Research On Air Quality Recognition Model Based On Scene Image

Air pollution is a serious threat to human health,and air quality has become the focus of public attention.Therefore,it is very important to effectively and timely air quality monitoring for public health protection and pollution control.At present,environmental quality is mainly obtained from air quality monitoring stations,but they are sparsely distributed and monitoring cannot cover all areas.Image-based air quality estimation provides a new method for real-time air quality.With the popularization of mobile phones and the development of computer vision technology,there are ways to recognize air quality by taking environmental images at hand.People can take pictures through smart terminals such as mobile phones,and obtain the air quality of the environment through images.Environmental managers can analyze the air quality in the grid area in real time.The core of this mode is the image-based air quality estimation model.This paper is based on deep learning algorithms and builds image data to study the method of image-based air quality estimation.Furthermore,this paper studies the role of weather conditions in image air quality estimation.Research integrates weather features into air quality image recognition multi-task deep learning methods.This paper designs and implements an air quality estimation system based on environmental images.This paper includes the following three parts:Image-based ResNet air quality estimation model.Using the constructed air quality scene image data set,a ResNet-based deep convolutional neural network model(AQC-Net)is proposed to estimation air quality index through image features.This paper adds a self-supervision module to the model and combines the global context information of the feature map.We use the interdependence between the channel mappings to reconstruct the features,enhance the interdependent channel mappings,and improve the ability to express features.This paper compares AQC-Net with SVM,VGG and ResNet on the same data set.Experimental results show that the accuracy of the AQC-Net model for air quality classification reaches 72%,which is 12%,3.7% and1.9% higher than the SVM,VGG and ResNet models.Add weather feature analysis and classification recognition process based on deep learning to the air quality image recognition model.Because weather factors have an important impact on air quality,add a weather feature analysis model based on image recognition.This model is pre-trained on the Image Net data set,combined with migration learning to train and classify the weather-air quality image data set(WAQI),and divide the images into 5 categories: sunny,cloudy,overcast,precipitation and floating dust.Experiments show that the classification accuracy of the model is 97.4%,which is 1.15% and 1.99% higher than the classical algorithms such as Random Forest and Goog Le Net,respectively,which verifies that the model is more suitable for the analysis of weather characteristics.Multi-task and image-based air quality estimation model incorporates weather features.This paper uses a multi-task model set,combining air quality features and weather features in the image.The main task is air quality recognition,and the auxiliary task is weather recognition,finally obtain the AQI of the environment in the image.The model extracts air quality and weather feature data from images acquired by the image collection terminal.The model can identify 5 weather conditions and 6 air quality indexes.The model is trained and tested on the image dataset(WAQI)constructed in this paper,and WAQI is compared with single-task models such as Goog Le Net,Vgg Net and ResNet.This research show that in the air quality estimation task,the recognition accuracy of the multi-task model is 22.97%,20.94% and 14.33% higher than that of Goog Le Net,Vgg Net,and ResNet respectively.In addition,compared with the model AQC-Net,which is proposed in the first part,the accuracy rate is increased by 12.45%.Finally,the paper presents the design and implementation method of an air quality recognition system based on scene images.