Research On Algorithm Of Road Defect Detection Based On Machine Vision

Good road condition is the precondition to ensure the traffic safety.It has become the primary task of road maintenance to obtain the information of road defects quickly and accurately and avoid the threat of road safety.At present,the cost of manual detection method is high,the risk coefficient is large,and it is easy to be affected by human factors,resulting in the phenomenon of missed detection and false detection.Therefore,it is of great significance to realize the detection of road defects by technical means.As a classical image detection technology,full convolution neural network(FCN)is widely used in the field of deep learning.In order to improve the performance of the existing road defect detection algorithm,FCN is introduced into the road defect detection in this paper for algorithm optimization.The main contents are as follows:(1)To solve the problem that FCN is not perfect for small feature extraction and spatial consistency is not considered,a multi-scale global FCN algorithm is proposed.The multi-scale dilated convolution structure is composed of three single-scale modules with different dilated rates in parallel,which can expand the sensing range without losing the resolution information,thus sensing small features.At the same time,the deep information and shallow information of the network are fused by using the transposed convolution layer to further expand the image details.On this basis,the conditional random field(CRF)structure is added to realize the end-to-end connection between the network and CRF,so as to calculate the loss value and complete the transmission of global pixel information.The experimental results show that the algorithm can not only effectively identify the characteristics of small defects,but also consider the global dependence between pixels,which significantly improves the detection performance of road defects.(2)In order to reduce the computational burden of convolutional neural network,the existing algorithms often replace the conventional convolution with the lightweight algorithm which takes the deepwise separable convolution as the core,but at the same time,it will bring the node characteristics degradation and the information flow is not smooth,which will affect the accuracy of network detection.In order to solve these problems,a lightweight algorithm is proposed,which combines the SENet channel weighting module with the depthwise separable convolution.By using the depthwise separable convolution,the information of each channel is learned and fused respectively,and the SENet module and the first depthwise separable convolution layer are used as channel weighting.By adjusting the characteristic parameters between channels,the feature recalibration is completed,so as to make up for the phenomenon of node feature degradation and information poor circulation.The experimental results show that the algorithm can reduce the computation and complexity of the model,and effectively avoid the loss of detection accuracy.(3)Based on the improved algorithm,the software of road defect detection is designed and developed by Py Qt.The software has the functions of single or multiple image detection,defect degree evaluation,data storage and query.The test results show that the software can realize the accurate automatic detection of road defects,and determine the condition of road defects by calculating the relevant parameters of the degree of defects,so as to provide accurate data for the regular maintenance of the road.The work done in this paper effectively improves the accuracy of road defect detection,reduces the complexity of the model and calculation,and has a certain practical value.