摘要
为了提高猪肉新鲜度检测的实时性,提出了基于Caffe框架与ResNet残差神经网络的猪肉新鲜度分级的新方法。根据理化试验结果将猪肉的新鲜度分为7级,并在理化试验前拍摄对应的猪肉照片作为样本进行网络训练。在网络训练完成后分别用同源和异源样本图片对系统分级准确率进行验证,结果显示系统分级的准确率均达到95%以上,说明该系统能够很好地对猪肉新鲜度进行分级。与传统的理化试验检测新鲜度的方法相比,在保证了分级准确率较高的同时,检测过程简单、实时性高、无损,是一种更高效的猪肉新鲜度分级方法。
In order to improve the real-time detection of pork freshness, a new method of pork freshness classification was proposed based on Caffe framework and ResNet residual neural network. According to the results of pork physical and chemical experiments, the freshness of pork was divided into seven grades, and the corresponding pork photos were taken as samples for network training. After the network training, the classification accuracy of the system was validated by homologous and heterogeneous samples, respectively. The results showed that the classification accuracy of the system reached more than 95%, indicating that the system could classify the freshness of pork very well. Compared with the traditional physical and chemical methods, this method is simple, real-time and non-destructive, and it is a more efficient method for pork freshness classification.
作者
邱洪涛
孙裴
侯金波
辜丽川
乔焰
焦俊
QIU Hong-tao;SUN Pei;HOU Jin-bo;GU Li-chuan;QIAO Yan;JIAO Jun(Anhui Agricultural University, Hefei 230036, China;Anhui Hongsen Networking Company Limited, Bozhou 236800, China)
出处
《江苏农业学报》
CSCD
北大核心
2019年第2期461-468,共8页
Jiangsu Journal of Agricultural Sciences
基金
国家自然科学基金项目(31671589
31371533
3177167)
安徽省攻关项目(1804a07020130)
安徽省科技重大专项(16030701092)
