基于机器视觉的生鲜牛肉冷藏时间识别研究

Identification of the refrigerated time for chilled beef based on machine vision

为解决生鲜牛肉冷藏时间人为识别误差大且无法量化的问题,该研究以4种冷藏时间共400块切片生鲜牛肉为研究对象,通过图像采集系统获取样本图像后搭建基于GoogLeNet的改进模型进行生鲜牛肉冷藏时间识别,并在此基础上引入迁移学习理论和数据增强技术来提高目标识别准确率。结果显示,该模型对测试集图像数据识别准确率为91.18%,迁移学习机制辅以数据增强技术能有效缓解复杂网络模型过拟合问题,并且在相同环境下的对比实验中,对生鲜牛肉冷藏时间的识别准确率优于反向传播式神经网络、深度卷积网络及GoogLeNet原始模型,可实现高效率、高准确性的生鲜牛肉冷藏时间识别,为生鲜牛肉的运输贮存安排及环境设置提供参考。

To address the problem of the artificial recognition error and quantify the chilled beef refrigerated time the time,this study took 400 slices of chilled beef as raw materials with four kinds of refrigerated time to explore it.After obtaining the sample images through the image acquisition system,an improved model based on GoogLeNet was set up to identify the refrigerated time.Moreover,the transfer learning theory and data enhancement technology were applied to improve the accuracy of target recognition.The results showed that the recognition accuracy of the model was 91.18%,and the transfer learning mechanism with data enhancement technology could effectively alleviate the over-fitting problem of the complex network model.The results of the comparative experiments showed that the recognition accuracy of the method in this study was better than that of the back propagation neural network,the VGG convolution neural network and the GoogLeNet.In summary,the method in this research could realize the recognition of chilled beef refrigerated time with high efficiency and satisfactory accuracy.It could also provide a reference for the transportation schedule and environmental arrangement of chilled beef in the near future.