摘要
为解决YOLOXs行人目标识别算法效率低、体积规模大的问题,采用轻量架构卷积神经网络替换主干网络,深度可分离卷积替换,特征增强网络中的3×3卷积层,提出轻量化增强网络,并应用于智能汽车行人目标识别试验。结果表明,轻量化增强网络模型能够在确保行人目标识别精度的基础上有效缩减参数量和所占内存,对参数缩减了44.1%,所占内存缩减了41.9%,该算法更适合于嵌入式与移动端设备的搭建,对智能汽车的开发具有一定的参考价值。
To solve the problem of low efficiency and large size of YOLOXs pedestrian target recognition algorithm.A lightweight convolutional neural network is used to replace the backbone network,and the 3×3 convolutional layer in the feature enhancement network is replaced by a lightweight convolutional neural network.The results show that the lightweight enhanced network model can effectively reduce the number of parameters and memory on the basis of ensuring the accuracy of pedestrian target recognition.The parameters are reduced by 44.1%and the memory is reduced by 41.9%.This is more suitable for the construction of embedded and mobile devices,and has certain reference value for the development of intelligent vehicles.
作者
谭超
朱荣钊
TAN Chao;ZHU Rongzhao(Information Technology College,Xiamen Huatian International Vocation Institute,Xiamen 361102,China;School of Computer and Information Engineering,Hubei University,Wuhan 430062,China)
出处
《安阳师范学院学报》
2024年第2期29-34,共6页
Journal of Anyang Normal University
基金
福建省教育厅课题研究项目(项目编号:JA13457)。
