摘要
在计算机视觉领域中,图像分割技术为基础的部分之一,图像分割的精确与否对后续图像处理存在重要的影响。对全卷积网络进行改进,将卷积网络、多尺度特征提取和空洞卷积进行结合,将不同卷积层抓取到的具有更强描述力的特征进行融合,并且增加网络的层数来提升网络的泛化能力。对网络在CamVid数据集上进行测试,其测试结果与FCN-16s的测试结果进行对比,证明了网络的精确度有所提升。
In the field of computer vision pne of the basic parts of image segmentation technology,the accuracy of image segmentation has an important impact on subsequent image processing.In this paper,the whole convolutional network is improved by combining the convolutional network,multi-scale feature extraction and hole convolution.Also,different convolutional layers are captured to merge with more descriptive features in which the number of layers of the network is increased.In addition,we also improve the generalization of the network.We conduct the experiments based on the CamVid dataset.The test results are compared with the test results of FCN-16 s,which proves that the accuracy of the network is improved.
作者
李轩
孙昕楠
LI Xuan;SUN Xin-nan(School of Electronic and Information Engineering,Shenyang Aerospace University,Shenyang 110136,China)
出处
《沈阳航空航天大学学报》
2020年第1期50-57,共8页
Journal of Shenyang Aerospace University
作者简介
李轩(1967-),男,吉林东丰人,副教授,博士,主要研究方向:信号处理与检测技术、图像处理,E-mail:1041291632@qq.com。
