基于Welch法功率谱和广义回归神经网络的禽蛋裂纹辨识

Eggshell Crack Identification Based on Welch Power Spectrum and Generalized Regression Neural Network(GRNN)

为建立一种快速无损检测禽蛋裂纹的方法,构建了基于磁致伸缩振子扫频式振动的禽蛋裂纹检测系统。系统以声学特性为基础,通过利用Welch法功率谱分析禽蛋振动音频信号,利用主成分分析法提取特征向量中的有用信息并构建基于广义回归神经网络(generalized regression neural network,GRNN)的禽蛋裂纹检测模型。实验对290枚鸡蛋进行检测(训练集200枚,测试集90枚)。结果表明,测试集中无损蛋与裂纹蛋的判别率分别达到96.7%和98.3%。研究表明,利用磁致伸缩振子扫频和Welch法功率谱分析,通过主成分分析法提取特征向量中的有用信息并结合GRNN模型检测禽蛋裂纹是可行的。

This study aimed to establish a quick method for non-destructive testing of cracked eggs. We firstly developed a detection system for cracked eggs based on sweep frequency vibration of the magnetostrictive vibrator. The system was based on the acoustic characteristics, and by Welch power spectrum analysis of vibration audio signal of eggs and extraction of useful information in the feature vector through the principal component analysis（PCA）, the detection model for egg cracks was constructed based on generalized regression neural network（GRNN）. A total of 290 eggs, including 200 eggs in the training set and 90 eggs in the test set, were detected in this study. The results showed that the recognition rates of intact eggs and cracked eggs reached 96.7% and 98.3%, respectively, in the test set. The research indicated the feasibility of using the magnetostrictive vibrator sweep and Welch power spectrum analysis and extracting useful information in the feature vector through PCA method coupled with GRNN neural network model for the detection of cracked eggs.