检索规则说明:AND代表“并且”;OR代表“或者”;NOT代表“不包含”;(注意必须大写,运算符两边需空一格)
检 索 范 例 :范例一: (K=图书馆学 OR K=情报学) AND A=范并思 范例二:J=计算机应用与软件 AND (U=C++ OR U=Basic) NOT M=Visual
作 者:杜云龙 强俊 王洪铭 肖光磊 孙宇 DU Yunlong;QIANG Jun;WANG Hongming;XIAO Guanglei;SUN Yu(School of Computer and Information,Anhui Polytechnic University,Anhui Wuhu 241000,China)
机构地区:[1]安徽工程大学计算机与信息学院,安徽芜湖241000
出 处:《重庆工商大学学报(自然科学版)》2025年第1期64-71,共8页Journal of Chongqing Technology and Business University:Natural Science Edition
基 金:安徽省高校优秀拔尖人才培育资助项目(GXYQZD2021123);国家级大学生创新创业计划项目(S202110363098).
摘 要:目的针对自然场景下交通标志检测存在的小目标精度低和目标特征信息不足等问题,提出一种使用残差网络(Residual Network,ResNet)和注意力机制(Attention Mechanism)的单步多目标检测算法SSD(Single Shot MultiBox Detector),经过残差网络和注意力机制提取的特征向量输送到一个轻量、高效的特征融合模块中,最后将输出的feature map送到检测器中进行检测,从而提升交通标志检测的正确率。方法首先,利用残差模块将特征进行1×1降维再3×3升维,然后将恒等映射和残差部分生成的特征图进行逐像素相加;其次,将CBAM(Convolutional Block Attention Module)引入到残差模块Conv4_x输出的特征图上,然后与残差模块Conv2_x,Conv3_x输出的特征图一起输入到高效的特征融合模块中进行特征融合,最后将融合后的特征图送入模型中检测以实现对交通标志的识别。结果通过仿真实验验证,改进后的单步多目标检测算法SSD在中国交通标志检测数据集上进行检测的平均精度为90.55%,能够有效地提取小目标特征的信息。相较于主流算法CenterNet、YOLOv3、YOLOv4、Faster RCNN、SSD分别提高了2.57%、3.4%、2.79%、3.8%、4.93%。结论优化后的目标检测方法相较于其他检测方法提取到了更多的特征信息,达到了更高的检测精度,在交通标志检测中具有良好的实用性和有效性。Objective In response to the issues of low accuracy in detecting small objects and insufficient target feature information in natural scenes for traffic sign detection,a single shot multibox detector(SSD)algorithm using residual network(ResNet)and attention mechanism was proposed.The feature vectors extracted by residual networks and attention mechanisms were fed into a lightweight and efficient feature fusion module.Finally,the output feature map was sent to the detector for detection,thereby enhancing the accuracy of traffic sign detection.Methods Firstly,the features were dimensionally reduced by 1x1 and then increased by 3x3 using residual modules,and then the feature maps generated by the constant mapping and residual parts were summed pixel by pixel.Secondly,the convolutional block attention module(CBAM)was introduced to the feature map output by Conv4_x of the residual module.Then,the feature map output by the residual module Conv4_x and the feature maps output by the residual modules Conv2_x and Conv3_x were fed into the efficient feature fusion module for feature fusion.Finally,the fused feature map was sent to the model for detection to realize the recognition of traffic signs.Results Through simulation experiments,the improved SSD algorithm achieved an average precision of 90.55% for detection on the Chinese traffic sign detection dataset,effectively extracting feature information from small objects.Compared with mainstream algorithms including CenterNet,YOLOv3,YOLOv4,Faster RCNN,and SSD,the improved SSD algorithm improved the accuracy by 2.57%,3.4%,2.79%,3.8%,and 4.93%,respectively.Conclusion The optimized object detection method extracts more feature information and achieves higher detection accuracy compared with other detection methods,demonstrating good practicality and effectiveness in traffic sign detection.
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在链接到云南高校图书馆文献保障联盟下载...
云南高校图书馆联盟文献共享服务平台 版权所有©
您的IP:216.73.216.38