PCLDet: Prototypical Contrastive Learning for Fine-Grained Object Detection in Remote Sensing Images

Ouyang, Lihan; Guo, Guangmiao; Fang, Leyuan; Ghamisi, Pedram; Yue, Jun

doi:10.1109/tgrs.2023.3290091

Cited by 7 publications

(2 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Otsuki et al [56] extend this concept to multi-modal language understanding tasks, learning data cluster prototypes from various domains to enable contrastive learning. Ouyang et al [57] propose PCLDet to learn contrastive-aware features for fine-grained object detection.…”

Section: E Contrastive Learning With Prototypesmentioning

confidence: 99%

Few-Shot Object Detection on Remote Sensing Images

Deng

Fang

2022

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

Object detection in remote sensing images relies on a large amount of labeled data for training. The growing new categories and class imbalance render exhaustive annotation nonscalable. Few-shot object detection (FSOD) tackles this issue by meta-learning on seen base classes and then fine-tuning on novel classes with few labeled samples. However, the object's scale and orientation variations are particularly large in remote sensing images, thus posing challenges to existing few-shot object detection methods. To tackle these challenges, we first propose to integrate a feature pyramid network and use prototype features to highlight query features to improve upon existing FSOD methods. We refer to the modified FSOD as a Strong Baseline which is demonstrated to perform significantly better than the original baselines. To improve the robustness of orientation variation, we further propose a transformation-invariant network (TINet) to allow the network to be invariant to geometric transformations. Extensive experiments on three widely used remote sensing object detection datasets, i.e., NWPU VHR-10.v2, DIOR, and HRRSD demonstrated the effectiveness of the proposed method. Finally, we reproduced multiple FSOD methods for remote sensing images to create an extensive benchmark for follow-up works.

show abstract

Section: E Contrastive Learning With Prototypesmentioning

confidence: 99%

Few-Shot Object Detection on Remote Sensing Images

Deng

Fang

2022

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

show abstract

“…Similarly, Zhang et al [12] proposed the I2MDet model, leveraging frequency and semantic image information to improve detection accuracy. Li et al [13] employed prototype contrast learning for fine-grained vehicle detection. Feng et al [14] designed SDANet to enhance vehicle detection in complex settings.…”

Section: Introductionmentioning

confidence: 99%

SatDetX-YOLO: A More Accurate Method for Vehicle Target Detection in Satellite Remote Sensing Imagery

Zhao,

Guo,

Shao

et al. 2024

IEEE Access

View full text Add to dashboard Cite

Satellite remote sensing technology significantly contributes to intelligent transportation by optimizing traffic planning via global perspectives and rich data, enhancing traffic efficiency and reducing environmental impact. However, current target detection models frequently exhibit low accuracy in vehicle detection tasks due to complex background interference in satellite imageries and a need for critical semantic information. To improve vehicle target detection accuracy, this study introduces SatDetX-YOLO, a vehicle detection model for satellite remote sensing images based on YOLOv8. The model involves reconstructing the backbone network with FasterNet for enhanced feature extraction, a redesigned decoupled head for improved computational efficiency and complex data processing, and incorporating the Deformable Attention Module (DAM) to increase sensitivity to small targets and feature correlation capture. Employing the Maximum Probabilistic Distance IoU (MPDIoU) loss function enhances adaptability and generalization to diverse vehicle targets. Experimental results demonstrate that under comparable FPS, SatDetX-YOLO's Precision (P), Recall (R), and Mean Average Precision (mAP) improved by 3.5%, 3.3%, and 3.2%, respectively. Despite a minor reduction in FPS, the model significantly enhances detection accuracy, striking a balance between accuracy and speed.

show abstract