Research On Dust Image Recognition Method Based On Improved Residual Network

With the continuous development of economic level,health is becoming increasingly important to us.Nowadays,the diseases caused by inhalable particulate matter are gradually increasing.People’s health problems have became a topic of great concern.Especially for the elderly and children,it is more likely to cause a variety of diseases.There are many large and small particles in the atmospheric environment,many of which are inhalable by human body,among which dust is a very important part of these substances.The source of dust is complex,mainly including road dust and construction dust.In the actual application process,the traditional dust monitoring method has many troubles and inconveniences.With the rapid development of computer technology and the widespread application of image recognition technology,intelligent recognition of dust pollution through video surveillance will be an important means to prevent dust pollution in the future.In the past ten years,the application of deep learning algorithms in the field of video monitoring has been greatly developed,but there are few studies on the recognition of dust images using deep learning methods.For this reason,the research on dust recognition based on deep learning methods has become a topic worthy of research.The object of dust identification is to search dust pollution under the cross camera through dust image.It is a technology that uses computer vision technology to judge whether there is dust pollution image in video or image.In this paper,the traditional dust recognition method has a low recognition rate and insufficient key information extraction in the feature extraction process.The research on the dust recognition method is carried out.The main contents are as follows:(1)Design of dust identification model based on improved residual network.Firstly,the characteristics of dust are analyzed,and resnet50 network is applied to the dust data set.Secondly,the network structure of resnet50 is improved,and the first improved network model is formed,which is named Update＿Res.Adding spatial pyramid pooling to solve the problem that the input image size is not fixed,not only increases the scale invariance of the dust recognition model,but also suppresses the occurrence of overfitting during the training process.And change the strategy of pyramid pool to average pooling.Applying the method of expanding the feature map to the backbone network,removing the last spatial downsampling of the network,is conducive to extracting more fine-grained features,improving the performance of the model,and thus improving the recognition rate.Then,in order to solve the problem of insufficient key information extraction in the process of dust feature extraction,on the basis of Update＿Res network,attention mechanism module is introduced.The improved network model in the second step is formed and named Final＿Res.Using Update＿Res is used as the backbone network,and the attention mechanism module is introduced into the last two residual blocks of the network;then the features of the corresponding levels are extracted from the two residual blocks,and then the features of the two different levels are fused to generate the final features.In addition,the improved activation function adopts a dynamic activation function,which dynamically adjusts the parameters of Re LU according to the input characteristics,and enhances the nonlinear expression ability of the network model.(2)Dust image recognition process based on improved residual network.First of all,there is no public data set for dust image recognition.Therefore,all kinds of pictures about dust scenes are collected by Internet and camera for data collection,and the training set and test set are formed.Then,the data set is expanded by using the method of data enhancement,and more data is generated in the training set and test set to increase the amount of data in the dust data set,It solves the problem of no data set and insufficient data in this study,enhances the generalization ability of the model,and improves the ability of the network model to adapt to various application scenarios.Then,the dust image is processed by image preprocessing,including graying,normalization and denoising.Finally,the training process of the two models is described.Finally,experiments are conducted to verify the effectiveness of the method proposed in this paper in the recognition of dust.The experiment shows that the method in this paper has a higher accuracy in the recognition of dust images.