Detection for Eggshell Crack Based on Acoustic Feature and Support Vector Machine

Abstract

The paper has proposed a new method based on acoustic feature and support vector machine. A sound signal acquisition system is designed based on microcontroller, the power spectra is received for good shell eggs and crack eggs. 4 parameters, such as the average power spectrum area (x1), power spectrum area of range value (x2), the first average formant amplitude (x3) and the first formant amplitude range value (x4), are extracted. These 4 parameters are regarded as input vector for support vector machine (SVM). The advantages and disadvantages for classification performance because of different kernel functions and different training sample size are compared, and ultimately the radial basis function (RBF) function is regarded as the best kernel function for the optimal classification results, and then the penalty coefficient C and the normalization coefficient are optimized, the overall recognition rate reached 97.37% or more, running time is about 0. 3s.The results show that SVM has a perfect performance in eggshell crack detection.