计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (18): 147-153.DOI: 10.3778/j.issn.1002-8331.1802-0090

• 模式识别与人工智能 • 上一篇    下一篇

受损图片分类的双并行交叉降噪卷积神经网络

王  达1,2,周雪峰2,徐智浩2,蔡奕松2,陈  洲1,2   

  1. 1.沈阳工业大学 信息科学与工程学院,沈阳 110870
    2.广东省智能制造研究所 广东省现代控制技术重点实验室,广州 510070
  • 出版日期:2018-09-15 发布日期:2018-10-16

Double parallel cross denoising convolution neural network for damaged image classification

WANG Da1,2, ZHOU Xuefeng2, XU Zhihao2, CAI Yisong2, CHEN Zhou1,2   

  1. 1.School of Information Science and Engineering, Shenyang University of Technology, Shenyang 110870, China
    2.Guangdong Key Laboratory of Modern Control Technology, Guangdong Institute of Intelligent Manufacturing, Guangzhou 510070, China
  • Online:2018-09-15 Published:2018-10-16

摘要: 应对大量受干扰图像的分类问题,提出了一种双并行交叉降噪卷积模型,该模型由两部分并行交叉网络结构组成,分别对应改进的自编码方式和并行交叉卷积神经网络,同时在该模型训练的过程中,使用批量正则化和改进激活函数的方法。经实验验证,与同类模型相比,该模型首先具有降噪能力强、鲁棒性好、泛化能力强和准确率高的特点,其次避免过拟合,加快收敛速度。在图片不同程度受损的情况下,它也可较好地完成图像目标识别分类任务。

关键词: 图像分类, 自编码器, 卷积神经网络, 双并行交叉

Abstract: In order to deal with the classification problem of a large number of disturbed images, a double parallel crossover noise reduction convolution model is proposed. The model consists of two parts of parallel cross-network structure, corresponding to improved self-coding and parallel cross-convolutional neural network. At the same time, the model uses the method of batch normalization and improvement of traditional activation function in the model training. It is proved by experiments that the model has the characteristics of strong noise reduction, good robustness, strong generalization ability and high accuracy in comparison with the same model. Secondly, it avoids overfitting and speeds up convergence. In the case of different degrees of picture damage, it can also accomplish the task of image target recognition classification well.

Key words: image classification, autoencoder, convolution neural networks, double parallel cross