Research On Image And Video Classification Method For Air Quality Measurement

Air pollution prevention and control is a major strategic demand of our country at present,and air quality monitoring is an important basis for air pollution prevention and control.The traditional air quality measurement methods have some shortcomings,which can not meet the current requirements of fine and intelligent air pollution monitoring.In recent years,with the rapid development of machine learning and computer vision,the measurement method of air quality based on vision information has been paid more and more attention.In this paper,the problem of image and video classification for air quality measurement is deeply studied,and the innovative achievements are as follows:(1)A new air quality measurement method based on video data is proposed.In the process of image data acquisition,a single image is easy to be affected by light,jitter and other noise interference,and the field of vision of a single image is limited,so it is often difficult to evaluate the air quality accurately.This paper proposes to take the video data as the input to measure the air quality.By fusing the results of multiple frames,the video data can fully smooth the noise interference of single frame,and through the superposition of multiple frames,its receptive field is significantly larger than that of a single image,and more visual information can be collected,which is conducive to the prediction of the results.(2)A new air quality measurement algorithm based on dual-channel 3D convolution network is proposed.In order to solve the problem of small distance between classes of air quality data,according to the characteristics of outdoor environment,this paper proposes to use semantic segmentation to divide the outdoor environment into different regions.Through the design of a dual-channel network,with the assistance of the regional information provided by the semantic segmentation channel,the upper channel is guided to extract the information of different regions,and the stable prediction is carried out by combining the features of the two channels.(3)An outdoor video dataset for air quality measurement is established and annotated.A large number of experiments are carried out on this dataset,and the results show that the method proposed in this paper is accurate and effective.