A colorimetric and photothermal dual-mode nanozyme platform: Multimodal deep learning guided real-time freshness sensing and predicting of aquatic products under challenging lighting conditions

A colorimetric and photothermal dual-mode method was conducted for hypoxanthine detection. • The 2D Fe/Ni-MOF-based sensing platform demonstrated excellent sensitivity, accuracy and biocompatibility. • The dual-mode method was developed for the early evaluation of freshness in aquatic products. • A LAMF-Net model was trained with accuracy >97% for freshness prediction under challenging lighting conditions. • Rapid freshness assessment was enabled via self-developed intelligent online detection platform. Timely and accurate freshness assessment of aquatic products with complex matrices remains significant challenge. Hypoxanthine (Hx)-based monitoring methods using colorimetric sensing with smartphone are widely used but affected by variable lighting conditions throughout the food supply chain, causing detection results remain inaccurate. Here, a colorimetric/photothermal dual-mode Hx detection and freshness prediction platform was constructed by cascading a thermostable 2D Fe/Ni-MOF nanosheet with xanthine oxidase, enabling stable and sensitive on-site detection of Hx. A linear response to Hx was exhibited in the range of 0.2–100 μM, with detection limits of 0.12 μM (colorimetric mode) and 4.21 μM (photothermal mode), respectively. To address ambient lighting effects, seven lighting conditions (~50, ~100, ~500, ~1000, ~5000, ~10,000, and ~50,000 lx) were simulated to construct a dataset integrating lighting, colorimetric, and photothermal information for the model training and testing. A light-adaptive multimodal fusion network based on YOLOv11 and Multi-Layer Perceptron (MLP) achieved a prediction accuracy of 98% in freshness classification, remarkably higher than that of the single colorimetric (89%) and photothermal (93%) modes, when trained on 4536 labeled images and 108 photothermal data points, demonstrating strong robustness under complex and varying lighting conditions. Coupled with the self-developed “Food Track” application, this work not only provides a practical and reliable approach for the early freshness prediction in aquatic products, but also highlights the potential of multimodal fusion algorithms to handle lighting variations in food supply chains. Download: Download high-res image (120KB) Download: Download full-size image