YOLO-CE: an underwater low-visibility environment target detection algorithm based on YOLO11

The Journal of Supercomputing, Journal Year: 2025, Volume and Issue: 81(5)

Published: April 10, 2025

Language: Английский

---

FishDet-YOLO: Enhanced Underwater Fish Detection with Richer Gradient Flow and Long-Range Dependency Capture through Mamba-C2f

Electronics, Journal Year: 2024, Volume and Issue: 13(18), P. 3780 - 3780

Published: Sept. 23, 2024

The fish detection task is an essential component of marine exploration, which helps scientists monitor population numbers and diversity understand changes in behavior habitat. It also plays a significant role assessing the health ecosystems, formulating conservation measures, maintaining biodiversity. However, there are two main issues with current algorithms. First, lighting conditions underwater significantly different from those on land. In addition, light scattering absorption water trigger uneven illumination, color distortion, reduced contrast images. accuracy algorithms can be affected by these variations. Second, wide variation species shape, color, size brings about some challenges. As have complex textures or camouflage features, it difficult to differentiate them using To address issues, we propose algorithm—FishDet-YOLO—through improvement YOLOv8 algorithm. tackle complexities environments, design Underwater Enhancement Module network (UEM) that jointly trained YOLO. UEM enhances details images via end-to-end training species, leverage Mamba model’s capability for long-distance dependencies without increasing computational complexity integrate C2f create Mamba-C2f. Through this design, adaptability handling tasks improved. RUOD DUO public datasets used train evaluate FishDet-YOLO. FishDet-YOLO achieves mAP scores 89.5% 88.8% test sets DUO, respectively, marking 8% 8.2% over YOLOv8. surpasses recent state-of-the-art general object

Language: Английский

---

Attention-Based Lightweight YOLOv8 Underwater Target Recognition Algorithm

Sensors, Journal Year: 2024, Volume and Issue: 24(23), P. 7640 - 7640

Published: Nov. 29, 2024

Underwater object detection is highly complex and requires a high speed accuracy. In this paper, an underwater target model based on YOLOv8 (SPSM-YOLOv8) proposed. It solves the problems of computational complexities, slow speeds low accuracies. Firstly, SPDConv module utilized in backbone network to replace standard convolutional for feature extraction. This enhances efficiency reduces redundant computations. Secondly, PSA (Polarized Self-Attention) mechanism added filter enhance polarization features channel spatial dimensions improve accuracy pixel-level prediction. The SCDown (spatial-channel decoupled downsampling) downsampling then introduced reduce cost by decoupling space operations while retaining information process. Finally, MPDIoU (Minimum Point Distance-based IoU) used CIoU (Complete-IOU) loss function accelerate convergence bounding box regression experimental results show that compared with YOLOv8n baseline model, SPSM-YOLOv8 (SPDConv-PSA-SCDown-MPDIoU-YOLOv8) reaches 87.3% ROUD dataset 76.4% UPRC2020 dataset, number parameters amount computation decrease 4.3% 4.9%, respectively. frame rate 189 frames per second thus meeting requirements algorithms facilitating lightweight fast edge deployment.

Language: Английский

---

CIS: A Coral Instance Segmentation Network Model with Novel Upsampling, Downsampling, and Fusion Attention Mechanism

Journal of Marine Science and Engineering, Journal Year: 2024, Volume and Issue: 12(9), P. 1490 - 1490

Published: Aug. 28, 2024

Coral segmentation poses unique challenges due to its irregular morphology and camouflage-like characteristics. These factors often result in low precision, large model parameters, poor real-time performance. To address these issues, this paper proposes a novel coral instance (CIS) network model. Initially, we designed downsampling module, ADown_HWD, which operates at multiple resolution levels extract image features, thereby preserving crucial information about edges textures. Subsequently, integrated the bi-level routing attention (BRA) mechanism into C2f module form C2f_BRA within neck network. This effectively removes redundant information, enhancing ability distinguish features reducing computational redundancy. Finally, dynamic upsampling, Dysample, was introduced CIS better retain rich semantic key feature of corals. Validation on our self-built dataset demonstrated that significantly outperforms baseline YOLOv8n model, with improvements 6.3% 10.5% PB PM 2.3% 2.4% mAP50B mAP50M, respectively. Furthermore, reduction parameters by 10.1% correlates notable 10.7% increase frames per second (FPS) 178.6, thus meeting operational requirements.

Language: Английский

---

Conceptual Validation of High-Precision Fish Feeding Behavior Recognition Using Semantic Segmentation and Real-Time Temporal Variance Analysis for Aquaculture

Biomimetics, Journal Year: 2024, Volume and Issue: 9(12), P. 730 - 730

Published: Nov. 30, 2024

Aquaculture plays an important role in the global economy. However, unscientific feeding methods often lead to problems such as feed waste and water pollution. This study aims address this issue by accurately recognizing fish behaviors provide automatic bait casting machines with scientific strategies, thereby reducing farming costs. We propose a behavior recognition method based on semantic segmentation, which overcomes limitations of existing dealing complex backgrounds, splash interference, target overlapping, real-time performance. In method, we first segment targets images using segmentation model. Then, these segmented are input into our proposed By analyzing aggregation characteristics during process, can identify behaviors. Experiments show that has excellent robustness performance, it performs well case background occlusion targets. aquaculture industry efficient reliable for behavior, offering new support intelligent delivering powerful solutions improve management production efficiency. Although algorithm shown good performance recognition, requires certain lighting conditions density, may affect its adaptability different environments. Future research could explore integrating multimodal data, sound information, assist judgment, enhancing model promoting development aquaculture.

Language: Английский

---