摘要
提出了一种基于多层小波变换和纹理分析的蛋壳破损检测方法。该方法对获取的鸡蛋透射图像G分量在不同水平上进行小波分解,计算和分析各水平高频细节子图像的纹理特征参数,实验确定最有效的8个特征参数作为BP网络输入,建立结构为8-20-2的BP神经网络蛋壳破损分类模型。实验表明,该方法对无破损蛋、线状破损蛋、网状破损蛋和点状破损蛋的判别正确率分别为95%、90%、95%、80%,平均识别率为90%。
A new method of crack detection in eggs was proposed with multi-level wavelet transform and texture analysis technology. First, G gray level image of all egg images were decomposed into approximation and detail sub-images at various levels by wavelet transform. Then, the feature vector which was composed of wavelet texture energy features, and the gray-level co-occurrence matrix features of the detail sub-images were analyzed and computed. Finally, with the most appropriate and effective eight parameters as inputs, the best BP neural network （8 input nodes, 20 hidden nodes, 2 output nodes）was employed to detect egg crack and classify eggs. The results of experiment proved that the correct discerning rate to detect eggs without crack and eggs with linear crack, meshy crack, point crack is respectively 95 %, 90 %, 95 % and 80 %, and the average correct rate to detect crack in eggs is 90 %.
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2009年第2期170-174,共5页
Transactions of the Chinese Society for Agricultural Machinery
基金
国家"863"高技术研究发展计划资助项目(2007AA10Z214)
关键词
鸡蛋
破损
检测
小波变换
BP神经网络
Egg, Crack, Detection, Wavelet transform, BP neural network
