基于神经网络架构搜索的铭牌目标检测方法  

Object detection for nameplate based on neural architecture search

作　　者：邓渭铭 杨铁军 李纯纯[1] 黄琳[1] DENG Wei-ming;YANG Tie-jun;LI Chun-chun;HUANG Lin(Guangxi Key Laboratory of Embedded Technology and Intelligent System,Guilin University of Technology,Guilin Guangxi 541004,China;College of Intelligent Medicine and Biotechnology,Guilin Medical University,Guilin Guangxi 541199,China)

机构地区：[1]桂林理工大学广西嵌入式技术与智能系统重点实验室,广西桂林541004 [2]桂林医学院智能医学与生物技术学院,广西桂林541199

出　　处：《图学学报》2023年第4期718-727,共10页Journal of Graphics

基　　金：国家自然科学基金项目(62166012,62266015);广西自然科学基金项目(2022GXNSFAA035644);广西嵌入式技术与智能系统重点实验室主任基金项目(2020-1-8)。

摘　　要：为了提高构建深度卷积神经网络(CNN)的自动化程度并进一步提高目标检测精度,提出了一种改进的基于DenseNAS的神经网络架构搜索方法以自动构建铭牌检测CNN。首先,基于改进DenseNAS的Head层,设计了可搜索的、融合深浅层特征的子网模块(CSP-Block1和CSP-Block2)。然后,基于CSP-Block1和CSP-Block2构建的搜索空间,搜索铭牌检测CNN的Backbone和Head。实验结果表明,该方法在一个铭牌5分类的数据集上,耗时约9.35 GPU hours搜索出了最佳神经网络,在测试集上检测精度mAP≈97.3%,比YOLOv5等SOTA方法更高。In order to enhance the automation of building deep convolutional neural network(CNN)for object detection and further improve the detection accuracy,an improved DenseNAS-based neural architecture search method was proposed to automatically build a CNN for nameplate detection.First,the searchable subnet modules(CSP-Block1 and CSP-Block2)were designed to fuse deep and shallow layer feature mapping by enhancing the Head layer of DenseNAS.Subsequently,the search space was established based on the CSP-Block1 and CSP-Block2 to explore the Backbone and Head of CNN for nameplate detection.The experimental results demonstrated that the proposed method required about 9.35 GPU hours to search the optimal neural network on a nameplate dataset consisting of 5 classes,and that the detection accuracy mAP was about 97.3%on the test set,exceeding those of state-of-the-art methods,such as YOLOv5.

关 键 词：神经网络架构搜索 卷积神经网络 CSP结构 铭牌 目标检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]