基于计算机视觉鉴别肉松与肉粉松

为实现快速鉴别肉松和肉粉松,文中提出了一种基于计算机视觉和BP人工神经网络的快速无损检测方法。先对60组肉松和60组肉粉松图像进行灰度化并剪切成长宽为899×772像素,然后在剪切后的灰度图像中提取灰度均值、总熵值、灰度等级矩阵的标准差、基于灰度共生矩阵的对比度、相关度、纹理二阶矩和均匀度,在剪切后的二值图像中提取分形维数,共计8个纹理指标,再将45组肉松和44组肉粉松作为训练集输入BP人工神经网络进行训练,剩余样本作为测试集进行测试。研究结果表明:构建的BP神经网络总分类准确率为80.65%,其中有2组肉松被误判为肉粉松,有4组肉粉松被归为肉松。该研究成果可为销售点快速无损鉴别肉松与肉粉松提供了一种技术方法。

Identifying dried meat floss and dried meat powder based on computer vision

In this paper,a rapid nondestructive test method was established by computer vision and back propagation(BP) artificial neural network.It used in identifying dried meat floss and dried meat powder.The images of sixty dried meat floss and dried meat powder each were converted to gray and cut into 899×772 pixels.The gray mean,the total entropy,the standard deviation of gray scale matrix and contrast,correlation,energy and homogeneity based on gray level co-occurrence matrix were extracted from the cut gray images and the fractal dimension was extracted from the cut binary images.These eight texture indicators of forty-five dried meat floss and forty-four dried meat powder pre-processed images were used as the training set in training BP artificial neural network,and the rest samples were used as the test set.Results demonstrated that the total classification accuracy of the final BP artificial neural network was 80.65%.Two dried meat floss and four dried meat powder were miss classified.The study provided a rapid nondestructive analytical system for identifying dried meat floss and dried meat powder.