城市街景图像语义分割方法研究  

Research on semantic segmentation method of cityscapes image

作　　者：姜长泓 梁奇强 李秀华[1] 梁超[1] JIANG Changhong;LIANG Qiqiang;LI Xiuhua;LIANG Chao(School of Computer Science&Engineering,Changchun University of Technology,Changchun 130012,China)

机构地区：[1]长春工业大学计算机科学与工程学院,吉林长春130012

出　　处：《长春工业大学学报》2022年第4期447-453,共7页Journal of Changchun University of Technology

基　　金：吉林省科技发展计划重点研发项目(20200403037SF)。

摘　　要：针对Deeplabv3+网络在语义分割任务中存在计算量大、训练速度慢的问题,使用改进后的轻量级卷积神经网络Xception作为骨干特征提取网络,进而极大地减少了模型参数数量;针对Deeplabv3+网络在城市街景图像上存在对小目标分割效果较差的问题,引入CBAM注意力机制,增强目标区域的权重比例,从而有效地提高网络性能。最后使用Cityscapes数据集进行实验,并与几种常见的分割方法做了对比研究。Aiming at the problems of large computation and slow training speed in the semantic segmentation task of Deeplabv3+network,an improved lightweight convolutional neural network Xception was used as the backbone feature extraction network,which greatly reduced the number of model parameters.Aiming at the problem that Deeplabv3+network has poor segmentation effect on small targets in urban street view images,CBAM attention mechanism is introduced to enhance the weight ratio of target regions,thus effectively improving the performance of the network.Finally,Cityscapes data set is used for experiments,and compared with several common segmentation methods.

关 键 词：语义分割 轻量级卷积神经网络 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]