基于语义信息的场景识别方法研究  

Research on Scene Recognition Method Based on Semantic Information

作　　者：杨清溪 张丽红[1] YANG Qingxi;ZHANG Lihong(College of Physical and Electronic Engineering, Shanxi University, Taiyuan 030006, China)

机构地区：[1]山西大学物理电子工程学院,山西太原030006

出　　处：《测试技术学报》2021年第6期521-528,共8页Journal of Test and Measurement Technology

基　　金：山西省研究生教育改革课题资助项目(2019JG031);山西大学物理电子工程学院课程思政建设资助项目(WDKCSZ202009)。

摘　　要：在场景识别任务中,由于场景图像类内变化大,类间相似度高,不同场景类别之间表现出相似的外观和对象分布,从而容易导致场景识别任务的失败.为解决该问题,本文提出一种基于语义分割及高效网络相结合的场景识别模型.该模型由语义分支和RGB分支两部分组成,语义分支在语义分割基础上进一步提取图像上下文信息,RGB分支采用高效网络来提取图像的全局特征,通过注意力机制将两个分支的输出特征进行融合,最终输入线性分类器以实现场景识别的预测.将提出的网络模型在ADE20K,MIT Indoor 67和SUN3973个数据集进行训练与测试,实验结果表明,提出的模型可以显著减少网络参数数量,同时提高场景识别的准确率.In the scene recognition task,due to large changes within the scene image class and high similarity between classes,different scene categories show similar appearance and object distribution,which easily leads to take faliure.To solve this problem,a scene recognition model based on the combination of semantic segmentation and EfficientNet is proposed.The model is composed of semantic branch and RGB branch.On the basis of semantic segmentation,image context information is further extracted by semantic branch.The global features of the image are extracted by an EfficientNet used by RGB branch.The output features of the two branches are merged through the attention mechanism and finally inputted into the linear classifier to realize the prediction of scene recognition.The proposed network model is trained and tested on three data sets ADE20K,MIT Indoor 67 and SUN397.Results show that the number of network parameters can be significantly reduced,and the accuracy of scene recognition can be improved with the model in this paper.

关 键 词：场景识别 语义分割 高效网络 注意力机制 多模态特征融合 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]