Gradient Double Threshold Otsu Research Based On Evolutionary Algorithm And Its Application In Crack Detection

Nowadays,concrete construction facilities play a more and more significant roles in various aspects of our lives including roads,bridges,houses,tunnels,reservoirs,dams,and so on.However,concrete buildings become fragile and even damaged attributing to various factors such as cyclic load,fatigue cumulative damage,climate,natural disasters and so on.The cracks in concrete as an important warning of degradation of civil structures indicates that just-in time concrete cracks monitoring is an important task at this moment.Traditional detection methods can be mainly categorized into artificial vision inspection and ultrasonic detection.However,the two methods mentioned above are less feasible and flexible in practical application.The former one does not only require a great amount of manpower and but also limits to environmental constraints for monitoring.While the monitoring equipment of latter method is relatively expensive.Therefore,it is necessary to apply computer vision technology to concrete crack detection.Image segmentation referred as the primary task in a majority of computer vision experiments obtaining more and more attention of the researchers in this area.The purpose of image segmentation is to simplify the image based on clear regionalization.The simplified image enables to reduce the difficulty of feature extraction through removing some unnecessary feature details in the following classified work.Edge detection is a method of Image Segmentation.By detecting the edge of the image,it can achieve the goal of image regionalization.Compared with other image segmentation methods,edge detection can not only extract the outer edge of the sub-region,but also extract the inner edge(texture,etc.)of the sub-region.An edge detection algorithm based on evolutionary algorithm and gradient double threshold Otsu(EAGDT)is proposed in this paper,aiming to apply it into the detection and identification of concrete cracks.The main contents and key points of this paper are as follows:(1)Combining the Wiener filter and the Gaussian filter aims to improve the anti-noise property of EAGDT edge detection algorithm.Adopting Adaptive Adjustment Algorithm for Non-Uniform Illumination Images Based on 2D Gamma Function(2DGF),aims to reduce the negative influence of Illumination and Shadow parts on EAGDT Edge Detection.(2)An optimal double threshold algorithm based on gradient amplitude and maximum inter-class variance is proposed in this paper.The algorithm is used for threshold selection of EAGDT edge detection algorithm.The first step is to calculate an overall image gradient magnitude inter-class variance to select a high threshold;Then,calculate the gradient amplitudes inter-class variance of the smaller gradient amplitudes.The purpose of this step is to select the low threshold.By searching for the maximum value of the sum of variance between the two classes,the high and low threshold values will be obtained.Finally,both high threshold and low threshold can be used for EAGDT algorithm.(3)The paper proposed an improved Adaptive Evolutionary algorithm(IAEA)based on adaptive adjustment of three parameters: crossover probability,mutation probability and mutation search range.The purpose is to solve the double threshold based on gradient amplitude and maximum inter-class variance.Compared with the traditional evolutionary algorithm,the improved algorithm can shorten the iteration time and achieve quickly converge to the global optimal solution.(4)EAGDT edge detection algorithm is applied into concrete crack detection.The carryout of edge extraction enables to highlight the edge feature,which can be used for the preparation for the recognition and classification of the crack image.The next step is that HOG is used to extract the edge image from EAGDT.After that,GLCM is used to extract the gray image feature.Finally,using the extracted feature train the SVM classifier.The classifier enables to identify whether cracks exist in concrete images.The experimental results show that EAGDT can not only obtain more edge information but also suppress noise.The edge image obtained by EAGDT algorithm shows the advantages in strong continuity of edge detection,accurate edge location,detection of edge details,anti-noise and strong robustness.In concrete crack detection,high accuracy can be achieved.