基于小波能量谱和BP神经网络的皮蛋破损检测

Preserved egg crack detection based on wavelet energy spectrum and BP neutral network

根据皮蛋蛋壳的声学特性,采用小波变换和BP神经网络相结合的处理方法,对皮蛋进行破损检测。首先对采集到的声音信号进行6层小波分解,计算每层分解信号的小波能量谱,构造小波能量谱分布的特征向量,然后将其作为神经网络的输入向量,基于MATLAB创建了网络结构为6-20-2的BP神经网络。检测结果表明,该方法对好壳皮蛋的识别率为88.5%,对损壳皮蛋的识别率为83.3%。

This paper presents a method based on wavelet energy spectrum and BP neutral network to detect preserved egg cracks. Firstly, the acoustic response was collected and decomposed into 6 levels with wavelet transform. Then the wavelet energy spectrum was extracted on every level. The eigenvector based on wavelet power spectrum was formed, which was used as the input of BP neutral network. The 6-20-2 BP neutral network was built based on MATLAB. The experimental example achieved a crack detection rate of 88. 5% with normal-shell preserved eggs, and 83. 3% with cracked-shell preserved eggs.