Design Of Road Scene Recognition System Based On Deep Learning In Complex Weather

As the quality of life improves,the traffic volume increases dramatically,and so does the accident rate.Intelligent driving technology can assist human operation and reduce the accident rate,so the development of intelligent driving technology is of great significance.In this paper,the target detection algorithm is improved based on deep learning method,so that the algorithm has excellent detection effect on road scenes.The main work is as follows:(1)The data set was established,and the samples were obtained through Pascal VOC and other open source databases,and then the data set samples were preprocessed.In order to enrich the data set and enhance the generalization ability of the system,the data set was preprocessed by six image processing methods in this paper to prepare for the subsequent experiments.(2)The recognition accuracy and speed of AlexNet,VGGNet and ResNet are compared and analyzed,and the network model with superior performance is selected as the basis to build the feature extraction network.Several common optimization algorithms are introduced,and the effects of different activation functions and different optimization algorithms on network performance are compared through experiments,paving the way for the construction of subsequent network models.This paper introduces the current mainstream target detection algorithms,such as R-CNN,Fast R-CNN,Faster R-CNN,R-FCN,YOLO and SSD,and analyzes the difference of accuracy and network performance of each algorithm,Therefore,the algorithm is selected.(3)Aiming at the problem of large number of VGG feature extraction network parameters and slow training speed,this paper proposed to improve the original VGG-16 feature extraction network of SSD based on the residual network RESNET-50.The first 24 layers of the network of RESNET-50 were selected and 6 additional convolution layers were added to build a new network architecture.Based on this,a new target detection method is built.Compared with SSD target detection algorithm,the improved algorithm has a certain degree of improvement in recognition speed and accuracy.The results show that the improved algorithm in this paper has a better performance in the recognition of road scenes in complex weather.