一种基于改进PSPNet的城市街景语义分割方法  

An improved PSPNet-Based Semantic Segmentation Method for Urban Street Scenes

作　　者：叶波 潘硕 李景文 姜建武 YE Bo;PAN Shuo;LI Jing-wen;JIANG Jian-wu(College of Geomatics and Geoinformation,Gunlin University of Technology,Guilin Guangxi,541006,China;Guangxi Key Laboratory of Intelligent Sensing Service of Ecological Spatial and Temporal Big Data,Guilin Guangxi 541006,China;Guangxi Institute of Industrial Technology,Nanning Guangxi 530200,China)

机构地区：[1]桂林理工大学测绘地理信息学院,广西桂林541006 [2]广西生态时空大数据智能感知服务重点实验室,广西桂林541006 [3]广西产业技术研究院,广西南宁530200

出　　处：《计算机仿真》2024年第9期105-110,120,共7页Computer Simulation

基　　金：国家自然科学基金(41961063);广西高校中青年教师科研基础能力提升项目(2022KY0250)。

摘　　要：在当前街景语义分割任务过程中,传统的语义分割方法容易产生事物边缘分割不精准以及无关背景特征影响严重的问题。针对这一问题,基于PSPNet模型提出一种融合语义特征和边缘特征的语义分割网络E-PSPNet。模型由语义分割子网络和边缘检测子网络构成,在语义分割子网络中嵌入注意力机制,加强有效特征的获取,忽略无关背景特征信息,并利用边缘检测子网络获取到更准确的轮廓特征,最后通过特征融合模块对两种特征进行融合得到最终结果。在Cityscapes数据集上进行消融实验,该模型平均交并比对比原模型提升了3.3%,并与现有模型进行实验对比,实验结果证明E-PSPNet模型可以有效改善街景边缘分割不精准以及背景无关特征影响严重的问题。In the current process of streetscape semantic segmentation task,the traditional semantic segmentation methods are prone to the problems of imprecise edge segmentation and serious influence of irrelevant background features.The model consists of a semantic segmentation sub-network and an edge detection sub-network,in which an attention mechanism is embedded in the semantic segmentation sub-network to enhance the acquisition of effective features and ignore irrelevant feature information,and the edge detection sub-network is used to obtain more accurate contour features.Finally,the final result is obtained by fusing the two features through the feature fusion module.Some ablation experiments were conducted on the Cityscapes dataset,and the model improved the average delivery by 3.3% compared to the original model of the comparison,and the experimental comparison with the existing model,and the experimental results proved that the E-PSPNet model can effectively improve the problem of imprecise edge segmentation of streetscape and the serious influence of background irrelevant features.

关 键 词：语义分割 边缘检测 注意力机制模块 城市街景 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]