Deep Learning Model Based on E-Noses Sensors in Food Production Controlling System

Abstract

In the agricultural food manufacture region, ensuring and evaluating food quality is vital because it directly affects human health and the profitable price of the invention. The aroma of the product is a crucial characteristic that reflects its quality. A prominent trend in this area is the utilization of electronic noses (e-noses) for automated replication of smells. This involves deploying multiple sensors to detect specific compounds that contribute to the product's odor and overall quality. The reliable assessment of food quality depends on the proper functioning of these sensors, which provide digital data used for classifying food quality. However, addressing this issue has led to the implementation of various strategies, often focusing on correcting data from digital sensors. In our research, we propose an innovative approach using a Deep Learning model to leverage digital time series data from sensors for classification tasks. To maintain overall prediction accuracy, we employ a Multiple Layer Perceptron (MLP) neural network for classification prediction tasks. This method trains the proposed MLP classifier on a dataset from food production that includes 11 digital sensors (such as hydrogen sulfide, ammonia, and hydrogen sensors) across various types of beef cuts, including brisket. As a result, unlike traditional machine learning models, our approach can effectively handle data generated from sensors and address the different classes –categories-(excellent(1),good(2),acceptable(3),spoiled(4))thereby enhancing food quality assessment. Consequently, this study demonstrates the effectiveness of our proposed model through a case study focused on predicting the quality of beef cuts, yielding promising results that can be applied to general food quality assessment.