基于改进SSD的被遮挡车辆目标检测算法研究
Research on occluded vehicle detection algorithm based on improved SSD
朱雅琪 1王洪亮1
作者信息
- 1. 南京理工大学机械工程学院,江苏 南京 210094
- 折叠
摘要
针对复杂环境下被遮挡车辆目标检测能力差的问题,提出一种基于改进SSD的车辆目标检测算法.替换骨干网络为ResNet50 网络,利用残差学习的思想获得更好的模型训练效果,引入特征融合模块增加网络的语义信息并增加注意力机制对重要特征进行强调,以提高被遮挡车辆目标的检测能力,利用Focal loss损失函数解决模型训练时正负样本失衡问题.实验结果表明,所提算法检测精度更高,同时对于被遮挡车辆目标的检测效果也较好.
Abstract
In order to solve the problem of poor detection ability of occluded vehicle targets in complex environment,an improved SSD(Single Shot MultiBox Detector)based vehicle target detection algorithm was proposed.The backbone network was replaced with ResNet50 network,and the idea of residual learning was used to obtain better model training effect.In order to improve the detection ability of occluded vehicle target,the feature fusion module was introduced to increase the semantic information of the network and the attention mechanism was added to emphasize important features.The Focal loss function was used to solve the problem of positive and negative sample imbalance during model training.The experimental results showed that the proposed algorithm has higher detection accuracy and better detection effect on partially obscured vehicle targets.
关键词
深度学习/遮挡车辆/特征融合/注意力机制Key words
deep learning/occluded vehicle/feature fusion/attention mechanism引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据