Infrared UAV detection based on multi-channel interactive attention mechanism and edge contour enhancement

Fig. 6. Qualitative results of different detection methods （for better visualisation， the target area is enlarged in the lower left corner. Correctly detected targets and false alarm regions are shown with red and yellow circles， respectively）

Download full size

View in Article

Fig. 7. Qualitative results obtained by the MCIA， ECE and MLFF modules （for better visualization， the target area is enlarged in the lower right corner）

Download full size

View in Article

Table 1. IoU, Pd, Fa values obtained by different methods on NUDT-SIRST dataset

View table

View in Article

Table 1. IoU, Pd, Fa values obtained by different methods on NUDT-SIRST dataset

Model	NUDT（Tr=50%）	NUAA（Tr=50%）	IRSTD-1k （Tr=50%）
Model	Pd/ Fa/ IoU	Pd/ Fa/ IoU	Pd/ Fa/ IoU
Top-Hat^［27］	78.41/166.7/20.72	79.84/1012/7.143	75.53/1346/8.74
IPI^［25］	74.49/41.23/17.76	85.55/11.47/25.67	80.75/16.68/24.98
RIPT^［24］	91.85/344.3/29.44	79.08/22.61/11.05	77.47/28.41/14.33
MPCM^［23］	84.32/356.8/27.28	83.27/17.74/12.35	69.73/29.47/11.68
PSTNN^［26］	66.13/44.17/22.40	77.95/29.11/14.85	22.40/74.15/54.37
ACM^［21］	95.68/9.34/68.28	92.93/3.45/72.46	90.35/12.42/60.47
MTU-Net^［22］	97.35/3.89/83.83	98.55/1.30/73.12	91.52/1.71/63.12
RDIAN^［20］	97.98/8.49/78.23	98.23/1.45/69.71	89.06/1.34/62.21
ALC-Net^［19］	96.51/9.26/81.43	92.18/37.23/67.84	84.36/62.12/60.25
MCIAECE-Net	98.83/2.09/85.11	98.09/1.21/69.89	91.64/1.08/61.16

Table 2. Ablation Study of MCIA, ECEM and MLFF on the NUDT-SIRST Dataset

View table

View in Article

Table 2. Ablation Study of MCIA, ECEM and MLFF on the NUDT-SIRST Dataset

Baseline	MCIA	ECE	MLFF	Pd	Fa	IoU
√	×	×	×	96.88	4.49	78.23
√	√	×	×	96.93	3.3	81.77
√	×	√	×	97.56	4.14	80.38
√	×	×	√	98.20	2.88	80.04
√	×	√	√	98.51	2.29	80.59
√	√	×	√	97.56	2.65	83.27
√	√	√	×	97.09	4.31	80.83
√	√	√	√	98.83	2.09	85.11

Table 3. Performance comparison of different methods
View table
View in Article
Table 3. Performance comparison of different methods
Method Parameters/MB GFLOPS FPS
ACM 1.52 0.55 36.14
ALC-Net 0.52 1.48 29.49
MTU-Net 12.75 6.22 110.01
RDIAN 0.22 3.72 100
MCIAECE 3.88 3.68 47.96

Table 4. Comparison of MCIA with other attention mechanisms
View table
View in Article
Table 4. Comparison of MCIA with other attention mechanisms
Method Pd Fa IoU
w/o MCIA 98.51 2.29 80.59
CBAM 98.64 3.53 82.45
SE 98.23 4.66 81.14
DCFE-Net 98.83 2.09 85.11

Table 5. Performance comparison of MLFF
View table
View in Article
Table 5. Performance comparison of MLFF
w/o ECE X_i-1 X_i Y_i+1 Pd Fa IoU
√ √ × √ 98.51 2.31 79.08
√ × √ √ 97.09 2.63 80.83
√ √ √ √ 98.83 2.09 85.11

Tools

Get Citation

Copy Citation Text

Su-Zhen NIE, Jie CAO, Qun HAO, Xu-Ye ZHUANG. Infrared UAV detection based on multi-channel interactive attention mechanism and edge contour enhancement[J]. Journal of Infrared and Millimeter Waves, 2025, 44(3): 335

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites