Research On Edge Detection Algorithm Based On Deep Learning

As one of the basic tasks of computer vision,edge detection has always been the focus of computer research and exploration.Especially now,with the extensive use of neural network and deep learning,a large number of scholars try to structure different networks for edge detection.The definition of edge emphasizes the change of image optical properties,so the trained network can achieve good detection results even without fine-tuning.However,the current research results still need to be fine-tuned in the corresponding data set,and still can not be used in practical applications.But in the era of deep learning,edge detection emphasizes the training of data set,and it should also be applied in practical application.This paper analyzes the existing edge detection methods,and finds that although the current edge detection method has better detection effect and has surpassed the effect of human annotation,there are still some problems.Most of these methods pay more attention to the correctness of edge detection while ignoring the crispness of the edge,so the detected edge is not very crisp,but in some high-level detection tasks and practical applications,there are higher requirements for the crispness of edge detection,such as optical flow task and highway crack detection task in this paper.Based on this problem,this paper optimizes and improves the existing methods,proposes a new edge detection algorithm.The experimental results show that the method in this paper is better than the existing ones.The main work of this paper is as follows:(1)Combined with the advantages of the Dexi Ned and CED,The paper proposed a new and more robust edge detection algorithm used single network structure which can detect the edge without any fine-tuning.In this paper,the original extremely dense network is transformed into a single network structure,which can not only correctly detect the edge,but also greatly improve the crispness of edge detection.(2)We have carried out horizontal test and analysis on several open data sets,and the final experiment results show that the detection effect of the proposed method is improved by 3%-5% compared with the previous method.(3)In this paper,the edge detection is applied to the real life,that is,highway pavement crack detection.In this paper,several different detection methods are used to detect the cracks.The final experiment results show that the cracks detected by this method are clearer,more crispness,and more correctness.(4)Finally,the significance of some special network structures of edge detection is explored.The influence of different network layer on the model is compared,and the middle layer of the network is output.At the same time,this paper adjusts and trains the special up sampling method,iteration times,whether to use non maximum suppression and loss function in edge detection,and gives the corresponding experimental process and the final optimization suggestions.