基于注意机制的双分支黑烟车辆检测网络  

Dual Branch Network for Black Smoke and Vehicle Detection Based on Attention Mechanism

作　　者：郭韬远 任明武[1,2] GUO Taoyuan;REN Mingwu(School of Computer Science and Engineering,Nanjing University of Science&Technology,Nanjing 210094;MOE Key Laboratory of Intelligent Perception and System for High-Dimensional Information,Nanjing University of Science&Technology,Nanjing 210094)

机构地区：[1]南京理工大学计算机科学与工程学院,南京210094 [2]南京理工大学高维信息智能感知与系统教育部重点实验室,南京210094

出　　处：《计算机与数字工程》2022年第1期147-151,共5页Computer & Digital Engineering

摘　　要：当前交通道路中,存在许多有毒尾气排放超标的车辆,严重污染空气、损害人体健康。目前黑烟车辆检测多采用人工方法或者基于手工特征提取的传统机器学习方法,耗力耗时且难以全面实时监控。论文率先将基于卷积神经网络的目标检测框架CenterNet作为视频监控交通场景下黑烟车辆检测的基本解决方案,并针对实验结果进一步改进上述结构,提出基于注意机制的双分支黑烟车辆检测网络,使用双主干网络提取有针对性的特征表示,对于双主干网络的特征融合引入注意机制。实验结果表明,在黑烟车辆数据集下的AP达到黑烟92.53、车辆97.84,相较CenterNet算法分别提升了2.86、5.7。In the current traffic road,there are a large number of vehicles with tail gas holes that are thick with black smoke,which seriously pollute the air and damage human health.At present,the black smoke and vehicle detection adopts a manual meth⁃od or a traditional machine learning method based on manual features,which is time-consuming and low in accuracy.This paper takes the CenterNet based on ResNet18 as the basic solution for the detection of black smoke vehicles under the video surveillance scene,and further improves the above structure based on the experimental results,and proposes a dual-branch black smoke vehicle detection network based on the attention mechanism.The dual backbone network is used to extract the targeted feature representa⁃tion,and the attention mechanism is introduced for the feature fusion of the dual backbone network.The experimental results show that the AP under the black smoke and vehicle dataset reaches black smoke 92.53 and vehicle 97.84,which is 2.86 and 5.7 higher than the CenterNet algorithm.

关 键 词：黑烟车辆检测 卷积神经网络 目标检测 注意机制 双分支 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]