| 基于多路径网络的权值调整图像语义分割算法 |
| |
| 引用本文： | 秦晓飞,何玉帅,孙越,严浩通,林轩.基于多路径网络的权值调整图像语义分割算法[J].光学仪器,2020,42(1):46-51. |
| |
| 作者姓名： | 秦晓飞 何玉帅 孙越 严浩通 林轩 |
| |
| 作者单位： | 上海理工大学 光电信息与计算机工程学院, 上海 200093,上海理工大学 机械工程学院, 上海 200093,上海理工大学 机械工程学院, 上海 200093,上海理工大学 机械工程学院, 上海 200093,上海理工大学 机械工程学院, 上海 200093 |
| |
| 基金项目： | 国家重点研究发展计划（2016YFF0101400） |
| |
| 摘 要： | 为提高图像语义分割准确程度,针对场景解析中类别边缘分辨清晰度,提出了一种基于多路径网络的权值调整图像语义分割算法。通过引入多路径网络和权值调整并对图像场景中的物体类别具有的特征进行分析,提高图像的语义分割的准确程度;通过采用ADE20K数据集进行训练,提高边缘信息的分割效果,使模型具有更好的泛化能力。此算法加快了网络收敛速度。
|
| 关 键 词： | 图像语义分割 多路径网络 权值调整 边缘信息 |
| 收稿时间： | 2019/4/2 0:00:00 |
Re-coding semantic image segmentation method based on multi-path network |
| |
| QIN Xiaofei,HE Yushuai,SUN Yue,YAN Haotong and LIN Xuan.Re-coding semantic image segmentation method based on multi-path network[J].Optical Instruments,2020,42(1):46-51. |
| |
| Authors: | QIN Xiaofei HE Yushuai SUN Yue YAN Haotong and LIN Xuan |
| |
| Affiliation: | School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China,School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China,School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China,School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China and School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China |
| |
| Abstract: | In order to improve the accuracy of semantic image segmentation, a semantic segmentation algorithm based on the multi-path network is proposed for the resolution of category edges in scene analysis. By analyzing the characteristics of object classes in image scenes, the idea of the multi-path network and weight adjustment is introduced to improve the accuracy of the algorithm. This paper uses the ADE20K dataset training to significantly improve the segmentation effect. The experimental verification of the public ADE20K dataset proves that compared with the popular image semantic segmentation algorithm, the algorithm has good model convergence ability and generalization ability, and the image edge information segmentation is significantly improved. |
| |
| Keywords: | semantic image segmentation multi-path network weight redistribution edge information |
| 本文献已被 CNKI 维普 等数据库收录！ |
| 点击此处可从《光学仪器》浏览原始摘要信息 |
|
点击此处可从《光学仪器》下载全文 |
|
