<![CDATA[Efficient monitoring of crab larvae remains a critical challenge in aquaculture, as early-stage mortality is high due to the lack of practical and scalable detection systems. Although deep learning-based object detection has demonstrated strong performance for small aquatic organisms, many existing approaches are computationally intensive and unsuitable for mobile or resource-constrained hatchery environments. This study investigates the feasibility of lightweight deep learning models for mobile crab larvae detection in aquaculture environments. Using crab larvae at the zoea stage as a case study, lightweight YOLO-based architectures are evaluated to analyze the trade-off between detection accuracy and computational efficiency. The results indicate that extremely lightweight models offer minimal memory requirements and high deployment feasibility, but with limited detection accuracy. In contrast, more advanced lightweight architectures achieve substantially higher accuracy at the cost of increased model size and computational complexity. Rather than focusing solely on algorithmic comparison, this work emphasizes deployment-oriented insights for selecting appropriate lightweight detection models under practical mobile constraints. The findings demonstrate that lightweight deep learning provides a viable foundation for mobile aquaculture applications and establish a baseline for future optimization toward efficient on-device deployment.]]>
Copyrights © 2025
