Automated Classification of Egg Freshness Using Acoustic Signals and Convolutional Neural Networks

Abstract

The acoustic response of eggs has not been fully optimized for quality detection purposes. This research investigates the utilization of acoustic response as input features for classifying fresh and rotten eggs using Convolutional Neural Networks (CNN). Acoustic measurements were conducted using an experimental setup involving a striking rod, microphone, and microcontroller to capture acoustic responses from eggs struck vertically and horizontally. Features such as energy, entropy, zero crossing rate, spectral centroid, and Mel Frequency Cepstral Coefficients (MFCC) were extracted from the recorded acoustic signals. The study utilized a sample set of 200 intact chicken eggs without cracks, equally split between fresh and rotten eggs. Principal Component Analysis (PCA) revealed significant differences in acoustic responses between fresh and rotten eggs, with clear separation observed in horizontal measurements. Three primary classification models—Convolutional Neural Network (CNN), Deep Neural Network (DNN), and Support Vector Machines (SVM)—were evaluated for this classification task. In vertical tapping, CNN and DNN demonstrated stable performance with approximately 85% accuracy, while SVM slightly trailed with around 80% accuracy. However, in horizontal tapping, all three models achieved perfect classification with accuracy, precision, recall, and F1-score reaching 100%. This study demonstrates the effective use of acoustic response as input for Deep Learning models to distinguish between fresh and rotten eggs. The findings highlight the potential application of this technology in the food industry to ensure product quality before consumption, with significant implications for developing reliable automated solutions. Future research could expand validation using larger datasets and optimize sensor usage and data processing techniques to enhance detection consistency and reliability.