摘要
针对基于深度学习的DeepLabV3+语义分割算法在编码特征提取阶段大量细节信息被丢失,导致其在物体边缘部分分割效果不佳的问题,本文提出了基于DeepLabV3+与超像素优化的语义分割算法。首先,使用DeepLabV3+模型提取图像语义特征并得到粗糙的语义分割结果;然后,使用SLIC超像素分割算法将输入图像分割成超像素图像;最后,融合高层抽象的语义特征和超像素的细节信息,得到边缘优化的语义分割结果。在PASCAL VOC 2O12数据集上的实验表明,相比较DeepLabV3+语义分割算法,本文算法在物体边缘等细节部分有着更好的语义分割性能,其mIoU值达到83.8%,性能得到显著提高并达到了目前领先的水平。
To tackle the problem where by DeepLabV3+loses considerable detail information during feature extraction,which leads to poor segmentation results in the edges of the objects,this study proposed a semantics segmentation algorithm based on DeepLabV3+and optimized by superpixels.First,a DeepLabV3+model was chosen to extract semantic features and obtain coarse semantic segmentation results.Then,the simple linear iterative clustering algorithm was used to segment the input image into superpixels.Finally,high-level abstract semantic features and detailed information of the superpixels were fused to obtain edge optimized semantic segmentation results.Experiments conducted on the PASCAL VOC 2O12 dataset show that compared to DeepLabV3+,the proposed algorithm had superior performance in terms of detail parts such as edges of objects,and the value of mIoU reached 83.8%.The proposed algorithm thus outperformed other state-of-the-art algorithms in terms of semantic segmentation.
作者
任凤雷
何昕
魏仲慧
吕游
李沐雨
REN Feng-lei;HE Xin;WEI Zhong-hui;LU You;LI Mu-yu(Changchun Institute of Optics, Fine Mechanics and Physics,Chinese Academy of Sciences,Changchun 130033, China;University of Chinese Academy of Sciences, Beijing 100049, China)
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2019年第12期2722-2729,共8页
Optics and Precision Engineering
基金
吉林省科技发展计划资助项目(No.20180201013GX)