Eggshell Crack Detection with Rolling Eggs on a Corrugated Plate and LIBSVM

Abstract. In this research, the acoustic signals of eggs rolling on a corrugated plate were collected with an electret microphone that was mounted on the back of the plate. Time-domain signals were transformed into frequency-domain signals by Fourier frequency transformation (FFT). Peaks in four frequency ranges were selected as feature vectors between intact and cracked eggs. LIBSVM was employed to build a classification model using these feature vectors. The classification performance of LIBSVM was further improved by parameter optimization and Max-Min and Z-score normalizations. The highest area under curve (AUC) of 1 and classification accuracy of 100% for the prediction set indicated that support vector machine (SVM) was appropriate for this detection method. In addition, linear SVM was considered slightly superior to discriminant analysis (DA) based on Mahalanobis distance and logistic regression (LR). The complete testing procedure for a single egg required about 0.9 s. It is concluded that this research provides a feasible and simple method for eggshell crack detection.