SAM-Enhanced Visual Measurement for Fine-Grained Detection and Density Estimation of Salmon Lice in Aquaculture

Accurate measurement of salmon lice density and life-cycle stages is essential for effective health management in aquaculture, yet existing vision-based approaches mainly focus on coarse localization and fail to provide reliable density estimation under complex underwater conditions. To address this limitation, this paper proposes a novel visual measurement framework that integrates an instance-level detection model with the Segment Anything Model (SAM) for fine-grained salmon lice analysis. The proposed method first localizes individual lice candidates and then leverages SAM to obtain precise instance segmentation, enabling accurate estimation of lice density and morphological characteristics across different developmental stages. Compared with conventional detection-only approaches, the integration of SAM significantly improves robustness to target overlap, fuzzy boundaries, and illumination variations commonly encountered in underwater environments. Extensive experiments on real fishery data demonstrate that the proposed framework consistently outperforms state-of-the-art methods in both localization accuracy and density estimation, validating its effectiveness and practical applicability for intelligent aquaculture health monitoring.