摘要
针对当前检测算法很难在资源有限的硬件平台上运行且小目标检测率不高等问题,提出一种基于改进SSD轻量化的车辆检测方法。引入新型轻量化网络MobileNetV3、重新设计特征层交叉融合以及更改损失函数的方法使算法在检测速度上明显提高,且在改进特征提取层和损失函数后,算法的均值平均精度(mAP)提高了2.4%,能够满足在检测识别的任务中对检测准确率和实时性的要求。
Aiming at the problems of current detection algorithms are difficult to run on hardware platform with limited resources and low detection rate of small targets, a vehicle detection method based on improved SSD lightweight is proposed.By introducing new lightweight network, MobileNetV3,redesigning feature layer cross fusion and changing loss function, the detection speed of the algorithm is significantly improved.After improving the feature extraction layer and loss function, mean average precision(mAP)of the algorithm increases by 2.4 %,which can meet the requirements of detection accuracy and real-time in the task of detection and recognition.
作者
陈家栋
雷斌
CHEN Jiadong;LEI Bin(College of Machinery and Automation,Wuhan University of Science and Technology,Wuhan 430081,China;Institute of Robot Intelligent System,Wuhan University of Science and Technology,Wuhan 430081,China)
出处
《传感器与微系统》
CSCD
北大核心
2022年第10期117-121,126,共6页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(61305110)。
