[10]Yang Mingjing,Zheng Huiru,Wang Haiying,et al. Combining feature ranking with PCA:an application to gait analysis[C]//Proc of International Conference on Machine Learning and Cybernetics. Piscataway,NJ:IEEE Press,2010:494-499.
[11]Yang J Y,Chen Y P,Lee G Y,et al. Activity recognition using one triaxial accelerometer:a neuro-fuzzy classifier with feature reduction[C]//Proc of International Conference on Entertainment Computing.Berlin:Springer:2007:395-400.
[12]Khan A M,Lee Y K,Lee S Y,et al. Human activity recognition via an accelerometer-enabled-smartphone using kernel discriminant analysis[C]//Proc of the 5th International Conference on Future Information Technology. Piscataway,NJ:IEEE Press,2010:1-6.
[13]Cheng Wenchang,Jhan D M. Triaxial accelerometer-based fall detection method using a self-constructing cascade-AdaBoost-SVM classifier[J]. IEEE Journal of Biomedical&Health Informatics,2013,17(2):411-419.
[14]Choudhury T,Borriello G,Consolvo S,et al. The mobile sensing platform:an embedded activity recognition system[J]. IEEE Pervasive Computing,2008,7(2):32-41.
[15]Sun Yijun. Iterative RELIEF for feature weighting:algorithms,theories,and applications[J]. IEEE Trans on Pattern Analysis and Machine Intelligence,2007,29(6):1035-1051.
[2] Storm F A,Heller B W,MazzàC. Step detection and activity recognition accuracy of seven physical activity monitors[J]. PLo S One,2015,10(3):e0118723.
[3] Kai Kunze,Lukowicz P. Sensor placement variations in wearable activity recognition[J]. IEEE Pervasive Computing,2014,13(4):32-41.
[4] Xie Feng,Song A,Ciesielski V. Genetic programming based activity recognition on a smartphone sensory data benchmark[C]//Proc of IEEE Congress on Evolutionary Computation. Piscataway,NJ:IEEE Press,2014:2917-2924.
[5] Fahad L G,Tahir S F,Rajarajan M. Activity recognition in smart homes using clustering based classification[C]//Proc of the 22nd International Conference on Pattern Recognition. Piscataway,NJ:IEEE Press,2014:1348-1353.
[6] Iso T,Yamazaki K. Gait analyzer based on a cell phone with a single three-axis accelerometer[C]//Proc of the 8th Conference on HumanComputer Interaction with Mobile Devices and Services. New York:ACM Press,2006:141-144.
[7] Xie Shengkun. Principal component analysis based feature extraction methods applied to biomedical and communication network data[D].Philosophy:University of Guelph,2010.
[8]钱叶魁,陈鸣,叶立新,等.基于多尺度主成分分析的全网络异常检测方法[J].软件学报,2012,23(2):361-377.(Qian Yekui,Chen Ming,Ye Lixin,et al. Network-wide anomaly detection method based on multiscale principal component analysis[J]. Journal of Software,2012,23(2):361-377.)
[9] Xu Yong,Zhang D,Song Fengxi,et al. A method for speeding up feature extraction based on KPCA[J]. Neurocomputing,2007,70(4-6):1056-1061.
[10]Yang Mingjing,Zheng Huiru,Wang Haiying,et al. Combining feature ranking with PCA:an application to gait analysis[C]//Proc of International Conference on Machine Learning and Cybernetics. Piscataway,NJ:IEEE Press,2010:494-499.
[11]Yang J Y,Chen Y P,Lee G Y,et al. Activity recognition using one triaxial accelerometer:a neuro-fuzzy classifier with feature reduction[C]//Proc of International Conference on Entertainment Computing.Berlin:Springer:2007:395-400.
[12]Khan A M,Lee Y K,Lee S Y,et al. Human activity recognition via an accelerometer-enabled-smartphone using kernel discriminant analysis[C]//Proc of the 5th International Conference on Future Information Technology. Piscataway,NJ:IEEE Press,2010:1-6.
[13]Cheng Wenchang,Jhan D M. Triaxial accelerometer-based fall detection method using a self-constructing cascade-AdaBoost-SVM classifier[J]. IEEE Journal of Biomedical&Health Informatics,2013,17(2):411-419.
[14]Choudhury T,Borriello G,Consolvo S,et al. The mobile sensing platform:an embedded activity recognition system[J]. IEEE Pervasive Computing,2008,7(2):32-41.
[15]Sun Yijun. Iterative RELIEF for feature weighting:algorithms,theories,and applications[J]. IEEE Trans on Pattern Analysis and Machine Intelligence,2007,29(6):1035-1051.