Infrared Small Target Detection via Multi⁃Layer Convolution Fusion(Invited)

Peng Zhang; Lifen Shi; Ziyang Chen; Jixiong Pu

doi:10.3788/LOP241267

Laser & Optoelectronics Progress, Volume. 61, Issue 16, 1611014(2024)

Infrared Small Target Detection via Multi⁃Layer Convolution Fusion(Invited)

Peng Zhang¹, Lifen Shi², Ziyang Chen^1、*, and Jixiong Pu¹

¹Fujian Key Laboratory of Light Propagation and Transformation,College of Information Science and Engineering, Huaqiao University, Xiamen 361021, Fujian, China

²Officers College of PAP, Chengdu 610000, Sichuan, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Overall architecture of the model

Download full size

Fig. 2. Multi-layer convolutional fusion module

Download full size

Fig. 3. Node process

Download full size

Fig. 4. Multi-receptive field fusion module diagram

Download full size

Fig. 5. Infrared images and corresponding 3D images. (a) Original infrared image of two targets; (b) 3D image of infrared image of two targets; (c) original infrared image of three targets; (d) 3D image of infrared image of three targets

Download full size

Fig. 6. Grayscale distribution of detection results using different algorithms

Download full size

Fig. 7. Comparison of test original images, label images, and algorithm detection images

Download full size

Fig. 8. Ablation experiments of multi-layer convolutional fusion module. (a) Structure diagram with one less set of vertical node; (b) structure diagram with one more set of vertical node

Download full size

Table 1. Algorithm parameters

View table

Table 1. Algorithm parameters

Algorithm	Parameter
Top-Hat	Filter size：5×5
RLCM	Scale quantity：3，k₁=［2，5，9］，k₂=［4，9，16］
CDD	Target box scale：3×3-9×9，background Box scale：19×19
PSTNN	Block size：40×40，sliding step：40

Table 2. Average indicators of different methods

View table

Table 2. Average indicators of different methods

Method	False	Recall	Precision	F1	IOU_sample	IOU_pixel
Top-Hat	0.00060	0.4146	0.1095	0.1497	0.08202	0.04236
RLCM	0.00040	0.1771	0.0235	0.0414	0.02314	0.01369
CDD	0.00002	0.0812	0.8823	0.1493	0.08158	0.07327
FKRW	0.00007	0.1686	0.3480	0.2143	0.15639	0.10364
PSTNN	0.00009	0.8581	0.5692	0.6748	0.32021	0.28156
IAANet	0.00001	0.1736	0.9134	0.2851	0.18509	0.15971
ALCNet	0.00004	0.8117	0.7811	0.7952	0.65576	0.55878
ACM	0.00004	0.7502	0.7974	0.6950	0.57766	0.53588
Ours	0.00001	0.9409	0.9383	0.9385	0.84757	0.81431

Table 3. Ablation experiments of multi-layer convolutional fusion module

View table

Table 3. Ablation experiments of multi-layer convolutional fusion module

Algorithm	IOU_sample	IOU_pixel
Multi-layer convolutional fusion module minus a set of vertical node structure diagrams	0.61356	0.56931
Add a set of vertical nodes to the multi-layer convolutional fusion module	0.64562	0.60136
Ours	0.84757	0.81431

Table 4. Ablation experiments of different receptive fields

View table

Table 4. Ablation experiments of different receptive fields

Algorithm	IOU_sample	IOU_pixel
The combination of receptive fields with convolution kernel sizes of 1，3，and 5	0.80961	0.75496
The combination of receptive fields with convolution kernel sizes of 5，7，and 9	0.65137	0.59427
Ours（receptive fields with convolution kernel sizes of 3，5，and 7）	0.84757	0.81431

Table 5. Ablation experiments
View table
Table 5. Ablation experiments
Algorithm IOU_sample IOU_pixel
No multi-receptive field fusion module 0.70136 0.66370
No multi-layer convolutional fusion module 0.35641 0.30371
No CBAM 0.75253 0.72584
Ours 0.84757 0.81431

Tools

Get Citation

Copy Citation Text

Peng Zhang, Lifen Shi, Ziyang Chen, Jixiong Pu. Infrared Small Target Detection via Multi⁃Layer Convolution Fusion(Invited)[J]. Laser & Optoelectronics Progress, 2024, 61(16): 1611014

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites