Eggshell Crack Detection And Egg Classification Using Resonance And Support Vector Machine Methods

Abstract

Recent studies on the detection of eggshell cracks have mainly focused on the Fourier frequency analysis of a vibration-based response. This paper proposes a new detection methodology for eggshell cracks using a continuous wavelet transform and a support vector machine (SVM) technique. The proposed methodology consists of an experimental system and a data processing system. The experimental system was used to generate the impact force and to measure the acoustic signal, whereas the data processing system extracted timely signal features and classified the signals based on an SVM algorithm that is presented herein. The wavelet transform was introduced with the objective of overcoming the drawbacks of the conventional Fourier analysis and of finding new crack-sensitive features based on the local time and frequency information of the signal. Several wavelet-based features were extracted through the analysis of the energy distribution of the wavelet transform coefficients for cracked and intact eggs. These features include the first resonance scale (FS), the s-coordinate (SC) and the corrected t-coordinate (TC) of the centroid for the time-scale domain, and the weighted standard deviation (WD). These features, as well as their various combinations, were used to construct the SVM classifiers. The classification effects of the best single-, two-, and three-feature SVM classifiers and the four-feature SVM classifier are provided. With a detection scheme based on four measurements for every egg, the experimental example achieved the highest crack detection rate of 98.9%, and the smallest false rejection rate of 0.8%.