TY - GEN
T1 - A novel preprocessing approach with soft voting for hand gesture recognition with a-mode ultrasound sensing
AU - Wei, Sheng
AU - Zhang, Yue
AU - Pan, Jie
AU - Liu, Honghai
N1 - Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2022/8/10
Y1 - 2022/8/10
N2 - To explore the potential of gesture recognition based on the A-mode ultrasound (AUS) interface in human-computer interaction (HCI), according to the characteristics of AUS signal, a novel preprocessing approach is designed, feature extraction is performed by the window analysis method, and four methods, Linear Discriminant Analysis (LDA), k-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Artificial Neural Network (ANN) for classification. The experimental results show that the single feature with the best results can achieve 91.63% accuracy on KNN. Meanwhile, by feature combination, we can achieve 91.91% accuracy on the KNN classifier, it is 3.60% higher than the highest recognition rate of 88.03% among linear fitting features, called KB features. Further, we learn the integration of soft voting for four classifiers, LDA, KNN, SVM, and ANN, and achieve the highest recognition rate of 92.32% on single features and can achieve 93.09% decoding rate on combined features, which is 4.01% higher than 89.08% among KB features with the soft voting method. The experimental results show that AUS has outstanding performance in gesture decoding.
AB - To explore the potential of gesture recognition based on the A-mode ultrasound (AUS) interface in human-computer interaction (HCI), according to the characteristics of AUS signal, a novel preprocessing approach is designed, feature extraction is performed by the window analysis method, and four methods, Linear Discriminant Analysis (LDA), k-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Artificial Neural Network (ANN) for classification. The experimental results show that the single feature with the best results can achieve 91.63% accuracy on KNN. Meanwhile, by feature combination, we can achieve 91.91% accuracy on the KNN classifier, it is 3.60% higher than the highest recognition rate of 88.03% among linear fitting features, called KB features. Further, we learn the integration of soft voting for four classifiers, LDA, KNN, SVM, and ANN, and achieve the highest recognition rate of 92.32% on single features and can achieve 93.09% decoding rate on combined features, which is 4.01% higher than 89.08% among KB features with the soft voting method. The experimental results show that AUS has outstanding performance in gesture decoding.
KW - A-mode ultrasound
KW - classification method
KW - feature extract
KW - gesture recognition
KW - preprocessing
KW - soft voting
UR - http://www.scopus.com/inward/record.url?scp=85136920632&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-13841-6_34
DO - 10.1007/978-3-031-13841-6_34
M3 - Conference contribution
AN - SCOPUS:85136920632
SN - 9783031138409
T3 - Lecture Notes in Computer Science
SP - 363
EP - 374
BT - Intelligent Robotics and Applications - 15th International Conference, ICIRA 2022, Proceedings, Part IV
A2 - Liu, Honghai
A2 - Yin, Zhouping
A2 - Liu, Lianqing
A2 - Jiang, Li
A2 - Gu, Guoying
A2 - Wu, Xinyu
A2 - Ren, Weihong
PB - Springer
T2 - 15th International Conference on Intelligent Robotics and Applications, ICIRA 2022
Y2 - 1 August 2022 through 3 August 2022
ER -