基于特征采样引导和集成RFELM的道路高排放源识别模型  

On-road high-emitter identification model based on guided feature sampling and ensemble RFELM

作　　者：周汉胜 段培杰 李泽瑞 周金华[1] ZHOU Hansheng;DUAN Peijie;LI Zerui;ZHOU Jinhua(School of Biomedical Engineering,Anhui Medical University,Hefei 230023,China;Institute of Artificial Intelligence,Hefei Comprehensive National Science Center,Hefei 230088,China;AHU-IAI AI Joint Laboratory,Anhui University,Hefei 230601,China)

机构地区：[1]安徽医科大学生物医学工程学院,安徽合肥230032 [2]合肥综合性国家科学中心人工智能研究院,安徽合肥230088 [3]安徽大学安徽大学与合肥综合性国家科学中心人工智能研究院联合实验室,安徽合肥230601

出　　处：《现代电子技术》2024年第6期124-130,共7页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(62103125);国家自然科学基金资助项目(62033012);安徽省博士后研究人员科研活动资助项目(2021A484)。

摘　　要：机动车排放的污染气体会对环境造成严重危害,其中尾气排放超标的车辆是主要污染来源,因此实现对道路高排放源的有效识别具有重要意义。针对尾气遥测数据,提出一种基于特征采样引导和集成随机傅里叶特征极限学习机(RFELM)的道路高排放源识别模型。首先对遥测数据进行多次随机采样,构建多组训练子集;然后对每组训练子集进行多次特征采样,并训练对应的子分类器,根据组内最优子分类器的输入特征更新特征采样的概率与特征权重;最后对所有子分类器的验证分数进行排序,筛选出一定比例的RFELM组成分类器集合,采用加权投票法预测数据的标签。实验结果表明,相比于RFELM和随机森林等算法,所提模型在真实的道路遥测数据上具有更好的识别效果,还有着更强的抗噪能力。The pollution gas emitted by vehicles causes serious harm to the environment,among which the vehicles with excessive exhaust emissions are the major sources of pollutions.Therefore,it is of great significance to realize the effective identification of high-emitters on the road.A high-emitter identification model based on guided feature sampling and ensemble random Fourier feature extreme learning machines(RFELM) is proposed to classify the on-road remote sensing data.The remote sensing data is randomly sampled several times to construct multiple training subsets.Then,each training subset is sampled several times to train corresponding subclassifiers.The sampling probability and weight of feature are updated according to the input features of the optimal subclassifiers in the group.The validation scores of all subclassifiers are sorted,a certain proportion of RFELM is selected to form the classifier set,and the weighted voting method is used to predict the labels of the test data.The experimental results show that in comparison with RFELM,random forest and so on,the proposed model has better recognition performance and stronger noise resistance on real road remote sensing data.

关 键 词：道路高排放源识别 遥测数据 特征采样 集成学习 随机傅里叶特征极限学习机 子分类器 

分 类 号：TN957.523-34[电子电信—信号与信息处理] X734.2[电子电信—信息与通信工程]