摘要
多种应用依赖于数据理解的准确性,而语义图像分割有效地解决了这个问题,它为基于像素级别的场景理解提供了必要的上下文信息。鉴于ResNeXt50相比于一般的卷积操作具有更强的特征提取能力,提出了一种基于ResNeXt50的U-Net网络结构ECAU-Net。在融合过程中,通过引入超强通道注意力(ECA)模块进一步增强特征表示对场景分割的判别能力。除此之外,在整体网络结构中引入空洞卷积,在不改变卷积核大小的情况下扩大图像的感受野范围,从而最大化地提高网络性能。实验结果表明,在CamVid数据集上,ECAU-Net相较于U-Net在Acc, Acc class, MIoU和FWIoU这4个评价指标上分别提高了2.1%,8.6%,8.2%和3.2%。
Many applications rely on the accuracy of data understanding,and semantic image segmentation effectively solves this problem,which provides the necessary context information for scene understanding based on pixel level.A U-Net network structure,ECAU-Net,is proposed based on ResNeXt50.Compared with the general convolution operation,ResNeXt50 has a stronger feature extraction ability.In fusion process,Efficient Channel Attention module(ECA-Net) is introduced to further enhance the ability of feature representation to discriminate scene segmentation.In addition,the introduction of dilated convolution in the overall network structure expands the receptive field of the image without changing the size of convolution kernel,thereby maximizing the performance of the network.The experimental results show that on the CamVid dataset,compared with U-Net,ECAU-Net has improved the Acc,Acc class,MIoU and FWIoU respectively by 2.1%,8.6%,8.2% and 3.2%.
作者
王瑞绅
宋公飞
王明
WANG Ruishen;SONG Gongfei;WANG Ming(Nanjing University of Information Science and Technology,School of Automation,Nanjing 210000,China;Nanjing University of Information Science and Technology,Collaborative Innovation Center of Atmospheric Enviroment and Equipment Technology,Nanjing 210000,China;Key Laboratory of Advanced Control and Optimization for Chemical Processes,Shanghai 200000,China)
出处
《电光与控制》
CSCD
北大核心
2023年第1期92-96,102,共6页
Electronics Optics & Control
基金
国家自然科学基金面上项目(61973170)
中央高校基本科研业务费专项资助项目(2020AC0CP02)。
关键词
语义图像分割
空洞卷积
超强通道注意力模块
U-Net
semantic image segmentation
dilated convolution
efficient channel attention module
U-Net
