复杂场景下基于复合胶囊网络的交通标志识别  

Traffic sign recognition based on compound capsule network under background of complex scene

作　　者：陈立潮[1] 张倩茹 曹建芳 潘理虎[1] CHEN Li-chao;ZHANG Qian-ru;CAO Jian-fang;PAN Li-hu(College of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China;Computer Department,Xinzhou Teachers University,Xinzhou 034000,China)

机构地区：[1]太原科技大学计算机科学与技术学院,山西太原030024 [2]忻州师范学院计算机系,山西忻州034000

出　　处：《计算机工程与设计》2021年第9期2627-2633,共7页Computer Engineering and Design

基　　金：山西省自然科学基金项目(201701D121059)。

摘　　要：针对复杂场景中交通标志被遮挡,用单一模型特征提取方法获取的图像信息不充分的问题,提出一种复合胶囊网络的交通标志识别方法。将残差网络中的多尺度思想引入胶囊网络卷积层,在主胶囊层中加入双通道池化,对动态路由算法的计算方式进行优化,提高特征提取的效果和预测值的输出概率。在GTSRB数据集上的测试结果表明,改进后的复合胶囊网络识别精度达到99.21%,相对于其它识别算法性能有所提升,验证了该模型的有效性。Aiming at the problem that the traffic signs are blocked in the complex scenes,and the image information obtained using the single model feature extraction method is insufficient,a method of traffic sign recognition based on compound capsule network was proposed.The multi-scale idea of residual network was introduced into the convolution layer of capsule network,and the dual channel pooling was added to primary capsule layer.The calculation method of dynamic routing algorithm was optimized to improve the effect of feature extraction and the output probability of prediction values.The test results on the GTSRB dataset show that the improved compound capsule network has a recognition accuracy of 99.21%,which improves the perfor-mance compared to other recognition algorithms,thus verifies the effectiveness of the model.

关 键 词：交通标志识别 遮挡图像 胶囊网络 特征提取 动态路由算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]