Research On Crack Detection Algorithm Of Concrete Pavement Based On Machine Learning

In recent years,with the improvement of computer computing performance,machine learning has begun to be widely applied in various fields of research.Among them,traditional machine learning algorithms can achieve good results in gesture recognition and handwritten digit recognition,but in the traditional civil engineering industry.In the middle,due to the randomness of the cracks on the concrete pavement,it is difficult to extract the features of the picture,and it is difficult to reduce the dimension.Under the condition of high-dimensional samples,the traditional machine learning method is used for crack identification,and the accuracy is low.Compared with the traditional machine learning method,Under the condition of high-dimensional sample size,deep learning is more advantageous in training speed and accuracy.The convolutional neural network CNN is a representative deep learning algorithm,which is widely used in image detection.Therefore,based on the principle of feature extraction and machine learning,this paper compares various feature extraction algorithms,preprocesses crack images,introduces traditional machine learning algorithms and deep learning algorithms,and uses traditional machine learning methods SVM.With the deep learning algorithm CNN,the pre-processed concrete crack picture was trained and detected,and finally the user interface interface program of concrete pavement crack detection was developed.(1)Acquisition and pretreatment of crack samples: The crack shooting process was designed.The obtained crack pictures were copied into 4 groups,and the features were extracted by Laplacian operator,Sobel operator and Canny operator respectively.The group only performs grayscale processing as the primary comparison group,and combines the online dataset to obtain 4 sets of training sample sets and test sample sets.(2)SVM algorithm based on traditional machine learning for crack detection and recognition: 4 sets of training samples are trained by SVM algorithm of linear kernel function and RBF kernel function respectively,and the corresponding 4 sets of test samples are used for crack detection.(3)CNN algorithm based on deep learning for crack detection and identification: Four sets of training samples and test samples were tested for cracks by migration learning method based on convolutional neural network inception V3 model.And compare and analyze with traditional machine(4)Development based on CNN crack detection platform: Combining machine learning,user graphical interface and image processing,developed a crack detection platform with good scalability,and finally proposed the process of evaluating road health with this procedure.