基于注意力的多尺度水下图像增强网络  

Multi-scale Underwater Image Enhancement Network Based on Attention Mechanism

作　　者：方明[1,2] 刘小晗 付飞蚺[1] FANG Ming;LIU Xiaohan;FU Feiran(School of Computer Science and Technology,Changchun University of Science and Technology,Changchun 130022,China;School of Artificial Intelligence,Changchun University of Science and Technology,Changchun 130022,China)

机构地区：[1]长春理工大学计算机科学技术学院,长春130022 [2]长春理工大学人工智能学院,长春130022

出　　处：《电子与信息学报》2021年第12期3513-3521,共9页Journal of Electronics & Information Technology

基　　金：山东省支持青岛海洋科学与技术试点国家实验室重大科技专项(2018SDKJ0102-6)。

摘　　要：水下图像往往会因为光的吸收和散射而出现颜色退化与细节模糊的现象,进而影响水下视觉任务。该文通过水下成像模型合成更接近水下图像的数据集,以端到端的方式设计了一个基于注意力的多尺度水下图像增强网络。在该网络中引入像素和通道注意力机制,并设计了一个多尺度特征提取模块,在网络开始阶段提取不同层次的特征,通过带跳跃连接的卷积层和注意力模块后得到输出结果。多个数据集上的实验结果表明,该方法在处理合成水下图像和真实水下图像时都能有很好的效果,与现有方法相比能更好地恢复图像颜色和纹理细节。Due to the absorption and scattering,color degradation and detail blurring often occur in underwater images,which will affect the underwater visual tasks.A multi-scale underwater image enhancement network based on attention mechanism is designed in an end-to-end manner by synthesizing dataset closer to underwater images through underwater imaging model.In the network,pixel and channel attention mechanisms are introduced.A new multi-scale feature extraction module is designed to extract the features of different levels at the beginning of the network,and the output results are obtained via a convolution layer and an attention module with skip connections.Experimental results on multiple datasets show that the proposed method is effective in processing both synthetic and real underwater images.It can better recover the color and texture details of images compared with the existing methods.

关 键 词：水下图像增强 深度学习 注意力机制 多尺度特征 

分 类 号：TN911.73[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]