基于核函数的SVM猪肉颜色等级评定研究

Study on Color Classification Assessment for Pork Based on Kernel Function of SVM

为了实现用计算机和机械设备进行猪肉颜色自动化分级,本研究对猪肉样品照片进行图像处理,提取其中颜色特征参数,并进行色彩空间参数换算。通过对基于核函数的3种SVM多分类方法进行比较,选择出最适合于猪肉颜色的SVM多分类评定方法。对比结果显示,采用单独的HSV数据及RGB与HSV联合数据进行分类,分类效果好于RGB数据。RBF核函数"二叉树"SVM多分类模型,经过样本学习后,其分类的正确率可达98%;同时考虑经验风险和置信风险,其分类正确率达80%。

To realize the auto classification for pork color with computer and equipment, the processing for pork sample images, drawing of color characteristics and spatial data conversion was studied. With the comparison of three SVM multi-classification methods based on kernel function, the most suitable SVM classification assessment methods for pork color were found. Comparison results showed that the using HSV and the combination usage of RGB and HSV had higher classification efficiency than com- puter recognized RGB. The accuracy of pork multi classification would be 98% after sample studies with the binary tree model of RBF kernel function. With the consideration of empirical risk and confi- dence risk at the same time, the accuracy was guaranteed to reach 80%.