1D-CNN Model for Fast Assessment of Beef Freshness from Noisy Electronic Nose Signal

Maintaining and ensuring the quality of beef throughout the supply chain is critical for both the food safety and consumer satisfaction. Because of the high perishable nature of beef, accurate and fast assessment of freshness is crucial. Electronic noses (E-noses), devices which contain arrays of gas sensors, present an accurate, non-destructive, and rapid solution. However, the raw signals from the gas sensors are often noisy which requires pre-processing techniques such as signal reconstruction thereby hindering practical implementation. This paper addresses this challenge by utilizing a 1D-Convolutional Neural Network (1D-CNN) model in assessing beef freshness directly from raw and noisy E-nose signals. This approach eliminates the need for signal reconstruction techniques that are often time-consuming. By training the model from raw E-nose signal, the model learns the necessary parameters for handling the noisy input generated by the gas sensor array. This study showcases the effectiveness of the approach by evaluating its performance against a real-world dataset of E-nose signals collected from 12 cuts of beef with 4 degrees of freshness. The model achieves an accuracy of 97.22% and an F1-score of 0.9611 on the test dataset suggesting its strong ability to discriminate beef freshness from the noisy E-nose signals. This implies strong potential of the 1D-CNN model for practical applications in the food industry particularly in quality control and assurance of highly-perishable goods.