C16 | Spectroscopic control of fish products: simultaneous recognition of species and thawed status in large-scale distribution

M. Marcolin¹, S. Currò¹, T. Breda¹, A. Franceschi¹, M. Bernardini², S. Strazzacappa², J.C. Ferlito³, M. Gazziero³, Novelli¹, F. Fontana¹, L. Fasolato¹, S. Balzan¹. | ¹Dip. di Biomedicina Comparata e Alimentazione, Università degli Studi di Padova, Legnaro; ²Alì SPA, Padova; ³ITPhotonics S.r.l., Fara Vicentino.

Purpose. The study aims to evaluate the applicability of near-infrared spectroscopy (NIRS) as a tool to support the traceability of fish products marketed in large-scale retail chains, with particular attention to those pre-packaged in plastic film. Methods. A total of 218 fish products distributed by 41 retail outlets were selected, pre-wrapped in plasticised polyvinyl chloride film for food use. For each sample, spectral data were collected both with the film and in direct contact with the product, using a portable NIR instrument (PoliSpecNIR, ITPhotonics, Italy; 900-1680 nm, 2 nm resolution) . This untargeted, rapid and environmentally friendly method is capable of providing a spectral fingerprint of the chemical-physical composition of the samples, with both qualitative and quantitative applicability for food analysis at every stage of the supply chain, and has been recognised as a technique to support the inspection assessment of the physical state (fresh/thawed) of fish products by Implementing Regulation (EU) 2022/2503. The samples analysed were divided into three datasets: i) two according to physical status (fresh or thawed) and ii) one for the attribution of the eight total species: Gadus macrocephalus (No. 41), G. morhua (No. 6), Pollachius virens (n°33), Lophius piscatorius (n°20) and L. vomerinus (n°11), Sepia officinalis (n°57) and Salmo salar (n° 47). The spectral data derived from the scans were analysed using chemometric methods to evaluate the effect of the film. An initial unsupervised exploratory approach was carried out using Principal Component Analysis (PCA) as shown in the figure, and two classification models were developed on the raw data using linear and kernel Support Vector Machines, optimised with the implementation of “grid search”. The predictive model was based on a hold-out, 60% training and 40% validation with proportional division of classes, aimed at reducing the influence of external factors and increasing the robustness of the classification models themselves. Significant wavelengths for data analysis were identified using the Boruta algorithm for status and species. Results. The linear classification performance of the fresh and thawed datasets showed an accuracy (AC) of over 87% for both acquisition methods, with and without film, recording a significant result in all cases with a P-value of less than 0.05. The species identification results were similarly significant with an AC of around 80%. Conclusions. The film has a negligible impact on the reflectance signal of the sample, allowing the development of promising classification models for tracking both fresh and thawed products and for species identification, even in prepared products that have lost their anatomical integrity. The ability to obtain reliable spectroscopic fingerprints through food film reduces acquisition times and product handling, promoting food control and safety. The use of this data acquisition system has important implications for the creation of NIRS models and libraries useful for the authentication, traceability and quality management of product batches at the point of sale and the implementation of self-monitoring.