cs.CVOct 13, 2026
SaveTraditional and Recent Approaches for Infrared Small Target Detection : Overview, Challenges and Perspectives (Part 2)
Organizations: Laboratoire MIA La Rochelle Université, La Rochelle, 17000, France · ISEN Toulon , Toulon, , France · Laboratoire L3i La Rochelle Université, La Rochelle, 17000, France · Department of Electrical Engineering IIT Jammu, Jagti, 181221, India
Abstract
Since the rise of supervised deep learning models, researchers have attempted to solve infrared small target detection using various models like convolutional neural networks, GANs, Transformers, Mixture of Experts and recent graph/hypergraph neural networks. Moreover, these deep neural network methods have surpassed the traditional unsupervised models and the low-rank and sparse decomposition unsupervised models reviewed in "Traditional and Recent Approaches for Infrared Small Target Detection : Overview, Challenges and Perspectives (Part 1)". In this context, the aim of this paper is to provide a complete survey of the deep learning models in infrared small target detection. First, we categorize the different approaches found in the literature in terms of their modeling strategies. Then, we discuss them in terms of the critical situations that they claim to handle. Furthermore, we present the specific performance evaluation protocols. Then, we conclude with several promising directions for future research.
Figures & tables
Figure 1: U-shape structures and a multi-scale nested interaction module. First row: UNet ( 34 ) and UNet++ ( 59 ) designed for biomedical images. Second row: DNA-Net ( 53 ) and iSmallNet ( 58 ) designed for infrared small target detection (Image from Hu et al. ( 58 ) ).
| Deep Neural Networks Methods | Acronyms (Authors) |
|---|---|
| Detection based Algorithms | CNN (YOLOv5) [Detection] (Zhou et al. ( 132 ) ), TCRNet (YOLOv3) [Detection] (Mc Intosh et al. ( 133 ) ) |
| (19 papers) | YOLO series: IRSTD-YOLO (Tang et al. ( 134 ) ), YOLOSR-IST (Li and Shen ( 135 ) ), YOLO-MBL (Shen et al. ( 136 ) ) |
| YOLO-TFFM (Xu et al. ( 137 ) ), SAYOLO (Guo et al. ( 138 ) ), BOH-YOLO (Zhu and Zhou ( 139 ) ) | |
| YOLO-LCF (Chen et al. ( 140 ) ), IR-WSANet (Jin et al. ( 141 ) ), HSFL (Wang et al. ( 142 ) ) | |
| YOLO-MSF (He et al. ( 143 ) ) | |
| DETR series: RT-DETR-CST (Duan et al. ( 144 ) ), SEF-DETR (Liu et al. ( 102 ) ), IRST-RTDETR (Chen et al. ( 145 ) ) |
Table 1: Deep Neural Networks Supervised Detection and Segmentation-based Methods in Infrared Small Target Detection: An Overview. For each category, only the key references are indicated. Please see the corresponding section for more references.
| Deep Neural Networks Methods | Acronyms (Authors) |
|---|---|
| Learning Mechanisms | Conventional Learning Mechanisms |
| Single point supervised learning ( 190 ; 111 ; 191 ; 110 ) , Weakly supervised learning ( 192 ; 193 ; 194 ) , Source-free domain adaptation learning ( 195 ; 196 ; 197 ) | |
| Contrastive Learning ( 90 ; 198 ) , Dual-Stream Edge-Target Learning ( 199 ) | |
| Data Augmentation ( 63 ; 200 ; 64 ; 69 ; 71 ; 65 ; 68 ) | |
| Attention Mechanisms | Conventional Attention Mechanisms |
| Pyramid Attention Mechanisms, Hierarchical Attention Mechanisms ( 201 ) , Spatial Attention Mechanisms ( 202 ; 203 ) |
Table 2: Deep Neural Networks Supervised Segmentation-based Methods in Infrared Small Target Detection: An Overview. For each category, only the key references are indicated. Please see the corresponding section for more references.
| Methods | Acronyms |
|---|---|
| Conventional Learning Mechanisms | MTMLNet [Single-Frame] (Yang et al. ( 274 ) ), MTMLNet+ [Single-Frame] (Luo et al. ( 290 ) ), TDCNet [Multi-Frame][MIRST] (Fang et al. ( 275 ) ) |
| LODDIS [Multi-Frame][MIRST] (Chen et al. ( 291 ) ) | |
| GGL-Net (Zhao et al. ( 292 ) ), S Net (Hu et al. ( 293 ) ) | |
| Single Point Supervised Learning | PAL (Yu et al. ( 110 ) ), LELCM (Yang et al. ( 191 ) ) |
| [Plug and Play] | LESPS (2023) [Single-Frame] (Ying et al. ( 111 ) ), LESPS (2024) (Shao et al. ( 79 ) ) |
| GMVG-DIR (Shi et al. ( 190 ) ), GSACP (Zhou and Zhang ( 294 ) ), EDGSP (Huan et al. ( 295 ) ) |
Table 3: Learning Mechanisms: An Overview (44 papers).
| Methods | Acronyms |
|---|---|
| Conventional Attention Mechanisms | ALCNet [Single-Frame] (Dai et al. ( 316 ) ), DNA-Net [Single-Frame][Dense] (Li et al. ( 53 ) ), DAEA (Lin et al. ( 324 ) ), DCANet (Zang et al. ( 54 ) ), STPSA-Net (Liu et al. ( 325 ) ), STAM-UNET (Shahraki et al. ( 326 ) ), FAA-Net (Zhuang et al. ( 327 ) ), DENet (Zuo et al. ( 328 ) ), DPS-Net (Liu and Huang ( 329 ) ), NCENet (Wu et al. ( 330 ) ), DCGANet (Chen et al. ( 331 ) ), CSQNet (Yuan et al. ( 332 ) ), CARG (Zhang et al. ( 333 ) ), C2IRSTD (Wang et al. ( 334 ) ), SSCFNet (Wang et al. ( 123 ) ), PHFNet (Li et al. ( 335 ) ), LCAE-Net (Wang et al. ( 336 ) ), CS-LDRNet (Lei et al. ( 337 ) ), SWA-HFFM (Miao et al. ( 338 ) ), CCRANet [Single-Frame] (Wang et al. ( 339 ) ), BDSNet (Wang et al. ( 340 ) ), BCEANet (Li et al. ( 341 ) ) Gabor Modulation: GMN (Bian et al. ( 342 ) ) Adaptive: A2-STMC (Wu et al. ( 343 ) ) Dual-Attention: MCC-Net (Zhou et al. ( 344 ) ) Curvature: SCMT-Net (Yang et al. ( 124 ) ), CHFNet (Zhang et al. ( 345 ) ) Multi-Attention: MFAM (Li et al. ( 346 ) ) |
| Top-K Attention Mechanisms | LGTS (Huang et al. ( 347 ) ), TriHANet (Li et al. ( 348 ) ), HPFNet (Zhang et al. ( 117 ) ), DifRankNet (Fang et al. ( 349 ) ) |
| Pyramid Attention Mechanisms | FPN (Tong et al. ( 350 ) ), NS-FPN [Detection] (Yuan et al. ( 351 ) ), FE-FPN (Wang et al. ( 352 ) ), PAU2-Net (Peng et al. ( 44 ) ), MCFD (Zong and Shan ( 353 ) ), Nested FPN (Lu et al. ( 354 ) ), ISPANet (Wang et al. ( 355 ) ), MAPFF (Yang et al. ( 356 ) ) |
| Hierarchical Attention Mechanisms | HAFNet (Zhang et al. ( 201 ) ) |
| Spatial Attention Mechanisms | SANet (2023) (Zhu et al. ( 202 ) ), DSAD (Li et al. ( 203 ) ), CSAN-UNet (Zhong et al. ( 357 ) ), CenterADNet (Sun et al. ( 358 ) ), DSDF-Net (Dong et al. ( 359 ) ), SaKD (Ma et al. ( 360 ) ) |
| Direction-Induced Attention Mechanisms | RDIAN [Single-Frame] (Sun et al. ( 204 ) ), LDCNet (Tang et al. ( 205 ) ), ISNet (Zhang et al. ( 74 ) ), P2P-HDNet (Ni et al. ( 361 ) ), EDDNet (Ji et al. ( 362 ) ), DTUM (Li et al. ( 363 ) ), DCONet (Zhang et al. ( 364 ) ) |
Table 4: Attention Mechanisms: An Overview (81 papers).
| Methods | Acronyms |
|---|---|
| Adaptive Mechanims | TAMNet (Wang et al. ( 208 ) ), MDAFNet [Dense] (Li et al. ( 390 ) ), AER-net (Zhang et al. ( 45 ) ) |
| CEFANet (Fang et al. ( 381 ) ), AMID (Yang et al. ( 391 ) ), ISTD-APSNet (Li et al. ( 126 ) ) | |
| Recurrent Mechanims | RRCANet (Liu et al. ( 83 ) ) |
| Diffusion Mechanisms | EDPrior (DPNet) (Tong et al. ( 209 ) ), APCD (Rong et al. ( 210 ) ), DCFR-Net (Fan et al. ( 211 ) ) |
| DCDNet (Li et al. ( 392 ) ) | |
| Multi-Branch Mechanisms | DBCE (U-Net) (Xiao et al. ( 39 ) ), MMLNet (Li et al. ( 115 ) ), AFDMNet (Wang et al. ( 393 ) ) |
Table 5: Adaptive and Multi-Cues Mechanisms: An Overview (79 papers).
| Methods | Acronyms |
|---|---|
| Causal Guidance | CHAL (Duan et al. ( 442 ) ) |
| Prior Guidance | PSGF (Zhang et al. ( 212 ) ), BPGA (Lin et al. ( 443 ) ), ECFNet (Fang et al. ( 444 ) ), MEP-CSNet (Wang et al. ( 445 ) ) |
| RHNet (Fang et al. ( 446 ) ) | |
| Dual Prior: DPMSNet (Wang et al. ( 447 ) ), DPC (Yu et al. ( 448 ) ) | |
| Gaussian Prior: GPConv [Detection] (Liu et al. ( 449 ) ), PG-C2F (Cui et al. ( 450 ) ), GA-Conv (Wei et al. ( 451 ) ) | |
| Segmentation Prior: MGA-Net (Ding and Wu ( 452 ) ), MISA [MIRST] (Wu et al. ( 453 ) ) |
Table 6: Guidance Mechanisms: An Overview (119 papers).
| Methods | Acronyms |
|---|---|
| Robust Algorithms | RISTDnet (Hou et al. ( 225 ) ), MISTNet , SANet (2026) (Zhang et al. ( 545 ) ) |
| Refining False Detections Algorithms | RefineFD (Du et al. ( 546 ) ), RefineDet (Xu et al. ( 547 ) ) |
| Sparsity Algorithms | SAGCP (SAGCPDet) (Wu et al. ( 227 ) ), ACSNet (Wu et al. ( 228 ) ) |
| Aliasing Rectification Algorithms | FGARNet (Peng et al. ( 229 ) , EAA-Net (Zhang et al. ( 230 ) ) |
| Low-illumination Algorithms | Nv-Net (Chen et al. ( 81 ) ), CFD-Net (Zhang et al. ( 231 ) , FDI-Net (Chen et al. ( 232 ) ) |
| Lightweight Algorithms | Faster R-CNN [Detection] (Bai et al. ( 233 ) ), TBC-Net [Single-Frame] (Zhao et al. ( 234 ) ), TY-RIST [Detection] (Zhao et al. ( 548 ) ), IADF-YOLO (Sun et al. ( 549 ) ) |
Table 7: Robust and Lightweight Algorithms: An Overview (31 papers).
| Methods | Acronyms |
|---|---|
| Wavelets | HaarConv (YOLOv5-s) [Detection] (Song et al. ( 284 ) ), HaarTransNet (Fan et al. ( 285 ) ) |
| IRPruneDet (Zhang et al. ( 322 ) ) | |
| WMRNet (Zhang et al. ( 566 ) ), FSCNet (Li et al. ( 567 ) ), WCDMF-Net (Li et al. ( 80 ) ) | |
| DFAwareNet (Cheng et al. ( 568 ) ), SDANet (UNet) (Zhang et al. ( 569 ) ) | |
| WaveTD (Ma et al. ( 570 ) ), WTAPNet (He et al. ( 571 ) ), SWAN (Jing et al. ( 121 ) ) | |
| WRSSNet (Jiang et al. ( 572 ) ), Ivan-ISTD (Li et al. ( 4 ) ), IRPruneDeXt (Zhang et al. ( 573 ) ) |
Table 8: Wavelets Methods: An Overview (24 papers).
| Methods | Acronyms |
|---|---|
| Mamba State Space Model (SSM)-based Neural Networks | HMCNet (Li et al. ( 87 ) ), LVMG (Lu and Zhang ( 10 ) ), IRMamba (Zhang et al. ( 5 ) ) |
| EAMNet (Jiang et al. ( 587 ) ), Mamba-IRSTD (Zhang et al. ( 588 ) ), MRNet (Ye et al. ( 589 ) ) | |
| CPMamba (Wu et al. ( 129 ) ), BDSFusion (Li et al. ( 99 ) ), MiM-ISTD [Single-Frame] (Chen et al. ( 590 ) ) | |
| STLMamba [MIRST] (Qiu et al. ( 591 ) ), TAPM-Net [MIRST] (Xie et al. ( 592 ) ), MOU-Mamba (Liu et al. ( 11 ) ) | |
| SMILE (Li et al. ( 593 ) ), MRL-Mamba (Li et al. ( 594 ) ), SBMambaNet (Yu et al. ( 595 ) ) | |
| SAMamba (Xu et al. ( 283 ) ), FANM [Dense] (Xu et al. ( 77 ) ) |
Table 9: Mamba State Space Model (SSM)-based Neural Networks and Transformers-based Neural Networks: An Overview (67 papers).
Figure 2: Comparison between Performance and the Number of Parameters for LW-DSTransNet compared to early SOTA Data-driven Algorithms (Image from Huang et al. ( 626 ) ).
Figure 3: Impact of Plug and Play Single Point Supervised Learning Modules on SOTA Data-driven Algorithms (Image from Yu et al. ( 110 ) ).
Figure 4: Impact of Plug and Play Single Point Supervised Learning Modules on SOTA Data-driven Algorithms (Image from Liu et al. ( 113 ) ).
More in this topic
Infrared Small Target Detection
Infrared small target detection finds faint, often sub-pixel objects against cluttered thermal backgrounds while limiting false alarms.
See all 43 papers