표 3. | Table 3. 기존편파 SAR 데이터셋(HH, HV)과 변환편파(LH)데이터가 포함된 데이터셋을 활용하여 다양한 CNN 모델의 이진분류 성능 비교 | Comparison of various CNN Models for binary classification performance using original polarization (HH, HV) SAR dataset and the dataset with the included transformed polarization (LH).

Dataset	Classifier	Precision(%)	Recall (%)	F1-score(%)	Accuracy(%)
Original	ResNet-18	87.81	82.99	85.27	83.58
Origin+LH		93.23	88.23	90.65	90.19
Original	ResNet-34	87.81	81.63	84.51	82.90
Origin+LH		93.01	89.62	91.25	90.73
Original	ResNet-50	89.45	79.24	83.97	82.71
Origin+LH		92.24	89.27	90.70	90.15
Original	ResNeXt-50 (32×4d)	89.14	80.00	84.20	82.87
Origin+LH		92.55	89.15	90.81	90.27
Original	DenseNet-121	90.63	81.30	85.66	84.42
Origin+LH		92.16	89.69	90.89	90.31
Original	DenseNet-161	89.71	80.00	84.49	83.21
Origin+LH		93.12	87.50	90.21	89.75
Original	EfficientNet-b0	85.11	77.99	81.36	79.53
Origin+LH		92.45	88.15	90.22	89.71
Original	EfficientNet-b1	87.15	80.00	83.35	81.71
Origin+LH		92.80	88.42	90.54	90.04