Table 6 Ablation study results of the proposed HSICNet model across Indian Pines, Pavia University, and Salinas datasets.

From: HSICNet a novel deep learning architecture for hyperspectral image classification in remote sensing and environmental monitoring

Configuration

Dataset

OA (%)

AA (%)

Kappa

F1-score

HSICNet (full model)

Indian Pines

98.63

97.42

0.98

0.976

Pavia University

99.14

98.30

0.99

0.986

Salinas

99.35

98.91

0.99

0.991

Without the spatial branch

Indian Pines

96.11

94.60

0.95

0.948

Pavia University

96.92

95.45

0.95

0.953

Salinas

97.22

96.01

0.96

0.963

Without spectral branch

Indian Pines

95.43

93.87

0.94

0.935

Pavia University

95.78

94.51

0.94

0.943

Salinas

96.45

95.19

0.95

0.951

Without spectral-spatial fusion

Indian Pines

94.70

92.90

0.93

0.924

Pavia University

94.88

93.67

0.93

0.936

Salinas

95.73

94.41

0.94

0.944

Without attention mechanism (basic fusion)

Indian Pines

96.78

95.31

0.96

0.956

Pavia University

97.02

95.82

0.96

0.961

Salinas

97.89

96.88

0.97

0.973

Without PCA, dimensionality reduction

Indian Pines

97.13

95.96

0.97

0.964

Pavia University

97.69

96.53

0.97

0.969

Salinas

98.11

97.32

0.98

0.978