Evidence explorer
Read the study by layer.
Move between the research question, method, recorded result, and interpretation boundary without losing context.
Seafood freshness assessment is essential for quality control, shelf-life management, and supply-chain decisions, yet conventional sensory, chemical, and microbiological measurements are often destructive or laboratory-dependent. This systematic review synthesizes evidence on deep learning (DL)-based non-destructive seafood freshness assessment, focusing on sensing modality, model architecture, predictive performance, reference-standard validity, methodological robustness, and deployment readiness. Scopus and Web of Science records published in English from 2021 to 2026 were screened. Of 1,057 identified records, 306 remained after filtering and deduplication; 32 full texts were assessed and 28 studies were included. Four technology families emerged: red-green-blue (RGB) computer vision; hyperspectral, near-infrared, Raman, and fluorescence spectroscopy; colorimetric or fluorescent intelligent sensors; and multimodal or olfactory-sensor fusion. Reported performance was often high, with classification accuracies above 95% and regression R² values above 0.90 in several studies, but comparisons were limited by heterogeneous species, freshness labels, datasets, and validation designs. Studies anchored to total volatile basic nitrogen (TVB-N), total viable count/aerobic plate count (TVC/APC), K-value, thiobarbituric acid reactive substances (TBARS), or biogenic amines provided stronger freshness validation than storage-time or visual-class labels alone. Methodological appraisal classified 22 studies as some concern and six as high concern, with no study classified as low concern. Mobile, edge, portable-sensor, and robotic implementations show translational promise, but external and industrial validation remain limited. Technological maturity therefore depends on biological validity, independent validation, computational feasibility, and real-world deployment, not predictive accuracy alone.