Method for Fast Detection of Infrared Targets Based on Key Points

Zhuang Miao; Yong Zhang; Ruimin Chen; Weihua Li

doi:10.3788/AOS202040.2312006

Acta Optica Sinica, Volume. 40, Issue 23, 2312006(2020)

Method for Fast Detection of Infrared Targets Based on Key Points

Zhuang Miao^1,2, Yong Zhang^1、*, Ruimin Chen^1,2, and Weihua Li^1,2

¹Key Laboratory of Infrared System Detection and Imaging Technology, Shanghai Institute of Technical Physics, Chinese Academy of Sciences, Shanghai 200083, China

²School of Electronic, Electrical, and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Architecture of FKPD model

Download full size

Fig. 2. Feature extraction network

Download full size

Fig. 3. Feature fusion network

Download full size

Fig. 4. Target prediction network

Download full size

Fig. 5. Comparison of infrared target detection results under different models. (a) Labeled image; (b) CenterNet-Res18; (c) YOLOv3-Darknet53; (d) Tiny-YOLOv3; (e) FKPD-384

Download full size

Fig. 6. Examples for FKPD infrared target detection

Download full size

Fig. 7. Comparison of detection effects on PASCAL VOC dataset

Download full size

Table 1. Training parameters used for FKPD model
View table
Table 1. Training parameters used for FKPD model
Parameter Infrared dataset PASCAL VOC dataset
Training epoch 150 150
Class number 5 20
Batch size 32 64
Default image size 384 pixel×384 pixel 384 pixel×384 pixel
Initial learning rate 1.25×10^-3 1.25×10^-3

Table 2. Detection results on infrared dataset

View table

Table 2. Detection results on infrared dataset

Model	mAP /%	AP /%
Model	mAP /%	Bird	Fighter	Airliner	Helicopter	Trainer
CenterNet-Res18	88.04	76.73	88.95	94.91	90.77	88.84
YOLOv3-Darknet53	93.02	87.70	93.97	95.97	94.84	92.66
Tiny-YOLOv3	80.08	66.58	83.16	93.85	84.92	71.90
FKPD-384	88.98	79.40	90.84	95.01	90.27	89.39

Table 3. False alarm rate of model
View table
Table 3. False alarm rate of model
Model FAR /%
CenterNet-Res18 1.23
YOLOv3-Darknet53 1.12
Tiny-YOLOv3 8.96
FKPD-384 1.24

Table 4. Detection results on PASCAL VOC dataset
View table
Table 4. Detection results on PASCAL VOC dataset
Model mAP /%
CenterNet-Res18 68.24
YOLOv3-Darknet53 76.80
Tiny-YOLOv3 58.40
FKPD-384 61.61

Table 5. Real-time analysis of FKPD
View table
Table 5. Real-time analysis of FKPD
Model Model size /MB FLOPs /10¹⁰ CPU inference time /ms
CenterNet-Res18 56.86 8.69 297.05
YOLOv3-Darknet53 246.18 27.93 844.29
Tiny-YOLOv3 34.72 2.34 129.08
FKPD-384 8.12 1.55 115.18

Table 6. Ablation experiment of FKPD

View table

Table 6. Ablation experiment of FKPD

Method	mAP /%	Model size /MB	FLOPs /10¹⁰	CPU inference time /ms
FKPD-384(baseline)	88.98	8.12	1.55	115.18
FKPD-384 w/o FFN	84.34	8.12	1.55	114.89
FKPD-384 w Dilation	87.72	8.12	1.55	115.67
Compacted-FKPD-384	83.97	1.50	0.40	67.98
FKPD-512	90.15	8.12	2.76	174.29

Tools

Get Citation

Copy Citation Text

Zhuang Miao, Yong Zhang, Ruimin Chen, Weihua Li. Method for Fast Detection of Infrared Targets Based on Key Points[J]. Acta Optica Sinica, 2020, 40(23): 2312006

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites