Low-cost multispectral sensor reveals cold chain breaks, meat type, and storage time in chicken meat samples

Abstract

A portable and low-cost spectrometric system has been proposed to discriminate cold chain (frozen) interruption during the supply chain, as well as, the sample type among four different tested chicken meat and products (nugget, wing, thigh and drumstick) and the storage time elapsed after the cold chain break. To achieve this, an AS7265x sensor chipset was used and controlled by a custom programmed microcontroller through a Qwiic I2C interface. Communication between the user and the sensor setup was made using a custom programmed interface developed using the Python. Frozen chicken samples were thawed under various temperature-time combinations (4, 8, 16, and 24 h at 4 and 24 °C) and then refrozen at -18 °C. The refrozen samples were stored for up to 60 days. Spectral measurements (410–940 nm) were taken before thawing and at 1, 10, 30 and 60 days of storage after refreezing. The spectra obtained were analysed using Principal Component Analysis and outliers were eliminated using Mahalobobis distance. Several classification models (including Decision Tree, Random Forest, AdaBoost, Support Vector Machine, k-Nearest Neighbors, Gaussian-Naïve Bayes, Multilayer Perceptron, Gaussian Process, Linear and Quadratic Discriminant Analysis and Partial Least Squares Discriminant Analysis) were trained where appropriate. The top models with the highest predictive ability were selected and merged to build Soft Voting Classifiers (SVC) for each feature to be predicted. The SVC models correctly classified the samples with 93%, 92% and 83% accuracy for cold chain interruption, chicken meat/product type and storage time after thaw-refreeze cycle, respectively. In addition, a graphical interface was also developed so that it can be used by end users for food safety concerns. The results showed the potential of the developed low-cost sensor to rapidly detect the potential food safety risks due to cold chain breakage and to trace back the problem by predicting the storage time after refreezing.