基于特征融合的残差卷积网络对交通标识识别  

Traffic Sign Recognition Based on Feature Fusion Residual Convolution Network

作　　者：陈春玮 欧浩源[1] Chen Chunwei;Ou Haoyuan(Guangdong Polytechnic college,Foshan 528000,China)

机构地区：[1]广东职业技术学院,广东佛山528000

出　　处：《科学技术创新》2022年第4期69-72,共4页Scientific and Technological Innovation

基　　金：广东职业技术学院2020年度教学创新改革资助项目“人工智能背景下高职院校的计算机类课程教学改革的研究”(编号:300455);2018年广东省普通高校青年创新人才类项目,项目编号:2018GkQNCX016。

摘　　要：近些年深度卷积神经网络在图像识别上的学习能力和应用效果引起业界和学术界广泛重视,本文在自动驾驶日益突出的背景下,提出一种基于特征融合上采样的残差卷积神经网络的模型,经过与传统的卷积神经网络和残差网络进行准确率的对比,它在德国交通数据集GTSDB上性能和准确率上表现出更有优势。经过实验证明特征融合的残差卷积的网络表达能力进一步提高了。In recent years, the learning ability and application effects of deep neural networks in image recognition have attracted widespread absorption by the public and academia. Under the background of automatic driving becoming more and more prominent, a Upsamplingconvolutional neural network model based on feature fusion is proposed. After comparing the accuracy with the traditional neural network model and residual network, it shows advantages in performance and accuracy on the German traffic data set GTSDB. The experiment proves that the network expression ability of feature fusion is further improved.

关 键 词：卷积神经网络 残差 特征融合 交通标识 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]