Research On Weather Phenomenon Recognition Algorithm Based On Deep Learning

In daily life,human activities are closely related to weather phenomena.Automatic real-time recognition of weather phenomena has important research value and broad application prospect in automobile auxiliary driving,intelligent transportation,intelligent monitoring system and other aspects.In recent years,with the rapid development of deep learning in the field of machine vision,as CNN can extract rich,abstract,and deep semantic information from weather images,the algorithm of weather phenomenon recognition based on deep learning is studied in this paper.In view of the challenges and problems existing in the current methods of weather phenomenon recognition,the main research work of this paper is as follows.(1)Firstly,a weather image database named Weather Data Set-6 containing more categories is built,and the data set was extended by Cycle GAN to obtain Weather Data Set-6Plus,which greatly increased the size of the weather image data set,realized the diversity and distribution balance of the weather image data,and improved the universality and generalization ability of the weather recognition model.(2)Secondly,a three-channel convolutional neural network(3C-CNN)model is presented to recognize the weather phenomenon based on the classical CNN model and method of transfer learning.Three different CNN branches are used to extract the sky features,ground features and global features of the weather images,and the extracted weather features in each region are fused by Concatenate function.Finally,the weather images are recognized and classified by Softmax classifier.The model has both high accuracy of weather recognition and fast recognition speed.In addition,the number of parameters and model size of 3C-CNN model are relatively small,which can basically meet the real-time recognition of weather phenomena on most mobile and embedded terminal devices.Therefore,this model has important practical value.(3)Finally,the algorithm of weather phenomenon recognition based on deep learning is applied in the field of automobile auxiliary driving,and an automobile auxiliary driving servo system based on weather phenomena recognition is proposed.Tornado framework is adopted to implement the deployment of weather recognition application service,provide weather recognition service for the automobile auxiliary driving servo system,and realize end-to-end real-time recognition of weather phenomena.