Detection Based on Crack Key Point and Deep Convolutional Neural Network

Abstract

Based on the features of cracks, this research proposes the concept of a crack key point as a method for crack characterization and establishes a model of image crack detection based on the reference anchor points method, named KP-CraNet. Based on ResNet, the last three feature layers are repurposed for the specific task of crack key point feature extraction, named a feature filtration network. The accuracy of the model recognition is controllable and can meet both the pixel-level requirements and the efficiency needs of engineering. In order to verify the rationality and applicability of the image crack detection model in this study, we propose a distribution map of distance. The results for factors of a classical evaluation such as accuracy, recall rate, F1 score, and the distribution map of distance show that the method established in this research can improve crack detection quality and has a strong generalization ability. Our model provides a new method of crack detection based on computer vision technology.