新型稀疏自动编码器组合的深度学习方法

针对自动编码器在强噪声环境下分类效果低的特征,提出了基于改进型稀疏自动编码器组合的深度学习方法。在采用计算相关熵的方法,增强了稀疏自动编码器对非高斯噪声的鲁棒性的基础上,利用卷积神经网络对自动编码器进行边缘降噪,接着将改进后的稀疏自动编码器和边缘降噪自动编码器相结合,得到新的稀疏边缘降噪自动编码器。实测数据的实验结果表明,新的稀疏边缘降噪自动编码器比现有的分类算法,计算时间更短、准确率更高、效果更明显。

Deep Learning Method Based on Improved Sparse Automatic-Encoder Combination

Since the poor classification results of traditional auto-encoders when applied to datasets with strong noise, we proposed a new method which is a combination of spare auto-encoder and deeplearning technique. On the basis that the spare auto-encoder’s robustness against non-Gaussian noise is increased by computing relative entropy, a convolutional neural network was used to reduce its marginal noise. Then combining the improved spare auto-encoder with marginalized denoising auto-encoder, the new spare marginalized denoising auto-encoder was obtained. The testing results show that our method has less runtime and higher accuracy than traditional ways.