一种面向识别的无监督特征学习算法

特征抽取是图像识别的关键环节，准确的特征表达能够产生更准确的分类效果。采用软阈值编码器和正交匹配追踪(OMP)算法正交化视觉词典的方法，以提高单级计算结构的识别率，并进一步构造两级计算结构，获取图像更准确的特征，以提高图像的识别率。实验表明，采用软阈值编码器和OMP算法能提高单级计算结构提取特征的能力，提高大样本数据集中图像的识别率。两级计算结构能够提高自选数据集中图像的识别率。采用OMP算法能提高VOC2012数据中图像的识别率。在自选数据集上，两级计算结构优于单级计算结构，与NIN结构相比表现出优势，与卷积神经网络CNN相当，说明两级计算结构在自选数据集上有很好的适应性。

A recognition-oriented unsupervised feature learning algorithm

Feature extraction is a key part of image recognition, and precise feature expression can generate more accurate classification. We improve the recognition rate of the single stage computational structure by adopting soft threshold encoder and the orthogonalizing visual dictionary of the orthogonal matching pursuit (OMP) algorithm. Besides, we build a two stage computational structure which extracts images’ features and increases the recognition rate. Experiments demonstrate that adopting soft threshold encoder and the OMP algorithm can increase the ability of extracting features of the single stage computational structure and enhance image recognition rate in big sample datasets. The two stage computational structure can improve recognition rate on self selection datasets. The OMP algorithm can improve recognition rate of the VOC2012 dataset. For self selection datasets, the two stage computational structure outperforms the single stage computational structure and network in network (NIN), and is equivalent to convolutional neural networks (CNN), indicating that the two stage computational structure is adaptive to self selection datasets.