Research On Image Classification Of Highway Cluster Fog Based On Convolutional Neural Network

In recent years,with the development of artificial intelligence technology,convolutional neural networks have been widely used in the field of image recognition.With the development of the Internet,image data has increased exponentially.How to quickly extract useful information from images has become very important.Therefore,the use of image classification technology to accurately classify images has important research significance for our research on the classification of cluster fog images.In recent years,my country’s economy has been in a stage of stable development.The mileage of expressways has continued to increase,and the total number of cars has also increased.The choice of expressways has the characteristics of high efficiency,economic convenience and comfort,which makes people choose expressways to go out.The odds become greater.However,due to the possibility of encountering foggy weather when traveling on expressways,the impact of fog on expressway traffic safety has always been a major problem to be solved,especially in the winter in northern my country,where there is a large temperature difference between day and night,especially Group fog is prone to occur.When there is a foggy weather on the highway,because the visibility will suddenly decrease and the friction between the wheels and the ground will decrease,the probability of vehicle rollovers and serial rear-end collisions will increase.It poses a threat to people’s lives and property.With the development of the expressway traffic network,the impact of fog on expressways has become greater and greater.The method of prohibiting traffic on expressways in foggy weather cannot fundamentally solve the problem of driving safety.It is extremely important to identify and classify highway pictures.In view of the fact that the driver cannot know the weather conditions of the highway section ahead in advance when the cluster fog occurs,this paper studies the classification algorithm of the highway cluster fog image based on the convolutional neural network.The main research work is as follows:This article first introduces the relevant basic theories and technologies of convolutional neural networks,introduces the basic structure and characteristics of convolutional neural networks,and also describes several more classic network models,and analyzes the development prospects of convolutional neural network algorithms.And research direction.Analyze CNN and SVM.CNN performs better in feature extraction.After the fully connected layer,SVM can be used to replace Soft Max classification to classify the cloud image.It has a higher recognition accuracy rate in the case of fewer data sets.And generalization performance.In the process of network training,the data set uses the ten-fold crossover method,and ten experiments are performed to analyze the results.The results are evaluated using three evaluation indicators: accuracy,recall and F1,which proves that the effect of cloud image classification has been effectively improved.At the same time,the accuracy of image classification when using different kernel functions is also analyzed.The results show that the accuracy of recognition and classification using Gaussian kernel function can reach 92.56%.Use this kernel function to construct the CNN-SVM model and compare the CNN model and SVM to obtain the classification accuracy of image recognition.In this paper,two methods of video extraction and Python web crawling images are used to create a highway with and without fog data sets,and some pictures that do not meet the conditions are artificially deleted,and there are a total of 1000 pictures.At the same time,in order to improve the accuracy of training and the occurrence of over-fitting,the data is enhanced by random interception,mirroring and other methods to make the training data more diversified and improve the accuracy of the training results.Next,the AlexNet network model is introduced,and the network is analyzed and summarized.An improved network model is proposed for the shortcomings of the network.Before the data is input to the network model,the image size is uniformly modified to the same size,and the images are batch normalized,so that the input data can have a similar distribution.At the same time,optimizing the network structure and using a smaller convolution kernel for image feature extraction can reduce the amount of parameters.Figure [38] table [13] reference [62]...