AI-based pigmentation fingerprinting for individual shrimp re-identification: Advancing precision aquaculture

Individual identification of specimens in crustacean aquaculture is fundamentally constrained by the moulting process, which periodically alters external phenotypic features and often necessitates invasive and stressful physical tagging. This study evaluates AI-based pigmentation fingerprinting as a non-invasive, welfare-focused alternative for the Pacific white shrimp (Penaeus (Litopenaeus) vannamei). We benchmarked six advanced deep-learning feature matching algorithms against a classical computer vision baseline to re-identify individuals across sequential moulting events. Using a dataset of individually reared shrimp imaged post-moult, we assessed model performance based on accuracy, runtime, and resource consumption. Our results demonstrate that abdominal chromatophore pigmentation patterns possess sufficient geometric stability to serve as biometric identifiers, with top models achieving 100% cross-moult re-identification accuracy, successfully compensating for non-rigid phenotypic features deformations. The graph-based LightGlue architecture emerged as the optimal tool for high-precision applications, balancing perfect accuracy with moderate computational demands (∼1.16 s per pair). For scenarios requiring real-time throughput, the lightweight D2-Net framework proved to be a rapid alternative (0.04 s per pair) with functional accuracy (76.32%) suitable for edge deployment. These findings establish pigmentation fingerprinting as a viable non-invasive approach for automated individual tracking in shrimp, promoting precision monitoring and animal welfare with potential applicability to other aquatic species exhibiting stable dermal patterns and non-rigid phenotypic changes.