Road Surface Condition Classification Using Deep Learning

In recent years,people have put forward higher and higher requirements for the safety of vehicle driving and handling.Due to the complexity and variability of the road scene in the actual driving,there is not a method with high recognition accuracy,good real-time performance and strong robustness to solve the problem of road state recognition.Therefore,it is of great practical significance for people to detect and identify the road condition accurately and quickly,and make timely warning in the case of bad road conditions.At present,image recognition method mainly uses manual feature extraction and then put it into machine algorithm for judgment and recognition,which is complex and time-consuming.The deep learning visual algorithm provides a good solution.It combines road image feature extraction and recognition and recognition into the same model,automatically extracts image features,simplifies the identification and identification steps,and reduces the workload of people.In this paper,the problem of road condition image recognition is taken as the research object.Based on the literature of related fields at home and abroad,the following two problems in road condition image recognition are studied.First,the uneven illumination intensity of the road image,the single-scale Retinex image enhancement algorithm due to the lack of illumination led to the image halo phenomenon and the image detail expression is not obvious,a color image enhancement technology based on the improved single-scale Retinex algorithm is proposed.First,the weighted least square method is used to enhance the details of the original color image.Then the original image is processed by a modified single-scale Retinax.The gain coefficients of the processed image layer and the detail image layer are constructed,and a new merged image is reconstructed.The experimental results show that the proposed algorithm can effectively remove the halo phenomenon,make the image details more prominent,and improve the automatic feature extraction in depth learning.Second,the deep learning method improves the recognition accuracy.In view of the need of fast real-time and high accuracy in the intelligent driving,the traditional image classification and recognition technology can not meet the needs well,and a suitableactivation function in the convolution neural network model has an important impact on the recognition results.Based on this,a road state classification method based on deep learning is proposed,and a new activation function is constructed by combining with the relu activation function.The experimental results show that the accuracy of classification and recognition of the road state database is 94.89%.The convolution neural network based on the improved activation function has better generalization ability and classification accuracy.