A comprehensive systematic review and bibliometric analysis of fish freshness assessment using computer vision

Horna Gomez I.A.; Mendez Valdivia C. · Applied Food Research (2026) · DOI: 10.1016/j.afres.2026.102673

Resumen

The perishability of fish and the limitations of traditional sensory, physicochemical, and microbiological evaluation methods have driven the development of automated systems based on computer vision. This study conducts a systematic review and bibliometric analysis of the evidence published between 2021 and 2025 on the automatic classification of fish freshness using computer vision, machine learning or deep learning, following the PRISMA protocol. Of 1123 publications indexed in Scopus, ScienceDirect, Web of Science, and PubMed, 18 met the inclusion criteria. Architectures such as CNN, DenseNet121, ResNet50, and the MobileNet family predominate, achieving accuracies exceeding 98% under controlled conditions. The bibliometric analysis revealed a geographic concentration in Turkey (33.33%) and Indonesia (22.22%), with only marginal representation from Latin America and a complete absence of studies from Africa. Common methodological gaps were identified: reliance on accuracy as the sole metric, heterogeneity in data partitioning schemes, a scarcity of standardized datasets, and the absence of methodological frameworks such as CRISP-DM. It is recommended to develop standardized protocols for evaluating different fish species, expand research into new geographic contexts, and further investigate emerging architectures such as Vision Transformers, which were not identified among the studies included in the systematic corpus.

Instituciones: Universidad Nacional de Ingeniería