YOLO-Shrimp: A Lightweight Detection Model for Shrimp Feed Residues Fusing Multi-Attention Features.

Area of Science:

• Aquaculture technology
• Computer vision
• Artificial intelligence

Background:

• Accurate monitoring of residual feed is crucial for cost-effective and environmentally sustainable intensive shrimp farming.
• Current manual methods for assessing residual feed are subjective, inefficient, and difficult to standardize.
• Automated detection of small, dense, and occluded feed particles presents significant technical challenges.

Purpose of the Study:

• To develop an efficient and accurate automated system for detecting residual feed in shrimp farming.
• To enhance the feature extraction capabilities of deep learning models for small and dense objects.
• To reduce computational complexity and parameter count in object detection models for aquaculture applications.

Main Methods:

• Proposed a lightweight object detection model named YOLO-Shrimp.
• Introduced a novel attention mechanism, EnSimAM, for multi-scale feature perception of small targets.
• Implemented an enhanced weighted intersection over union loss function (EnWIoU) for improved localization accuracy.
• Utilized the RepGhost module as the model backbone to reduce parameters and computational load.

Main Results:

• YOLO-Shrimp achieved mean Average Precision (mAP) scores of 70.01% (mAP@0.5) and 28.01% (mAP@0.5:0.95) on a real-world dataset.
• The model demonstrated a 19.7% reduction in parameter count and a 14.6% decrease in GFLOPs compared to the baseline.
• EnSimAM effectively improved feature extraction for small and dense residual feed particles.

Conclusions:

• YOLO-Shrimp offers a promising solution for automated residual feed detection in shrimp farming.
• The developed model enhances detection accuracy and efficiency while reducing computational resources.
• This technology can contribute to optimized feeding strategies, cost reduction, and environmental protection in aquaculture.