Convolutional Neural Network And Its Application In Image Crack Recognition

With the explosive growth and improvement of data volume and the improvement of computer hardware level and operational ability,the deep learning with convolutional neural network(CNN)as the main framework has gradually developed and been applied in many fields.The automatic recognition of crack images plays a vital role in road or tunnel maintenance or in the detection of fractured reservoirs.Therefore,it is also meaningful to apply CNN to the recognition of crack images.The main work of this paper consists of the following aspects:Firstly,in order to overcome some shortcomings that may exist in the optimization algorithm in CNN,an evolutionary gradient algorithm combining genetic ideology with gradient descent method is proposed,and its feasibility and superiority is shown by the experiment of CNN in handwriting recognition.Secondly,the proposed evolutionary gradient algorithm is applied to the recognition of crack images.Based on both the color change of crack images and the shape of cracks in images,an improved superpixel segmentation algorithm is proposed for extracting basic information such as crack location and shape in crack images.Thirdly,in order to realize the end-to-end recognition process for the crack image,the semantic segmentation algorithm,Segnet,based on convolutional neural network is used to train the crack image,and a large-scale precision label is established before the training.The crack database is used for the training of the complex network Segnet.The training results are better and the exact position of the crack in the image can be directly obtained.In general,the improved CNN proposed in this paper has achieved good results in dealing with handwritten digit recognition and crack image recognition,and uses the semantic segmentation algorithm Segnet to identify the crack end-to-end,the performance is better.