Simultaneous Detection of Sculpture Defects Based on Vision and Positioning Method

Abstract

The crack detection and effective maintenance of sculpture cultural relics have attracted more and more attention. However, for surface defects. Autonomous detection is mostly based on vision. The detection range of this type of method is limited to a common defect with a large crack width and easy identification. The conditions are too harsh. In reality, various types of defects usually appear, and they only occupy a small part of the inspection image. In addition, the difference between the parameters and the surrounding image parameters is small, which can easily lead to missed detection and false detection. In addition, most of the current researches only focus on defect detection. Little attention is paid to defect positioning, and this is the indispensable information for repairing and protecting sculptures. Part of the research proposed GPS positioning, but GPS signals are easily lost in a relatively complex geographic environment, and its infrastructure is not reliable and will increase Positioning costs. In this regard, this paper proposes a vision-based defect detection and positioning network method, which can be used in harsh conditions Detect, and locate defects, Which also set A supervised Deep Convolutional Neural Network is calculated. This paper also creates a training method to optimize its performance on the neural network. Experiments show the detection accuracy of this method is 80.7%, and the positioning accuracy of each image is 86% at 0.41 s (in the field. In the test experiment, it is 1200 pixels).