Table 12 Comparison of classification accuracy across different spectral features and methods with 50% window overlap during feature extraction.

From: An efficient deep learning approach with frequency and channel optimization for underwater acoustic target recognition

Feature

Method

Accuracy (mean ± std)

@time-train (s)

Avg.

p-value

MFLOPs

#Params(M)

1

2

3

4

5

STFT

ResNet18

72.55±1.36

@656.9s

69.41±2.29

@672.5s

70.15±1.30

@656.5s

79.06±0.46

@702.0s

73.44±1.36

@867.7s

72.92

0.0511

4288.81

11.17

RCMoE-b.

68.30±1.69

@792.1s

67.36±2.77

@776.7s

68.13±1.63

@751.8s

78.32±1.59

@770.2s

71.98±1.55

@825.1s

70.82

0.017

4288.82

11.19

CFTAnet

68.71±0.66

@615.3s

70.75±1.22

@585.8s

68.20±1.09

@580.1s

74.74±0.98

@573.1s

68.29±1.29

@629.0s

70.14

0.884

411.65

0.55

FCResNet5

70.12±0.72

@428.8s

69.63±1.21

@475.8s

69.01±0.75

@452.7s

77.57±0.57

@427.2s

71.44±0.68

@434.5s

71.55

–

196.12

0.65

Mel

ResNet18

68.72±1.76

@619.2s

70.14±1.61

@686.7s

69.13±1.43

@656.7s

77.34±1.06

@648.1s

73.75±1.03

@726.9s

71.82

0.000*

4288.81

11.17

RCMoE-b.

67.02±1.45

@1015.1s

65.98±3.50

@795.6s

68.11±1.57

@779.0s

76.25±1.15

@782.6s

71.91±1.95

@761.3s

69.45

0.004*

4288.82

11.19

CFTAnet

68.14±1.57

@593.0s

69.30±0.92

@582.9s

66.56±1.03

@550.9s

74.05±1.04

@547.1s

68.41±1.06

@568.6s

69.29

0.027

411.65

0.55

FCResNet5

69.03±1.17

@403.9s

70.54±1.43

@452.3s

67.73±1.05

@478.0s

76.75±0.45

@421.8s

70.81±0.77

@435.6s

70.97

–

196.12

0.65

CQT

ResNet18

67.60±1.28

@306.0s

65.52±1.60

@315.7s

64.02±1.17

@298.3s

69.28±1.32

@303.8s

68.12±1.53

@314.6s

66.91

0.000*

1514.76

11.17

RCMoE-b.

66.22±1.66

@369.1s

63.98±1.30

@358.9s

63.02±1.93

@329.5s

68.31±1.77

@313.9s

67.01±1.49

@322.0s

65.71

0.016*

1514.77

11.19

CFTAnet

66.04±2.16

@282.6s

65.68±1.66

@279.7s

63.89±1.18

@271.1s

68.71±1.89

@302.2s

65.59±1.46

@308.9s

65.98

0.225

142.23

0.51

FCResNet5

66.10±0.28

@305.1s

66.00±0.54

@294.8s

64.06±1.12

@283.6s

71.62±0.98

@254.0s

67.14±1.09

@296.5s

66.98

-

103.51

0.35

Gamma

-tone

ResNet18

65.82±1.37

@668.8s

64.88±2.37

@684.5s

60.93±1.24

@657.3s

70.57±0.97

@679.1s

66.76±0.91

@683.0s

65.39

0.000*

4288.81

11.17

RCMoE-b.

63.21±1.88

@794.8s

60.74±2.07

@753.7s

59.70±0.89

@704.1s

68.72±1.20

@752.6s

66.04±1.23

@812.4s

63.68

0.013*

4288.82

11.19

CFTAnet

62.55±1.67

@599.7s

64.11±1.01

@529.4s

61.86±1.58

@518.6s

65.35±1.31

@522.8s

64.06±1.02

@559.4s

63.59

0.054

411.65

0.55

FCResNet5

66.10±0.66

@408.7s

66.95±1.11

@413.2s

66.16±1.17

@415.0s

68.79±0.66

@429.1s

66.90±0.76

@405.3s

66.98

–

196.12

0.65

  1. * indicates p < 0.0167 (Bonferroni-corrected), denoting statistical significance.
Back to article page