Infrared Tiny Target Detection Method Based on a Multi-Hop Deep Network

Shule Yan; Runyu Chen; Nian Cai; Shaoqiu Xu; Jian Chen

doi:10.3788/LOP240610

Journals >Laser & Optoelectronics Progress >Volume 61 >Issue 22 >Page 2237008 > Article

Laser & Optoelectronics Progress
Vol. 61, Issue 22, 2237008 (2024)

Infrared Tiny Target Detection Method Based on a Multi-Hop Deep Network

Shule Yan¹, Runyu Chen¹, Nian Cai^1,*, Shaoqiu Xu¹, and Jian Chen²

Author Affiliations

¹School of Information Engineering, Guangdong University of Technology, Guangzhou 510006, Guangdong , China

²Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, Jilin , China

show less

DOI: 10.3788/LOP240610 Cite this Article Set citation alerts

Shule Yan, Runyu Chen, Nian Cai, Shaoqiu Xu, Jian Chen. Infrared Tiny Target Detection Method Based on a Multi-Hop Deep Network[J]. Laser & Optoelectronics Progress, 2024, 61(22): 2237008 Copy Citation Text

show less

Imaging characteristics of infrared targets in complex background. (a) Target imaging performance; (b) grayscale distribution of infrared images; (c) top view of grayscale distribution

Fig. 1. Imaging characteristics of infrared targets in complex background. (a) Target imaging performance; (b) grayscale distribution of infrared images; (c) top view of grayscale distribution

Download full size

Fig. 2. Overall framework of proposed method

Download full size

Fig. 3. Structure of MMFB and class activation mapping diagram of corresponding branches

Download full size

Fig. 4. ROC curves for different compared methods

Download full size

Fig. 5. Visualization of detection results

Download full size

Fig. 6. Visualization of infrared targets on IRDST dataset

Download full size

Method	Target	Recall /%	Precision /%	F1-score /%	$F_{a}$ /%	FNR /%	Time /ms
MDFA-CGAN	Tiny	56.8	52.4	54.5	41.6	43.2	27.1
MDFA-CGAN	Small	71.1	45.4	55.4	16.5	28.9	27.1
ACM	Tiny	19.9	26.7	22.8	44.1	80.1	30.3
ACM	Small	54.9	57.4	56.1	7.8	45.1	30.3
DNANet	Tiny	69.9	73.6	71.7	20.2	30.1	31.3
DNANet	Small	86.8	86.7	86.8	2.6	13.2	31.3
YOLO-FR	Tiny	92.1	60.6	73.1	48.3	7.9	6.4
YOLO-FR	Small	90.6	56.3	69.5	13.5	9.4	6.4
Proposed method	Tiny	97.0	94.5	95.7	4.5	3.0	11.3
Proposed method	Small	91.1	96.4	93.7	0.7	8.9	11.3

Table 1. Comparisons of different methods on infrared target detection

Baseline	MMFB	NWD	Target	Recall	Precision	F1-score	$F_{a}$	FNR
√			Tiny	95.1	79.0	86.3	20.4	4.9
√			Small	46.5	99.6	63.4	19.5	53.5
√	√		Tiny	95.9	84.7	90.0	4.2	4.1
√	√		Small	89.4	89.7	89.6	2.0	10.6
√		√	Tiny	92.0	90.4	91.2	5.3	8.0
√		√	Small	93.5	82.6	87.7	3.8	6.5
√	√	√	Tiny	97.0	94.5	95.7	4.5	3.0
√	√	√	Small	91.1	96.4	93.7	0.7	8.9

Table 2. Ablation experiments

Method	Recall	Precision	F1-score	$F_{a}$	FNR
MDFA-CGAN	41.6	44.8	43.1	55.2	58.4
ACM	35.5	35.9	35.7	64.1	64.5
DNANet	89.2	90.5	89.9	9.5	10.8
YOLO-FR	83.2	58.5	68.7	41.5	16.8
Proposed method	93.6	95.7	94.6	4.3	6.4

Table 3. Comparison on IRDST dataset

Shule Yan, Runyu Chen, Nian Cai, Shaoqiu Xu, Jian Chen. Infrared Tiny Target Detection Method Based on a Multi-Hop Deep Network[J]. Laser & Optoelectronics Progress, 2024, 61(22): 2237008

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享