Parallel Convolutional Neural Network For The Classification Of Uneven Illumination Pavement Images

In recent years,with the rapid development of highway traffic,crack detection as an important composition of pavement quality monitoring and conservation has been widespread concerned,how to identify the cracks quickly and accurately has gradually become a hot research.In this paper,author focus on the problem of how to improve the recognition accuracy by using the parallel convolution neural network.Based on the Lambert illumination model,the existing illumination invariant extraction algorithm is analyzed and studied.Aiming at the problem of the pavement cracks images affected by uneven illumination and the illumination invariant extraction method based on NSCT(Nonsubsampled contourlet).A new method of illumination invariant extraction is designed by studying the influence of illumination on pavement cracks images.Based on Lenet structure and high-level feature fusion theory,this paper studies and analyzes the parallel synchronous convolution neural network.In this paper,the validity of the proposed method of illumination invariant extraction and the parallel synchronous convolution neural network and are verified by using 2000 images with six types of pavement cracks affected by uneven illumination.Experiments were performed on Caffe platform GPU computing conditions.The results show that the original image supplemented by the illuminate invariant feature can effectively improve the classification accuracy of uneven illumination.This paper also optimizes the parameters of the network model to ensure its recognition effect.