Road Damage Recognition Method Based On Target Detection And Semantic Segmentation

In recent years,my country’s transportation industry has become one of the fastest-growing industries and has become a core part of the national economy.The construction of transportation infrastructure is linked to various links in society and plays a key role in the economic growth of cities and countries.The construction of a high-speed network with full coverage can connect some cities with a small population,improve the convenience of residents’ travel,and support the development of new urbanization.According to the plans of relevant departments,by the end of the 13 th Five-Year Plan period,my country’s total road mileage will reach a breakthrough of 150,000 kilometers.By 2020,my country will have more than 8 kilometers of average roads per square kilometer in urban built-up areas.Network density,urban road coverage rate reached 15%.The total number of kilometers of roads in our country is still rising,and the requirements for road grades are getting higher and higher.Regular inspection and maintenance of road pavement is of great significance to the development of my country’s transportation industry.Image-based road crack detection technology will gradually replace traditional manual detection methods and become the mainstream measurement method due to its fast detection speed,convenient method and high degree of automation.However,the current hardware systems and software algorithms based on image detection have limitations,which cannot achieve the requirements in practical applications.Both target detection and semantic segmentation have become key issues in the field of computer vision.Target detection mainly focuses on the objects that need to be recognized in the image,and points out the location and size of the object.Semantic segmentation refers to the automatic recognition of the image by the machine.The pixel level classifies objects and provides category,location and morphological information for objects.It has excellent application value in autonomous driving,remote sensing image processing,medical diagnosis,and fashion matching.Therefore,the development of semantic segmentation has affected the society.The development of various projects in the event.Based on this,in view of the many shortcomings of the original manual detection method,combined with the current research status of the existing machine vision system,digital image processing algorithm and deep learning technology,this paper has completed the following research content:First of all,based on the principle of target detection,using the DSSD(Deconvolutional Single Shot Detector)model as the architecture,a network structure capable of detecting road surface damage is studied.The model uses Res Net101 with high recognition accuracy as the basic network of the DSSD model,and adds an attention mechanism to it,and uses the combination of channel domain attention and spatial domain attention to achieve the addition of weights and spatial angles to the channel angle.The centralized summary on the above improves the detection effect of road cracks on small targets.At the same time,in order to reduce the proportion of simple data and increase the proportion of difficult-to-classify data,Focal loss is used to improve the overall detection effect,which is more conducive to the detection of cracks.Then,according to the results of the target detection,the block diagrams of various types of cracks in the picture are cut out,and the labelme tool is used to label the cracks in each block diagram.The content of the labeling includes the type and shape of the cracks,and the application of semantic segmentation is produced.Data sample.Finally,this paper studies an adversarial network application to segment the objects in the picture,and the obtained segmented image is re-spliced back to the entire image according to the coordinates at the time of cropping,so as to realize the semantic segmentation of the entire image.Through the use of adversarial neural networks,the accuracy of semantic segmentation in the segmentation of small targets is effectively improved,and the segmentation effect of the algorithm in image segmentation is improved.