Classification Of Weather Images Based On Feature Fusion Under Deep Learning Framework

Weather conditions have a significant impact on the imaging quality of the device in outdoor machine vision applications.The use of artificial intelligence technology to automatically recognize the weather conditions of the image can enhance the lowquality images of the imaging equipment and is also conducive to the normal operation of the application system.In the intelligent transportation system,real-time discrimination of weather conditions also helps cars to assist driving,optimize traffic strategies,and ease traffic congestion.Therefore,improving the automatic recognition accuracy of outdoor weather images and building a reliable outdoor weather image classification and recognition system are the current research hotspots in the field of image recognition,and they have broad engineering application prospects.Based on the characteristics and shortcomings of existing weather image classification work on outdoor surveillance image datasets,this paper builds a fusion model of traditional weather image features and deep weather image features based on the convolutional neural network structure in deep learning.The main research contents are as follows:(1)Aiming at the different classification criteria of the existing weather image datasets,first,by comparing and summarizing the research of meteorological observation data and the existing weather image classification work,the weather image classification in this paper is determined as cloudy and foggy.,Rainy days,snowy days,and sunny days;Secondly,combined with the problem of uneven image quality of existing weather image datasets,and covering only multi-background classification application scenarios,a multi-application scenario with higher quality was constructed.Background and single background dataset.Among them,the multi-background data set is filtered from the public monitoring data set,and the single-background data set is collected by itself.(2)In the selection of weather features,the structural features,color features,and dark channel features commonly used in traditional weather image classification algorithms are compared and extracted,and different types of traditional feature vectors are constructed,and the convolutional neural network is used to detect deep layers.Weather features are extracted and analyzed;in the classification experiment,the same classifier is used to classify different types of weather image features,and the classification performance is compared,and the basis for the fusion of traditional weather features and deep learning features is studied.(3)In terms of classification model construction,in view of the lack of existing weather image classification work,the classification scheme of fusion of traditional features and deep features is determined,and the feasibility of feature fusion is proved by the correlation between features.Combining different feature fusion methods and feature fusion positions,based on the convolutional neural network structure,four types of feature fusion models were constructed for weather image classification.The classification results were evaluated using reasonable evaluation indicators.Experiments proved that Effect of feature fusion model on weather image classification performance.(4)Combined with the idea of feature selection,the adaptive network weight optimization strategy is selected,the structure of the feature fusion network is improved,and the weight of the feature fusion layer of the fusion model is optimized to achieve the purpose of feature selection.The experimental results prove that weight optimization can effectively improve the accuracy of network recognition,and the location selection of the weight optimization module has an important impact on the optimization performance.