Research On Infrared Image Fusion And Recognition Extraction Method For Road Crack Detection

Road cracks are one of the common diseases that reduce the bearing capacity of the pavement and accelerate the damage of the pavement.Only by obtaining the full element information of cracks can we better provide a more accurate and favorable basis for road flaw detection,and it is also the basic requirement for road safety monitoring and evaluation.Because crack images are noisy due to shadows and uneven illumination caused by the environment,including threshold segmentation methods,crack extraction errors often occur.The clarity,integrity and accuracy of the detection results of the current crack detection methods need to be improved urgently.The thesis carries out research work from infrared and visible light image fusion method,road crack detection method,road crack width measurement and development of one-stop road crack detection system.The main research contents and innovations of the paper are as follows.(1)Aiming at the problem of image quality degradation caused by factors such as sensor noise,shadows and uneven illumination,the non-subsampling shearlet transform(NSST)image fusion method is used to enhance image clarity and improve image quality.First,the source images are denoised and enhanced according to the image characteristics of the infrared image and the visible light image;then,the images with unqualified quality are removed by the spatial frequency index;then,the source images of the infrared image and the visible light image are decomposed at multiple scales,respectively.Decompose into low-frequency sub-images at different scales;then use the weighted average and absolute value maximum rules to fuse the corresponding low-frequency sub-images at each scale;finally,use the inverse transform to reconstruct the fusion image.Compared with the non-subsampling contourlet transform algorithm,the AG,IE,MI,and evaluation indicators are improved by 0.12,0.06,0.12,and 0.01,respectively.The results show that the NSST method can generate better fused images,and the fused images contain more detailed image information.(2)Aiming at the problem of reduced extraction accuracy caused by noise and non-crack irrelevant region interference commonly found in road crack detection,a symmetric structure,depthwise separable convolution structure and connected domain component labeling based on convolutional neural network are proposed.Optimization algorithm for road crack segmentation model.5680 infrared images were collected for model training and testing.The test results show that the method can meet the detection effect of road cracks in different shadow images,and the detection accuracy can reach 92.23%.Compared with FCN,U-Net and Deeplab V3+networks,the similarity is improved by 5.97%,5.22% and 1.49%,respectively,and the results prove that the proposed algorithm outperforms the widely used algorithms.Therefore,this method is effective for crack detection in different road infrared images,which can improve the accuracy of segmentation results,reduce the recognition error rate,and improve segmentation accuracy and robustness.(3)The paper studies a method of crack width measurement using infrared visual images.First,for the binary image of the above image segmentation result,the fracture skeleton is extracted by the central axis transformation algorithm;then,the fracture area,perimeter,and skeleton length are calculated,and then the average crack width is calculated.Using 100 crack images to calculate the width,it is found that the actual average width of cracks and the development law of error and error rate,in which the width value [6.9,54],there is a threshold of 7.59,when it is smaller than the threshold,the error is greater than zero and close to zero,The more slender the crack is,the calculated value is greater than the actual average width of the crack;on the contrary,the error is less than zero and the calculated value is smaller than the actual average width of the crack.(4)The paper develops a one-stop infrared crack detection system that integrates infrared and visible light image fusion methods,road crack segmentation methods and crack width measurement functions.Through the input of infrared and visible light road crack images,a series of operations are performed to obtain fused images,segmentation results and crack sizes.The self-developed crack detection system was tested and achieved good results,providing users with a good human-computer interaction platform.