Towards Integration of Domain Knowledge-Guided Feature Engineering and Deep Feature Learning in Surface Electromyography-Based Hand Movement Recognition

Table 2

Average hand movement recognition accuracies in comparison with the state of the arts on NinaProDB1-NinaProDB5.

Dataset	Machine-learning (ML) model	Type of ML model	Input of ML model	Num. of movements for classification	Window length
					50 ms	100 ms	150 ms	200 ms
NinaProDB1	Random forests [20]	Shallow learning	Incell 5 engineered features	50	N.A.	N.A.	N.A.	75.3%
	GengNet [12]	CNN	Raw sEMG	52	N.A.	N.A.	N.A.	77.8%
	AtzoriNet [11]	CNN	Raw sEMG	50	N.A.	N.A.	66.6% ± 6.4%	N.A.
	WeiNet [13]	CNN	Raw sEMG	52	81.7%	83.4%	84.4%	85.0%
	HuNet [10]	CNN-RNN	Phinyomark feature set	52	N.A.	N.A.	86.8%	87.0%
	MV-CNN [14]	CNN	3 feature sets	52	85.8%	86.8%	87.4%	88.2%
	Evolved CNN [22]	CNN	Raw sEMG	52	N.A.	N.A.	N.A.	81.4%
	ChengNet [16]	CNN	Multi-sEMG feature image	52	N.A.	N.A.	N.A.	82.5%
	PFNet	CNN	Raw sEMG + DWPTC	52	85.1 ± 4.6%	86.1 ± 4.4%	87.0 ± 4.3%	87.8 ± 4.2%
NinaProDB2	Random forests [20]	Shallow learning	5 engineered features	50	N.A.	N.A.	N.A.	75.3%
	AtzoriNet [11]	CNN	Raw sEMG	50	N.A.	N.A.	60.3 ± 7.7%	N.A.
	ZhaiNet [37]	CNN	sEMG spectrogram	50	N.A.	N.A.	N.A.	78.7%
	HuNet [10]	CNN-RNN	Phinyomark feature set	50	N.A.	N.A.	N.A.	82.2%
	MV-CNN [14]	CNN	3 feature sets	50	80.6%	81.1%	82.7%	83.7%
	Evolved CNN [22]	CNN	Raw sEMG	50	N.A.	71.0%	N.A.	71.6%
	PFNet	CNN	Raw sEMG + DWPTC	50	82.4 ± 5.6%	83.4 ± 5.5%	84.4 ± 5.6%	85.4 ± 5.1%
NinaProDB3	SVM [20]	Shallow learning	5 handcrafted features	50	N.A.	N.A.	N.A.	46.3%
	MV-CNN [14]	CNN	3 feature sets	50	N.A.	N.A.	N.A.	64.3%
	PFNet	CNN	Raw sEMG + DWPTC	50	64.8 ± 8.9%	66.3 ± 9.0%	67.3 ± 8.9%	68.3 ± 9.2%
NinaProDB4	Random forests [21]	Shallow learning	mDWT features	53	N.A.	N.A.	N.A.	69.1%
	MV-CNN [14]	CNN	3 feature sets	53	N.A.	N.A.	N.A.	54.3%
	PFNet	CNN	Raw sEMG + DWPTC	53	60.0 ± 8.2%	65.8 ± 7.7%	69.1 ± 7.5%	71.7 ± 7.4%
NinaProDB5	SVM [21]	Shallow learning	mDWT features	41	N.A.	N.A.	N.A.	69.0%
	ShenNet [18]	Stacking-based CNN	TD, FD, and TFD feature images	40	N.A.	N.A.	N.A.	72.1%
	MV-CNN [14]	CNN	3 feature sets	41	N.A.	N.A.	N.A.	90.0%
	PFNet	CNN	Raw sEMG + DWPTC	41	89.1 ± 3.6%	89.6 ± 3.4%	90.2 ± 3.3%	90.3 ± 3.2%

N.A. denotes not applicable, and bold entries indicate our proposed method.