Cross-modal Feature Decoupling and Focalizing Network for Robust UAV-based Road Traffic Scenes Semantic Segmentation

方法	骨干网络	道路	建筑物	摩托车	汽车	卡车	树木	行人	交通设施	mIoU (%)
CNN-based
U-Net^[27]	UNet	93.50	68.50	12.70	67.10	68.10	67.90	0.00	10.80	52.70
UperNet^[28]	Swin-B	94.00	72.90	19.10	71.50	73.20	74.60	0.00	12.90	55.40
RTFNet^[8]	ResNet152	94.60	81.80	21.30	78.20	80.20	81.60	0.00	20.70	60.00
Attention-based
FEANet^[26]	ResNet152	94.70	82.30	53.90	83.00	83.00	82.50	0.00	47.80	67.60
EAEFNet^[25]	ResNet152	95.00	85.00	60.60	84.50	85.50	84.00	0.00	53.10	70.10
CMNeXt-B2^[29]	MiT-B2	94.90	81.40	53.60	81.00	83.00	82.60	22.30	49.80	70.00
CMNeXt-B4^[29]	MiT-B4	95.20	85.20	57.80	81.30	83.50	82.70	22.70	52.20	71.40
CMX-B2^[18]	MiT-B2	95.10	85.70	60.10	84.00	84.70	84.60	19.60	51.00	72.00
CMX-B4^[18]	MiT-B4	95.40	86.90	62.10	84.20	85.10	85.30	23.00	49.90	72.90
Mamba-based
CM-SSM^[20]	Mamba	94.70	84.50	61.80	86.00	85.30	83.70	33.90	57.90	74.30
Sigma^[19]	Mamba	94.80	86.30	64.60	86.50	86.50	84.10	35.40	58.10	75.30
CDFNet（本研究）	ResNet152	95.80	88.30	67.20	87.50	87.50	85.10	38.30	61.40	78.40

Table 2. Table 2CDFNet ablation study
View table
View in Article
Table 2. Table 2CDFNet ablation study
基线模型 CFDM FAD mIoU/%
✔ 72.10
✔ ✔ 76.20
✔ ✔ 74.80
✔ ✔ ✔ 78.40

Table 3. Table 3CDFNet ablation study
View table
View in Article
Table 3. Table 3CDFNet ablation study
方法可见光失效热红外失效模态正常
U-Net^[27] 45.90 33.10 52.70
FEANet^[26] 59.90 39.10 67.60
EAEFNet^[25] 65.00 41.70 70.10
CDFNet 67.10 46.60 78.40

Tools

Get Citation

Copy Citation Text

Qingwang WANG, Junlin OUYANG, Pengcheng JIN, Tao SHEN. Cross-modal Feature Decoupling and Focalizing Network for Robust UAV-based Road Traffic Scenes Semantic Segmentation[J]. Remote Sensing Technology and Application, 2025, 40(4): 864

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites