Research On Vehicle Black Smoke Feature Recognition Method Based On Road Monitoring

The particulate matter in motor vehicle exhaust not only pollutes the environment,but also endangers people’s health.Black smoke vehicles have become a key target for control due to their high pollution.The automatic detection of black smoke vehicles based on machine vision can effectively reduce the consumption of manpower and material resources.However,there are still problems like low detection accuracy and slow detection speed caused by vehicle shadows and road marks.Therefore,this thesis aims to improve the detection effect of black smoke vehicles and conducts research on road monitoring videos.In order to improve the recognition effect of weak smoke information,this thesis proposes a multi feature fusion black smoke recognition method.By comparing and analyzing the characteristics of black smoke,the color features of black smoke are extracted using color moments;Perform Gaussian smoothing and downsampling on the image,fuse different scales of the image,and extract LBP texture features separately;At the same time,three different aggregated Gabor kernels are weighted and averaged to extract multi-scale frequency domain features of the image.The three features are fed into the Ada Boost ensemble learning classifier for training,and the recognition results are fused using a weighted voting method.Compared with other smoke recognition algorithms,this method has significant advantages in recognition speed while ensuring recognition accuracy.In order to further improve the accuracy of black smoke feature recognition,this thesis proposes a residual network recognition model based on attention mechanism.This model takes Res Net34 as the basic network,and avoids neuron inactivation and output data distribution confusion by replacing the activation function and adjusting the network structure order,so as to improve the accuracy of black smoke recognition.To reduce computational complexity and improve recognition speed,this article introduces a deep separable module.At the same time,in order to reduce the attention of interference information,the CBAM attention mechanism module is also introduced to enhance the extraction of key features of black smoke.The experimental results show that the network model proposed in this thesis improves the performance of black smoke recognition and has advantages in various indicators compared to other existing methods.