Pavement Disease Detection System Based On Neural Network And Multi-scale Fusion

Highways are an important lifeline for maintaining the country’s transportation and play an important role in national development.Regular inspection is the first step in the process of maintenance.The traditional method of road diseases detection which uses human eyes has the disadvantages of high intensity and time-consuming.With the development of image processing,a variety of detection systems have been developed which focuses on the automation and intelligence.However,the existing detection strategies generally cannot get rid of the shortcomings of high hardware requirements,and generally use the form of data labels for segmentation,resulting in longer training time for the data set.Therefore,this paper proposes a new pavement disease detection system based on neural network and multi-scale fusion algorithm.In this paper,linear transformation and equalization based on the histogram of the disease image are performed to enhance the gray-scale feature,the image is divided into high and low frequency sub-bands using non-subsampled contourlet transformation,and the Sigmoid function is used to enhance the detailed features.The general denoising filter commonly used in feature extraction algorithms effectively reduces the running time.At the same time,the corresponding adaptive order adjustment model and edge extraction algorithm are constructed to obtain binarized feature images.Compared with traditional methods such as region growth and watershed Has the advantage of fast speed.At the same time,a feature classification system was constructed on the basis of the binary convolutional neural network,and a variety of databases were used to simulate on the Matlab,which has a classification model with an accuracy of 90.11%.At the same time,the measure method of road damage rate(DR)and pavement damage index(PCI)was designed。...