基于深度学习算法的机动车尾气排放黑烟监测  

Monitoring of black smoke from motor vehicle exhaust based on deep learning algorithm

作　　者：江东[1] 王晓龙[1] 陈政 陈旭 JIANG Dong;WANG Xiaolong;CHEN Zheng;CHEN Xu(National Institute of Measurement and Testing Technology,Chengdu 610021,China)

机构地区：[1]中国测试技术研究院,四川成都610021

出　　处：《现代电子技术》2023年第17期66-69,共4页Modern Electronics Technique

摘　　要：为提高对黑烟车监测能力,提出一种基于深度学习算法的机动车尾气排放黑烟监测方法。提取运动前景机动车位置,并结合霍夫直线检测方法明确机动车尾气排放候选黑烟区域;对候选黑烟区域图像实施特征降维处理,以处理后候选黑烟区域图像为输入,构建深度学习算法的卷积神经网络模型;利用卷积层提取输入图像边缘、颜色、纹理以及表面特征,在池化层求出目标特征损失函数,以最小化损失函数为目标对网络实施训练,完成机动车尾气排放黑烟监测。实验结果表明,该方法在不同天气场景下对机动车尾气排放黑烟监测的平均准确率为97.89%、召回率最高为89.94%,监测能力强。In order to improve the monitoring ability of black smoke vehicles,a monitoring method of black smoke from motor vehicle exhaust based on deep learning algorithm is proposed.The location of the motor vehicle in the moving foreground is extracted,and the candidate black smoke areas of the motor vehicle exhaust emissions are identified by combining with the Hough line detection method.The candidate black smoke area image is processed by feature dimensionality reduction,and the processed candidate black smoke area image is taken as the input to construct the convolutional neural network(CNN)model based on the depth learning algorithm.The convolutional layer is used to extract the edge,color,texture and surface features of the input image,and the objective feature loss function is obtained in the pooling layer.The network is trained with the goal of minimizing the loss function,so as to complete the black smoke monitoring of motor vehicle exhaust emissions.The experimental results show that,in different weather scenarios,the average accuracy rate of the method is 97.89%,its highest recall rate is 89.94%,and its monitoring ability is strong.

关 键 词：深度学习算法 尾气排放 黑烟监测 黑烟区域提取 主成分分析 深度特征 损失函数 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP301[电子电信—信息与通信工程]