Motion Intent Recognition in Intelligent Lower Limb Prosthesis Using One-Dimensional Dual-Tree Complex Wavelet Transforms

Table 4

Comparison of our method with other methods under user-dependent classification.

Reference	Subjects	Position of sensors	Feature extraction	Classifier	Type of motion state		Accuracy
					Steady	Transitional
Young et al. 2016 [16]	Eight transfemoral amputees	Prosthesis	Statistical features	DBN	5	8	90.00%
Liu et al. 2017 [14]	Three able-bodied two amputees	Prosthesis	Statistical features	HMM	5	\	95.80%
Zheng et al. 2017 [32]	Six transfemoral amputees	Prosthesis	Statistical features	SVM + QDA	\	8	94.90%
Su et al. 2020 [19]	Ten able-bodied	Healthy side	Statistical features	SVM	5	8	95.12%
Su et al. 2019 [21]	Ten able-bodied	Healthy side	Self-selected features from CNN	Softmax	5	8
	One amputee
Our method	Ten able-bodied	Healthy side	Five-layer low-frequency coefficients of 1D-DTCWT	SVM	5	—
					—	8
					5	8
	One amputee				5	—
					—	8
					5	8