基于可见光图像分析和支持向量机的鸭蛋新鲜度识别

研究鸭蛋内部的可见光图像信息(H和I)的变化来表征鸭蛋新鲜度的变化。采用可见光图像分析技术获取鸭蛋蛋心颜色参数, 将哈夫单位作为新鲜度的验证指标, 建立了基于支持向量机(SVM)的蛋心内部色泽参数的鸭蛋新鲜度预测模型。模型表明：当SVM类型为epsilon-SVR, 核函数为RBF模型特征参数C=27、σ=23时所建立的模型预测效果最好。其中模型预测效果参数RMSEC=0.9520, EMSEP=0.4205。鸭蛋新鲜度预测值与测定值之间具有良好的线性关系, 且预测值对实际值具有较大的搜索覆盖能力。该模型较好的解决了用普通的线性方法识别鸭蛋新鲜度存在的稳定性较差和置信度较小的问题。通过比较SVM和神经网络的识别结果可以发现：SVM对鸭蛋蛋心颜色的识别性能优于神经网络的识别性能(SVM: 98.92%>ANN: 93.77%)。

Detection on Duck Egg′s Freshness Based on Visible Light Image Analyzing and SVM

This article applied the duck egg internal visible light image information (H and I) to characterize the freshness of duck egg. The visible light image analysis technology was applied to obtain the duck egg heart color parameters. The Hafu unit was used as the confirmation index of egg freshness through experiment, then established the forecast model based on SVM (support vector machines) which was developed using the egg heart color parameters to predicate the egg freshness. The model indicated that when we choose the SVM type as epsilon-SVR, the nuclear function as RBF ,the model characteristic parameter C=27, σ=23 ,then the model forecast effect is the best. The model forecast effect parameters RMSEC reached 0.9520, and the EMSEP reached 0.4205. The predicted value of egg freshness and the practical value has good linear correlation and predicted value has a large covering and searching ability on the practical value. This model showed better stability and confidence interval compared with the ordinary linear method. Through the comparison of SVM and neural network recognition results we can get the conclusions that SVM has a better duck-egg freshness recognition performance than ANN by egg heart color's analysis(SVM reach to 98.92%>ANN reach to 93.77%).