A Metal Surface Damage Recognition Method For Augmented Reality Assisted Maintenance Systems

The small damages such as cracks and scratches on the surface of aerospace products pose a serious threat to the safety of life and property, and manual visual inspection is prone to omissions, leaving great safety hazards. Using augmented reality (AR) assisted maintenance systems to assist visual inspection is one of the effective solutions. However, the limitations of computing power in augmented reality devices and the real-time requirements of augmented reality pose significant challenges to small-scale object detection algorithms. Therefore, this paper proposed a metal surface damage recognition method for augmented reality assisted maintenance system. Firstly, for the appearance characteristics of surface damage in the steel image database NEU-CLS, the histogram equalization was employed for image enhancement to improve image quality. Afterwards, a SURF + K-means + Bag-of-Features + the-number-of-feature-points feature extraction and dimensionality reduction method was proposed to improve recognition efficiency while ensuring the robustness of the method. Finally, adaptive boosting learning framework was utilized to construct a surface damage recognition model which has good accuracy and efficiency for common metal surface damages.