Infrared Small Target Detection Algorithm Based on Real-Time Semantic Segmentation

Bin Shao; Hua Yang; Bin Zhu; Yi Chen; Rongping Zou

doi:10.3788/LOP221958

Laser & Optoelectronics Progress, Volume. 60, Issue 14, 1410006(2023)

Infrared Small Target Detection Algorithm Based on Real-Time Semantic Segmentation

Bin Shao^1,2,3, Hua Yang^1,2,3, Bin Zhu^1,2,3、*, Yi Chen^1,2,3, and Rongping Zou^1,2,3

Author Affiliations

¹College of Electronic Engineering, National University of Defense Technology, Hefei 230037, Anhui, China

²State Key Laboratory of Pulsed Power Laser Technology, Hefei 230037, Anhui, China

³Key Laboratory of Infrared and Low Temperature Plasma of Anhui Province, Hefei 230037, Anhui, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

Fig. 1. Network structure of RTIRSeg

Download full size

Fig. 2. Detail feature extraction network

Download full size

Fig. 3. Semantic information extraction network

Download full size

Fig. 4. Progressive feature fusion module

Download full size

Fig. 5. Segmentation results of proposed algorithm in multiple scenes. (a) Artificial construction background; (b) suburban background with sky and mountains; (c) less cloudy sky background; (d) clear sky background; (e) sky background with buildings; (f) (g) cloudy sky background

Download full size

Fig. 6. Segmentation effect of RTIRSeg in continuous frame of a video

Download full size

Fig. 7. ROC curves of six algorithms

Download full size

Table 1. Experimental environment
View table
Table 1. Experimental environment
Environment Configuration
CPU Intel（R）Xeon（R）CPU W-2223
GPU NVIDIA GeForce RTX 2080Ti × 1
Operating system Ubuntu 18.04
Memory 32 GB
Framework PyTorch v1.9.1
Library CUDA 10.2、CUDNN 7.6.5

Table 2. Ablation experiment

View table

Table 2. Ablation experiment

Method	IoU of target /%	FPS /（frame·s^-1）	FLOPs /10⁹	Param /10⁶
Baseline	60.9	107	15.38	14.76
+ Improved Feature Extraction	61.06	158	14.88	3.93
+ Improved Feature Fusion	68.32	118	11.41	3.04
+ Improved DiceLoss	68.63	117	11.41	3.04

Table 3. Comparison of parameters and performance of different networks
View table
Table 3. Comparison of parameters and performance of different networks
Network FLOPs /10⁹ Param /10⁶ Inference speed /（frame·s^-1）
FCN 247.13 49.49 81
ICNet 19.27 47.82 97
BiSeNet V2 15.38 14.76 105
STDCNet 10.57 8.57 103
TopFormer 0.62 1.37 121
RTIRSeg 11.41 3.04 117

Table 4. Evaluation of segmentation results of different methods

View table

Table 4. Evaluation of segmentation results of different methods

Network	Class	Acc	mAcc	Dice	mDice	IoU	mIoU
FCN	Background	99.99	81.69	99.99	84.86	99.99	76.76
FCN	Target	63.38	81.69	69.73	84.86	53.52	76.76
ICNet	Background	99.99	91.45	99.99	87.84	99.99	80.44
ICNet	Target	82.90	91.45	75.69	87.84	60.89	80.44
BiSeNet V2	Background	99.99	90.55	99.99	87.85	99.99	80.45
BiSeNet V2	Target	81.11	90.55	75.7	87.85	60.9	80.45
STDCNet	Background	99.99	80.09	99.99	83.13	99.99	74.77
STDCNet	Target	60.19	80.09	66.26	83.13	49.54	74.77
TopFormer	Background	99.99	78.60	99.99	81.29	99.99	72.77
TopFormer	Target	57.21	78.60	62.59	81.29	45.55	72.77
RTIRSeg	Background	99.99	90.42	99.99	90.58	99.99	84.15
RTIRSeg	Target	80.85	90.42	81.17	90.58	68.30	84.15

Tools

Get Citation

Copy Citation Text

Bin Shao, Hua Yang, Bin Zhu, Yi Chen, Rongping Zou. Infrared Small Target Detection Algorithm Based on Real-Time Semantic Segmentation[J]. Laser & Optoelectronics Progress, 2023, 60(14): 1410006

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites