基于拉普拉斯卷积网络和SMM分类器的小麦麦粒识别

Wheat kernel recognition based on Laplacian convolution network and SMM classifier

为了提高小麦麦粒识别的识别率,采用了拉普拉斯卷积网络(Convolution Network Based on Laplacian Eigenmap,LENet)和支持矩阵机(Support Matrix Machines,SMM)分类器相结合的方法对小麦麦粒进行识别。拉普拉斯卷积网络是一种无反馈的轻量型级联卷积神经网络,可以用来提取小麦麦粒的特征,该网络通过拉普拉斯特征映射来学习网络的参数,输出层通过块直方图编码和矩阵化处理实现,最终提取的特征使用SMM分类器进行分类。通过在建立的小麦麦粒图像数据库上的实验表明,该麦粒识别方法要优于一些传统特征提取分类方法,取得了较好的识别效果。

In order to improve the recognition rate of the recognition of the wheat kernel,Convolution Network Based on Laplacian Eigenmap（ LENet） and Support Matrix Machines（ SMM） were used for identification of wheat kernel. LENet is a convolutional network features lightweight concatenated convolutional neural networks without feedback that can be used to extract the wheat kernel and automatically learn the parameters of the network by Laplacian eigenmap feature mapping,the network＇s output layer can be realized by block histogram encoding and matrix processing. The acquired features were sent to SMM classifier for training and recognition. Experiments on the wheat grain image database show that the adopted method is better than some traditional feature extraction methods,and achieves better recognition results.