Research On Black Smoke Detection Algorithm For Moving Vehicles

Black smoke vehicle emissions will not only cause pollution to the air,but also bring different degrees of harm to human health.To achieve motor vehicle emissions supervision,China currently relies primarily on vehicle environmental protection inspection and road sampling.Video-based automatic detection technology for black smoke of moving vehicles can effectively reduce the consumption of human and material resources,and the subject has practical application significance.This thesis takes the motion vehicle in the road monitoring video as the research object,and researches the motion vehicle target detection algorithm and black smoke feature recognition algorithm,the main work is as follows:(1)In order to quickly extract the moving vehicle target in the video,this thesis analyzes the commonly used target detection networks and selects YOLOv5 as the base network,and finally selects YOLOv5 s as the motion vehicle detection algorithm in this thesis through experimental comparison of four network structures of YOLOv5.To further reduce the computation and improve the detection speed,the backbone network of YOLOv5 s is improved by using the lightweight network Mobile Netv3-Small,and the inference speed is improved by 40.9% and the number of floating-point operations is reduced by 61.8% while the accuracy rate remains basically the same as the original network of YOLOv5 s.(2)Vehicle smoke emission area interception is often affected by road traces and vehicle occlusion,which can lead to misclassification.To address the above problems,this thesis cleans the initial smoke emission area images by directional gradient features to remove images with too much interference information.Then the improved multi-scale chunking OTSU algorithm is used for feature extraction of black smoke images,which effectively solves the problem of weak smoke information loss.Finally,the classification of vehicle black smoke is performed by SVM classifier.Based on the experimental comparison between this thesis’ s method and the existing traditional black smoke recognition algorithm and deep learning black smoke recognition algorithm,this thesis’ s method has clear advantages in real-time while ensuring a certain accuracy and recall rate.