基于深度学习和稀疏表示的分类算法研究

Research on Classification Algorithm Based on Deep Learning and Sparse Representation

本文采用了一种基于深度学习的稀疏表示分类方法。该网络由一个带有卷积层的自编码器和一个全连接层组成。编码器网络的作用是学习图像鲁棒的深层特征,解码器的任务是进行分类。同时,位于编码器和解码器网络之间的稀疏编码层用于寻找来自编码器深层特征的稀疏表示,然后使用对测试集预测的稀疏编码进行分类。本文对传统的稀疏表示分类网络(SRC)以及结合深度学习的稀疏表示分类网络(DSRC)在UMD移动人脸数据集(UMDAA-01)上进行实验,实验结果表明深度学习与稀疏表示相结合的分类效果优于传统的SRC方法。

This paper proposes a sparse representation classification method based on deep learning. The network consists of an Auto-encoder with convolution and a fully connected layer. The role of the encoder network is to learn robust deep and abstract features of images, and the role of the decoder is to classify. Meanwhile, the sparse coding layer between the encoder and the decoder network is used for finding the sparse representation for these features from encoders. The sparse coding of the test set is then used for classification. In this paper, experiments on the traditional Sparse Rep-resentation Classification Network (SRC) and the Sparse Representation Classification Network (DSRC)combined with Deep Learning were conducted on the UMD mobile face dataset (UMDAA-01), and the results show that the combination of deep learning and sparse representation is better than the traditional SRC method.