应用特征融合预测富士苹果可溶性固形物含量

The prediction of soluble solids content of fuji apples based on feature fusion

针对传统苹果无损检测方法成本高,不利于携带等问题,使用富士苹果RGB图像的不同特征预测其可溶性固形物含量。通过统计方法和卷积神经网络提取苹果图像的颜色特征、纹理特征和局部特征。拼接以上特征,利用融合特征训练回归模型,得到预测结果。结果表明,基于融合特征的模型的预测决定系数Rp2=0.6557,优于基于单一特征的模型。

To address the problems of high cost and portability of traditional apple nondestructive testing methods,different features of RGB images of Fuji apples are used to predict their soluble solids content.The color features,texture features and local features of apple images are extracted by statistical methods and convolutional neural networks,and the prediction results are obtained by splicing the above features and training regression models using fused features.The results show that the prediction coefficient of determina⁃tion R2 p=0.6557 for the model based on fused features is better than that of the model based on single fea⁃tures.