IDDNet: Infrared Object Detection Network Based on Multi-Scale Fusion Dehazing DOI Creative Commons

Shizun Sun,

Shuo Han, Junwei Xu

et al.

Sensors, Journal Year: 2025, Volume and Issue: 25(7), P. 2169 - 2169

Published: March 29, 2025

In foggy environments, infrared images suffer from reduced contrast, degraded details, and blurred objects, which impair detection accuracy real-time performance. To tackle these issues, we propose IDDNet, a lightweight object network that integrates multi-scale fusion dehazing. IDDNet includes dehazing (MSFD) module, uses feature to eliminate haze interference while preserving key details. A dedicated loss function, DhLoss, further improves the effect. addition MSFD, incorporates three main components: (1) bidirectional polarized self-attention, (2) weighted pyramid network, (3) layers. This architecture ensures high computational efficiency. two-stage training strategy optimizes model's performance, enhancing its robustness in environments. Extensive experiments on public datasets demonstrate achieves 89.4% precision 83.9% AP, showing superior accuracy, processing speed, generalization, robust

Language: Английский

CSMR: A Multi-Modal Registered Dataset for Complex Scenarios DOI Creative Commons

C.J. Li,

Kun Gao, Zibo Hu

et al.

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(5), P. 844 - 844

Published: Feb. 27, 2025

Complex scenarios pose challenges to tasks in computer vision, including image fusion, object detection, and image-to-image translation. On the one hand, complex involve fluctuating weather or lighting conditions, where even images of same appear be different. other large amount textural detail given introduces considerable interference that can conceal useful information contained them. An effective solution these problems is use complementary details present multi-modal images, such as visible-light infrared images. Visible-light contain rich while about temperature. In this study, we propose a registered dataset for under various environmental targeting security surveillance monitoring low-slow-small targets. Our contains 30,819 targets are labeled three classes “person”, “car”, “drone” using Yolo format bounding boxes. We compared our with those used literature vision-related tasks, The results showed introducing through fusion compensate missing original also revealed limitations visual single-modal scenarios.

Language: Английский

Citations

0
IDDNet: Infrared Object Detection Network Based on Multi-Scale Fusion Dehazing DOI Creative Commons

Shizun Sun,

Shuo Han, Junwei Xu

et al.

Sensors, Journal Year: 2025, Volume and Issue: 25(7), P. 2169 - 2169

Published: March 29, 2025

In foggy environments, infrared images suffer from reduced contrast, degraded details, and blurred objects, which impair detection accuracy real-time performance. To tackle these issues, we propose IDDNet, a lightweight object network that integrates multi-scale fusion dehazing. IDDNet includes dehazing (MSFD) module, uses feature to eliminate haze interference while preserving key details. A dedicated loss function, DhLoss, further improves the effect. addition MSFD, incorporates three main components: (1) bidirectional polarized self-attention, (2) weighted pyramid network, (3) layers. This architecture ensures high computational efficiency. two-stage training strategy optimizes model's performance, enhancing its robustness in environments. Extensive experiments on public datasets demonstrate achieves 89.4% precision 83.9% AP, showing superior accuracy, processing speed, generalization, robust

Language: Английский

Citations

0