摘要
针对单步多框目标检测算法(SSD)中存在的误检、漏检以及检测精度不够高等问题,提出了一种改进的SSD目标检测算法。该算法通过空洞卷积替换conv4_3卷积层及之前的两次标准卷积,扩大感受野,使用反卷积对不同尺度的特征图进行融合,使融合形成的特征图具有丰富的上下文信息,最后为特征图添加注意力模型,有效提取感兴趣区域的特征。仿真实验结果表明,改进算法在VOC2007数据集上较原算法检测精度提升0.9%,检测结果更加准确,一定程度上改善了误检、漏检等问题,同时仍满足实时性的要求。
Aiming at the problems of false detection,missing detection and low detection accuracy in single shot multibox detector(SSD),an improved SSD object detection algorithm is proposed.The conv4_3 convolution layer and the previous two standard convolutions are replaced by dilated convolution to expand the receptive field.The deconvolution is used to fuse the feature maps of different scales,so that the feature maps formed by fusion contain rich context information.The attention model is added to the feature map to effectively extract the features of regions of interest.Simulation results show that the detection accuracy of the improved algorithm is 0.9%higher than that of the original algorithm,and the detection effect is better.To some extent,it solves the problems of false detection and missing detection,and still meets the requirements of real-time.
作者
王燕妮
刘祥
刘江
WANG Yanni;LIU Xiang;LIU Jiang(School of Information and Control Engineering,Xi’an University of Architecture and Technology,Xi’an 710055,China;Xi’an Modern Control Technology Research Institute,Xi’an 710065,China)
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2021年第4期145-153,共9页
Journal of Xi'an Jiaotong University
基金
陕西省自然科学基础研究资助项目(2020JM-499,2020JQ-684)。
关键词
目标检测
单步多框目标检测算法
空洞卷积
反卷积
注意力机制
object detection
single shot multibox detector
dilated convolution
deconvolution
attention mechanism