STDC-MA network for semantic segmentation

作     者：Lei, Xiaochun Lu, Linjun Jiang, Zetao Gong, Zhaoting Lu, Chang Liang, Jiaming Xie, Junlin 

作者机构：Guilin Univ Elect Technol GUET Sch Comp Sci & Informat Secur Guilin 541004 Guangxi Peoples R China Guilin Univ Elect Technol GUET Guangxi Key Lab Image & Graph Intelligent Proc Guilin 541004 Guangxi Peoples R China Guilin Univ Elect Technol GUET Sch Comp Sci & Informat Secur Guilin Guangxi Peoples R China 

出 版 物：《IET IMAGE PROCESSING》 (IET Image Proc.)

年 卷 期：2022年第16卷第14期

页      面：3758-3767页

核心收录：

学科分类：0808[工学-电气工程] 1002[医学-临床医学] 08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：Guangxi Key Laboratory of Image and Graphic Intelligent Processing [GIIP2004] Sichuan Regional Innovation Cooperation Project [2021YFQ0002] National Natural Science Foundation of China [61876049, 62172118] Student's Platform for Innovation and Entrepreneurship Training Program [202010595168, 202110595025, S202110595168] Nature Science key Foundation of Guangxi [2021GXNSFDA196002] 

主　　题：segmentation speed STDC-MA network highly demand segmentation accuracy intelligent transportation geophysical image processing semantic segmentation image representation high-level features lightweight structure low-level features semantic information feature extraction STDC-Seg network deep learning (artificial intelligence) image segmentation STDC-Seg structure efficient structure hierarchical multiscale attention mechanism high-level feature map autonomous driving image classification feature alignment module image resolution 

摘      要：Semantic segmentation is applied extensively in autonomous driving and intelligent transportation with methods that highly demand spatial and semantic information. Here, an STDC-MA network is proposed to meet these demands. First, the STDC-Seg structure is employed in STDC-MA to ensure a lightweight and efficient structure. Subsequently, the feature alignment module is applied to understand the offset between high-level and low-level features, solving the problem of pixel offset related to upsampling on the high-level feature map. The approach implements the effective fusion between high-level features and low-level features. A hierarchical multiscale attention mechanism is adopted to reveal the relationship among attention regions from two different input sizes of one image. Through this relationship, regions receiving much attention are integrated into the segmentation results, thereby reducing the unfocused regions of the input image and improving the effective utilisation of multiscale information. STDC-MA maintains the segmentation speed as the STDC-Seg network while improving the segmentation accuracy of small objects. STDC-MA was verified on the validation dataset of Cityscapes. The segmentation result of STDC-MA attained 78.32% mIOU with the input of 0.5x scale, 4.92% higher than STDC-Seg.