CEH-YOLO: A composite enhanced YOLO-based model for underwater object detection

Ecological Informatics, Journal Year: 2024, Volume and Issue: 82, P. 102758 - 102758

Published: Aug. 8, 2024

Advances in underwater recording and processing systems have highlighted the need for automated methods dedicated to accurate detection tracking of small objects imagery. However, unique characteristics optical images, including low contrast, color variations, presence objects, pose significant challenges. This paper presents CEH-YOLO, a variant YOLOv8, incorporating high-order deformable attention (HDA) module enhance spatial feature extraction interaction by prioritizing key areas within model. Additionally, enhanced pyramid pooling-fast (ESPPF) is integrated object attributes, such as texture, which particularly beneficial scenarios with or overlapping objects. The customized composite (CD) further improves accuracy inclusivity detection. Moreover, model uses WIoU v3 technique bounding box loss calculations, effectively addressing regression challenges related boxes under standard extreme conditions. experimental results show model's exceptional performance, achieving mean average precisions 88.4% 87.7% on DUO UTDAC2020 datasets, respectively. Notably, operates at rapid speed 156 FPS, fulfilling critical real-time needs. With concise size 4.4 M moderate computational complexity 11.6 GFLOPs, it highly suitable integration into systems.

Language: Английский

---

Unsupervised clustering optimization-based efficient attention in YOLO for underwater object detection

Artificial Intelligence Review, Journal Year: 2025, Volume and Issue: 58(7)

Published: April 23, 2025

Language: Английский

---

Multi-scale integration with semantic embedding and adaptive excitation transformer for underwater optical image enhancement

Optics & Laser Technology, Journal Year: 2025, Volume and Issue: 189, P. 112881 - 112881

Published: April 29, 2025

Language: Английский

---

SPMFormer: Simplified Physical Model-based transformer with cross-space loss for underwater image enhancement

Knowledge-Based Systems, Journal Year: 2025, Volume and Issue: unknown, P. 113694 - 113694

Published: May 1, 2025

Language: Английский

---

YOLO-GE: An Attention Fusion Enhanced Underwater Object Detection Algorithm

Journal of Marine Science and Engineering, Journal Year: 2024, Volume and Issue: 12(10), P. 1885 - 1885

Published: Oct. 21, 2024

Underwater object detection is a challenging task with profound implications for fields such as aquaculture, marine ecological protection, and maritime rescue operations. The presence of numerous small aquatic organisms in the underwater environment often leads to issues missed detections false positives. Additionally, factors water quality result weak target features, which adversely affect extraction feature information. Furthermore, lack illumination causes image blur low contrast, thereby increasing difficulty task. To address these issues, we propose novel algorithm called YOLO-GE (GCNet-EMA). First, introduce an enhancement module mitigate impact on Second, high-resolution layer added into network improve problems positives targets. Third, GEBlock, attention-based fusion that captures long-range contextual information suppresses noise from lower-level layers. Finally, combine adaptive spatial head filter out conflicting different Experiments UTDAC2020, DUO RUOD datasets show proposed method achieves optimal accuracy.

Language: Английский

---

Enhancing underwater target detection: Fusion of spatio‐temporal incompletely‐aligned AIS and sonar information via DTW and multi‐head attention mechanism

IET Radar Sonar & Navigation, Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 19, 2024

Abstract In the field of underwater target detection, passive sonar is an important means long‐distance detection. The detection information typically includes both surface and targets, whereas it a great challenge on effectively distinguishing between targets solely based information. Effective fusion AIS (Automatic Identification System) data can leverage their complementary nature to compensate for limitation However, are acquired different principles systems, which essentially multi‐source heterogeneous with obvious spatio‐temporal misalignment in nature. Existing methods normally struggle align time space subject complexity problem. this study, Dynamic Time Warping (DTW) algorithm applied domain. addition, deep learning multi‐head attention mechanism proposed achieve spatial alignment data, where matching same also be successfully achieved. It provides priori knowledge enhance by eliminating interference targets. Based mechanism, abstract features extracted from intermediate‐layer neural networks found effective represent typical motion trajectories, demonstrates effectiveness mechanism. experiment results show that method MatchingSucccessRate over 95%

Language: Английский

---

Multi-Scale Feature Fusion Enhancement for Underwater Object Detection

Sensors, Journal Year: 2024, Volume and Issue: 24(22), P. 7201 - 7201

Published: Nov. 11, 2024

Underwater object detection (UOD) presents substantial challenges due to the complex visual conditions and physical properties of light in underwater environments. Small aquatic creatures often congregate large groups, further complicating task. To address these challenges, we develop Aqua-DETR, a tailored end-to-end framework for UOD. Our method includes an align-split network enhance multi-scale feature interaction fusion small identification distinction enhancement module using various attention mechanisms improve ambiguous identification. Experimental results on four challenging datasets demonstrate that Aqua-DETR outperforms most existing state-of-the-art methods UOD task, validating its effectiveness robustness.

Language: Английский

---