基于路径增强SSD的遗失物体检测模型  

Lost Object Detector Based on PA-SSD

作　　者：徐好好 单志勇[2] 徐超 XU Hao-hao;SHAN Zhi-yong;XU Chao(School of Information Science and Technology,Donghua University;Ministry of Education,Digital Textile Research Center,Shanghai 201620,China)

机构地区：[1]东华大学信息科学与技术学院 [2]数字化纺织服装技术教育部工程研究中心,上海201620

出　　处：《软件导刊》2020年第11期17-20,共4页Software Guide

基　　金：国家自然科学基金项目(61602110)。

摘　　要：在日常出行中,乘客经常会将一些重要物品遗落在出租车后座上,而司机往往因为忽视使乘客出现损失。为对车内遗失物体进行检测,提出一种改进的SSD目标检测模型。在主干网络部分引入路径增强的特征金字塔网络(FPN),称为PA-SSD。将PA-SSD应用于常见遗失物品检测实验,结果表明,该算法检测速度为12fps,在验证集上的mAP为98.6%。基于PA-SSD的检测模型能高效且准确地识别乘客遗失物体,方便通知领取,减少乘客不必要的损失。In daily travel,passengers often leave some important items in the back seat of the taxi,and drivers often fail to notice that the loss of these items,which causes passengers property lose.In order to detect the lost objects in the car,this paper proposes an im⁃proved SSD detector which uses path augumented FPN in the backbone and it is called single shot multibox detection with path augu⁃mentation(PA-SSD).PA-SSD is applied to the detection of common lost items.The experimental results show that the detection speed of this detector is 12fps,and the mAP on the verification set is 98.6.PA-SSD can efficiently and accurately identify the lost objects,and it is easy to remind the passengers.

关 键 词：目标检测 卷积神经网络 SSD FPN 路径增强 

分 类 号：TP306[自动化与计算机技术—计算机系统结构]