摘要
传统图像去噪模型一般为浅层线性结构,特征提取能力有限,而现有基于深度学习的图像去噪模型存在去噪效率低、泛化能力弱等问题。针对上述缺点,以残差块、批归一化层和自编码器组成的残差卷积自编码块为基本网络结构,提出多功能去噪残差卷积自编码神经网络(DRCAENNm)和去噪残差卷积自编码神经网络(DRCAENN)两种基于深度学习的去噪网络模型。实验结果表明,DRCAENNm模型不仅具有盲去噪能力,还可以去除与训练数据类型不同的噪声,具有强泛化能力;DRCAENN模型收敛快,去噪效率远超其它网络模型。
The traditional image denoising models are generally shallow linear structures with limited feature extraction capabilities. However, existing image denoising models based on deep learning have the problems such as low denoising efficiency and weak generalization ability. In view of the above shortcomings, combined with the advantages of Convolutional Auto-Encoder Network and Residual Network, using the Residual Convolutional Auto-Encoder consisting of residual block, Batch Normalization layer and Auto-Encoder as the basic network structure, multi-functional denoising residual convolution Auto-Encoding neural network(DRCAENNm) and denoising residual convolution Auto-Encoding Neural network(DRCAENN) are proposed. The experimental results show that the DRCAENNm model not only has the ability to blindly denoise, but also can filter out the noises different from the training data type;DRCAENN model can converge faster and the denoising efficiency far exceeds other network models.
作者
罗仁泽
王瑞杰
张可
范顺利
LUO Ren-ze;WANG Rui-jie;ZHANG Ke;FANG Shun-li(State Key Laboratory of Oil and Gas Reservoir Geology and Development Engineering,Southwest Petroleum University,Chengdu Sichuan 610500,China;School of electrical information,Southwest Petroleum University,Chengdu Sichuan 610500,China;School of Earth Science and Technology,Southwest Petroleum University,Chengdu Sichuan 610500,China)
出处
《计算机仿真》
北大核心
2021年第5期455-461,共7页
Computer Simulation
基金
国家重点研发计划深地专项项目(2016YFC0601100)
四川省科技计划项目(2019CXRC0027)。
关键词
深度学习
图像去噪
卷积神经网络
自编码网络
残差网络
Deep learning
Image denoising
Convolutional neural network(CNN)
Auto-encoder(AE)
Residual network
