NDT of closed-shell oyster freshness by acoustic vibration signals

Abstract

This study introduces an innovative method combining acoustic vibration technology with machine learning (ML) to non-destructively assess the freshness of closed-shell oysters. An acoustic vibration system, developed in-house, gathers vibration signals, which are then processed through a fusion strategy that integrates two time-domain features, six frequency-domain features, and one time-frequency domain feature based on an improved MFCC. Utilizing these fused features, the Stacking ensemble learning algorithm was employed to integrate six mainstream machine learning classification algorithms, leveraging their strengths in signal analysis to build a high-performance freshness detection model. Cross-validation assessments reveal the model's accuracy at 98%, highlighting how multi-feature fusion and ensemble learning algorithms significantly improve detection precision. Comparative studies further demonstrate that fusion features notably enhance classification accuracy over using domain-specific features alone. The experimental results indicate that the proposed method successfully navigates the challenge posed by the oyster shell to internal detection, achieving dynamic decay detection of oyster freshness. This provides a new technological approach for quality control in shellfish products like oysters, contributing meaningfully to food safety and quality enhancement.