A fish counting model based on pyramid vision transformer with multi-scale feature enhancement

Ecological Informatics, Journal Year: 2025, Volume and Issue: unknown, P. 103025 - 103025

Published: Jan. 1, 2025

Language: Английский

---

Underwater instance segmentation: a method based on channel spatial cross-cooperative attention mechanism and feature prior fusion

Frontiers in Marine Science, Journal Year: 2025, Volume and Issue: 12

Published: April 1, 2025

In aquaculture, underwater instance segmentation methods offer precise individual identification and counting capabilities. However, due to the inherent unique optical characteristics high noise in imagery, existing models struggle accurately capture global local feature information of objects, leading generally lower detection accuracy models. To address this issue, study proposes a novel Channel Space Coordinates Attention (CSCA) attention module A Prior Fusion (CAPAF) fusion module, aiming improve segmentation. The CSCA effectively captures by combining channel spatial weight, while CAPAF optimizes removing redundant through learnable parameters. Experimental results demonstrate significant improvements when these two modules are applied YOLOv8 model, with [email protected] metric increasing 3.2% 2% on UIIS dataset. Furthermore, is significantly improved USIS10K datasets after other networks.

Language: Английский

---

Fish species identification on low resolution—a study with enhanced super-resolution generative adversarial network (ESRGAN), YOLO and VGG-16

PeerJ Computer Science, Journal Year: 2025, Volume and Issue: 11, P. e2860 - e2860

Published: April 30, 2025

An intelligent detection and recognition model for the fish species from camera footage is urgently required as fishery contributes to a large portion of world economy, these kinds advanced models can aid fishermen on scale. Such incorporating pick-and-place machine be beneficial sorting different in bulk without human intervention, significantly reducing costs large-scale fishing industries. Existing methods detecting recognizing have many limitations, such limited scalability, accuracy, failure detect multiple species, degraded performance at lower resolution, or pinpointing exact location fish. Modifying head compelling deep learning model, namely VGG-16, with pre-trained weights, used both find an image by implementing modified You Only Look Once (YOLO) incorporate bounding box regression head. We proposed using Enhanced Super Resolution Generative Adversarial Network (ESRGAN) algorithm neural network amplify resolution factor 4. With this method, overall accuracy 96.5% has been obtained. The experiment conducted based total 9,460 images spread across nine species. After further improving could integrated quickly sort according their

Language: Английский

---

DeepFins: Capturing dynamics in underwater videos for fish detection

Ecological Informatics, Journal Year: 2025, Volume and Issue: unknown, P. 103013 - 103013

Published: Jan. 1, 2025

Language: Английский

---

Application of Improved YOLOv8n-seg in Crayfish Trunk Segmentation

Israeli Journal of Aquaculture - Bamidgeh, Journal Year: 2024, Volume and Issue: 76(4)

Published: Dec. 17, 2024

The crayfish industry ( Procambarus clarkii ) is experiencing rapid growth. However, the processing sector continues to face challenges due a lack of advanced automation, relying heavily on manual visual inspection assess specifications and integrity, which limits efficiency precision in decision-making. To address issue intelligent grading P. , this work proposes GHB-YOLOv8-seg algorithm for segmenting main trunk shrimp based YOLOv8n-seg model. original network replaced through coupling Ghost HGNetV2, depth-separable convolution employed perform linear transformation features. This results reduction number parameters computational complexity while maintaining high accuracy. reduced; concurrently, introducing weighted bidirectional feature pyramid (BiFPN) enables model multi-scale fusion with greater alacrity, thereby enhancing model’s performance. Ultimately, was achieved by calculating pixel area after segmentation converting it actual body weight. demonstrated that improved reduced 60.5%, size 55.4%, mAP value increased from 98.9% 99.2%. study indicates facilitates precise lightweight trunk, can be integrated into diverse mobile devices.

Language: Английский

---

Hierarchical deep learning framework for automated marine vegetation and fauna analysis using ROV video data

Ecological Informatics, Journal Year: 2024, Volume and Issue: unknown, P. 102966 - 102966

Published: Dec. 1, 2024

Language: Английский

---