Infrared Tiny Object Detection Using Semi-Supervised Learning and Data Augmentation

Bin Zhang; Liangshun Wu; Yuguo Wang; Ling Peng; Juan Hu; Dawei Jiang

doi:10.1142/S0218001425510322

Infrared Tiny Object Detection Using Semi-Supervised Learning and Data Augmentation

Bin Zhang
, Liangshun Wu^*
, Yuguo Wang
, Ling Peng
, Juan Hu
, Dawei Jiang

^*Corresponding author for this work

Department of Computer Science

Research output: Journal Publications and Reviews › RGC 21 - Publication in refereed journal › peer-review

Abstract

Aiming at the core challenges of data shortage and high labeling cost in the infrared small target detection task, this paper puts forward an innovative method of integrating generative adversarial network and semi-supervised learning. By designing the small target generative adversarial network (STGAN) based on StyleGAN2, combining the self-attention mechanism and the comparative learning loss, the quality and diversity of generated data are effectively improved. At the same time, the knowledge distillation framework is used to optimize the student model by using the pseudo-tags generated by the teacher model, which significantly improves the detection performance in small sample scenes. The experimental results show that the nIoU (1% labeled data) of 0.709 is achieved by the proposed method on SIRST data set, and the FID index is reduced to 0.45, both reaching the current optimal level. The ablation experiment further verified the key roles of detection loss, self-attention module and REP-PAN feature fusion strategy, and provided a new idea for data expansion and model optimization in the field of small target detection. © 2026 World Scientific Publishing Company.

Original language	English
Article number	2551032
Number of pages	21
Journal	International Journal of Pattern Recognition and Artificial Intelligence
Volume	40
Issue number	2
Online published	13 Dec 2025
DOIs	https://doi.org/10.1142/S0218001425510322
Publication status	Published - Feb 2026

Funding

This study was supported in part by a grant from the President's Fund of Xinjiang University of Political Science and Law-“Research on Small Target Detection in Large-Scale Scenes of Nanjiang” (Grant No. XZZK2022002), China Higher Education Institution Industry-University-Research Innovation Fund-Joint Special Project (Phase VI:Yaocheng Digital-Intelligence Education Program)-A Real-Time Early Warning and Precision Profiling Evaluation Mechanism for Practice-Based Teaching Grounded in a Digital Instructional Management System.2025 Shandong Province Youth Natural Science Research Project-“Research on Key Technologies for Small Target Detection in Complex Scenarios” (Grant No. WLZR25001), 2025 Shandong Province Basic and Applied Basic Research Project-“Research on Small Target Detection in Large-scale Scenarios” (Grant No. WL-JC25008), 2025 Shandong Province Project on Artificial Intelligence in Teaching and Education Applications-Exploration and Practice of Industry-“Education Integration Mechanism Under the Background of AI + Education”(Project No. WL-AIJ2504003), Special Project for AI-Enabled Higher Education Research in 2025–“Commercial Value Transformation Pathways of AI-Enabled Educational Application Scenarios from the Perspective of Industry-Education Integration” (Project No. WL-AIGJ25002) and “AI-Enabled Communication Instruction for Children with Autism Spectrum Disorder” (Project No. WL-AIGJ25012). and the 2025 Open Project of Key Laboratory of Computational Neuroscience and Brain-Inspired Intelligence (Fudan University), Ministry of Education: "Brain-Inspired Federated Learning Methods for Edge Intelligence", and the Open Fund of Guizhou Key Laboratory of Artificial Intelligence and Brain-inspired Computing(Grant No. AILKF20250103).At the same time, The authors extend their sincere gratitude to the "Intelligent Perception and Decision Consulting Expert Team" of the China Society of Image and Graphics for their invaluable support, as well as to the Hong Kong Institute of Technology's Taiping (Young) Scholars Programme.

Research Keywords

GAN
infrared image
knowledge distillation
small sample
small target detection

Access Output

10.1142/S0218001425510322

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

Cite this

@article{5946b5b972dc44d3a948a1f57aa3bc00,

title = "Infrared Tiny Object Detection Using Semi-Supervised Learning and Data Augmentation",

abstract = "Aiming at the core challenges of data shortage and high labeling cost in the infrared small target detection task, this paper puts forward an innovative method of integrating generative adversarial network and semi-supervised learning. By designing the small target generative adversarial network (STGAN) based on StyleGAN2, combining the self-attention mechanism and the comparative learning loss, the quality and diversity of generated data are effectively improved. At the same time, the knowledge distillation framework is used to optimize the student model by using the pseudo-tags generated by the teacher model, which significantly improves the detection performance in small sample scenes. The experimental results show that the nIoU (1\% labeled data) of 0.709 is achieved by the proposed method on SIRST data set, and the FID index is reduced to 0.45, both reaching the current optimal level. The ablation experiment further verified the key roles of detection loss, self-attention module and REP-PAN feature fusion strategy, and provided a new idea for data expansion and model optimization in the field of small target detection. {\textcopyright} 2026 World Scientific Publishing Company.",

keywords = "GAN, infrared image, knowledge distillation, small sample, small target detection",

author = "Bin Zhang and Liangshun Wu and Yuguo Wang and Ling Peng and Juan Hu and Dawei Jiang",

year = "2026",

month = feb,

doi = "10.1142/S0218001425510322",

language = "English",

volume = "40",

journal = "International Journal of Pattern Recognition and Artificial Intelligence",

issn = "0218-0014",

publisher = "World Scientific Publishing Co. Pte Ltd",

number = "2",

}

TY - JOUR

T1 - Infrared Tiny Object Detection Using Semi-Supervised Learning and Data Augmentation

AU - Zhang, Bin

AU - Wu, Liangshun

AU - Wang, Yuguo

AU - Peng, Ling

AU - Hu, Juan

AU - Jiang, Dawei

PY - 2026/2

Y1 - 2026/2

N2 - Aiming at the core challenges of data shortage and high labeling cost in the infrared small target detection task, this paper puts forward an innovative method of integrating generative adversarial network and semi-supervised learning. By designing the small target generative adversarial network (STGAN) based on StyleGAN2, combining the self-attention mechanism and the comparative learning loss, the quality and diversity of generated data are effectively improved. At the same time, the knowledge distillation framework is used to optimize the student model by using the pseudo-tags generated by the teacher model, which significantly improves the detection performance in small sample scenes. The experimental results show that the nIoU (1% labeled data) of 0.709 is achieved by the proposed method on SIRST data set, and the FID index is reduced to 0.45, both reaching the current optimal level. The ablation experiment further verified the key roles of detection loss, self-attention module and REP-PAN feature fusion strategy, and provided a new idea for data expansion and model optimization in the field of small target detection. © 2026 World Scientific Publishing Company.

AB - Aiming at the core challenges of data shortage and high labeling cost in the infrared small target detection task, this paper puts forward an innovative method of integrating generative adversarial network and semi-supervised learning. By designing the small target generative adversarial network (STGAN) based on StyleGAN2, combining the self-attention mechanism and the comparative learning loss, the quality and diversity of generated data are effectively improved. At the same time, the knowledge distillation framework is used to optimize the student model by using the pseudo-tags generated by the teacher model, which significantly improves the detection performance in small sample scenes. The experimental results show that the nIoU (1% labeled data) of 0.709 is achieved by the proposed method on SIRST data set, and the FID index is reduced to 0.45, both reaching the current optimal level. The ablation experiment further verified the key roles of detection loss, self-attention module and REP-PAN feature fusion strategy, and provided a new idea for data expansion and model optimization in the field of small target detection. © 2026 World Scientific Publishing Company.

KW - GAN

KW - infrared image

KW - knowledge distillation

KW - small sample

KW - small target detection

UR - https://www.scopus.com/pages/publications/105024867672

UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-105024867672&origin=recordpage

U2 - 10.1142/S0218001425510322

DO - 10.1142/S0218001425510322

M3 - RGC 21 - Publication in refereed journal

SN - 0218-0014

VL - 40

JO - International Journal of Pattern Recognition and Artificial Intelligence

JF - International Journal of Pattern Recognition and Artificial Intelligence

IS - 2

M1 - 2551032

ER -

Infrared Tiny Object Detection Using Semi-Supervised Learning and Data Augmentation

Abstract

Funding

Research Keywords

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

Cite this