Research On The Algorithm Of Concrete Surface Crack Detection Based On Neural Network

Crack hazards have serious impacts in the transportation and construction industries such as concrete surfaces,bridges and pavements.To ensure the stability and maintainability of systems related to the transportation and construction industries,it is crucial to develop efficient crack detection processing algorithms.Meanwhile,crack images acquired by imaging devices in outdoor acquisition environments are usually characterized by low illumination,low contrast and blurred details due to light absorption and scattering during the acquisition process,and pre-processing crack images for visual enhancement is also a key step in crack detection.In order to improve the clarity of crack images and ensure the efficiency of the detection task,a crack image enhancement algorithm based on a degradation model is proposed;in addition,after an in-depth analysis of the shortcomings of traditional digital image processing methods in crack detection,two crack image detection algorithms are designed by combining coupled neural networks and convolutional neural networks,and the specific work is as follows.(1)A degradation model-based image enhancement preprocessing algorithm is proposed.The coarse texture distribution of degraded image is obtained by using Canny detection operator,and the blurred distribution map of degraded image is obtained on the basis of scene depth.Then the minimum channel map of the clear image is estimated based on the image characteristics combined with the quadratic constraint model of adaptive blur parameters,and the coarse transmittance is obtained using the atmospheric scattering model.The transmittance optimization is combined with adaptive lower boundary constraint and bilateral filtering,and the local atmospheric light is improved by morphological operations.Finally,the clarification of degraded images is achieved according to the recovery model.Experiments show that the proposed algorithm better solves the problems such as low illumination and low contrast of degraded images.(2)To address the problems of time-consuming,low accuracy and non-adaptive threshold of the existing methods based on coupled neural network crack detection,a modified pulse-coupled neural network model based on side-suppression and automatic parameter-adjusted fire-controlled modified simplified pulse-coupled neural network,PA-FCMSPCNN)for crack detection algorithm.First,in order to overcome the problem of inadequate stimulus influence of peripheral neurons,the visual side inhibition mechanism,which also has a biological background,is introduced into the model,and more accurate input stimuli and internal activity terms are obtained to achieve an optimized improvement of the model.Secondly,the adaptive detection thresholds are implemented in the RGB and HSV spaces of the cracked concrete images according to their color characteristics and luminance information.Then,a new parameter setting method is proposed,which effectively ensures the model stability and the rationality of the parameters.Finally,the detection results are refined by morphological operations to obtain refined crack detection results.The experiments show that the proposed method can obtain crack detection results with higher accuracy and more comprehensive detail retention,and also has good performance in quantitative evaluation.(3)A crack detection algorithm based on multi-scale parallel extraction and attention fusion network is proposed to address the problems of low generality of coupled neural network crack detection algorithm and loss of details and low accuracy of existing deep neural network crack detection methods.First,the high and low-level features of the crack scene are extracted by using multi-scale convolutional parallel neural networks with different depths;then,to improve the detection accuracy,the high and low-level features of the crack scene are effectively fused by combining the pixel attention mechanism for the features of the crack scene,and the effective fusion features for crack detection are obtained;finally,the crack detection output is performed by using nonlinear mapping.The experiments show that the proposed algorithm can obtain effective features for high-precision detection results,the details of crack detection results are clearer,and the supervised learning approach largely eliminates the noise interference of detection results and obtains better visual results;in addition,the proposed algorithm also has good performance in the evaluation of quantitative indexes such as regional contrast.