Research On Smoke Detection Algorithm Of Smoky Vehicles In Real Scenes

In 2021,the State clearly stated that one of the main goals of “The 14 th Five-Year Plan(2021–2025)(the Plan)for National Economic and Social Development” is to continuously reduce the total amount of major pollutant emissions,stabilize and moderately decrease carbon emissions after peaking,and continue to improve the ecological environment.Realizing the automatic identification of vehicle black smoke exhaust can contribute to the country’s victory in the defense of lucid waters and lush mountains,and is of great significance to the current environmental protection work.In order to solve the above problems,the research on the smoke detection algorithm of smoky vehicles in the real scene is carried out.The work is carried out in the following aspects:Medium and large vehicles are the key detection targets of black smoke emission.Vehicle detection of key detection target vehicles can effectively reduce the amount of data to be processed and facilitate subsequent feature extraction.In view of the above problems,this paper improves the Yolo V4 algorithm,introduces the CBAM attention mechanism,and improves the positioning accuracy and detection accuracy;The EIOU loss function is used to replace the original CIOU loss function to solve the problem of the limitation of the horizontal to vertical ratio of the original loss function in the frame regression loss;K-means++ clustering algorithm is used to redefine the size of the anchor boxes of the network.According to the aspect ratio of the vehicle target to be detected in this paper,the anchor frame of 9 sizes is regenerated,which makes it more suitable for the data image database in this paper.The improved vehicle detection algorithm map increases by 8.5% and the detection speed is23 FPS,which can meet the needs of real-time detection while ensuring accuracy.The black smoke emitted by vehicles is translucent,which is difficult to distinguish from the asphalt pavement in the background,and has an unstable shape.On the premise of ensuring the recognition accuracy,in order to effectively compress the feature dimension and improve the recognition speed,this paper proposes a black smoke recognition method based on the block discrete cosine transform feature of the maximum information coefficient.Firstly,Kmeans algorithm is used to cluster the black smoke at the tail of all black smoke vehicles to obtain a universal aspect ratio,and the tail smoke exhaust area of the target vehicle is extracted according to the aspect ratio;Then,discrete cosine transform features are extracted from the image by blocks,and the correlation between features and categories is measured according to the maximum information coefficient,and effective features are selected.At the same time,PCA is used to reduce the dimension and remove redundant features;Finally,SVM classifier is input to recognize the black smoke of vehicle exhaust.Compared with the mainstream traditional black smoke recognition algorithm,the accuracy of this algorithm is basically the same,the recall rate is increased by 4.08%,and the black smoke recognition time of a single car tail image is shorter;The classical deep learning algorithm has high accuracy and recall,but the recognition time of a single frame image is more than 2ms.The recognition time of a single image of this algorithm is 0.3ms.From the perspective of system real-time,this algorithm has certain advantages.