Research On Air Quality Classification And Measurement Method Based On Transformer

While enjoying all kinds of convenience brought by industrialization and urbanization,we must also be aware that in the process of rapid development,the ecological environment is deteriorating and has caused a series of problems.Among them,the hidden danger of declining air quality is closely related to each of us.The deterioration of air quality has many effects on people’s health,as well as exacerbating environmental problems such as acid rain and global warming.In order to prevent and control air pollution,the primary problem is to realize the detection of air quality.Among many air quality detection methods,there is no doubt that the detection method based on vision has a broad prospect.With the rapid development of deep learning technology in recent years,the field of computer vision has made great progress in image classification,detection and related low-level visual tasks.Meanwhile,with the popularization of various smart devices,it is no longer difficult to obtain large amounts of image data.Therefore,in recent years,the air quality measurement based on image method has achieved good results in some scenarios.The Transformer model has received widespread attention in th e field of natural language processing and computer vision.Among them,Vision Transformer’s excellent global feature extraction capability is suitable for solving image-based air quality measurement problems.Therefore,this paper study air quality measurement methods based on Transformer.The main innovative achievements of this paper include:(1)In view of the difficulty in extracting stable features in the image space domain to guide air quality measurement,this paper introduces the frequency domain feature extraction method and use Transformer architecture to achieve global feature extraction of air quality images in the frequency domain.(2)This paper designed a frequency domain feature channel screening module to generate the weight of the channel corresponding to the feature by self-learning,so that the model can produce different responses to features of different importance in the measurement task,so as to achieve more accurate measurement effects.(3)Aiming at the problem of high-precision air quality measurement,this paper explored the air quality measurement method based on multi-model ensemble learning,and achieved better results by integrating multiple basic classifiers on this problem.