A Novel Method Based on Convolutional Neural Network for Surface Flaw Point Size Calibration of Optics

Large aperture optics are widely used in high-power solid-state laser devices but are prone to damage and contamination, which degrade transmission performance and shorten device lifespan. Due to their complex fabrication, these optics are costly, making periodic repairs can reduce costs and improve the reliability operation. To facilitate efficient automatic repair, this work proposed a method based on Convolutional Neural Network (CNN)to calibrate the surface flaw points of optics. The size calibration model of flaws based on CNN is established and optimized. Optimal parameters for damage points calibration are determined as 0- 500μm size range, 30ms exposure time; and for contaminant points calibration model are determined as 0- 100μm size range, 20ms exposure time. Based on the dataset’s long-tail features, the feature distribution smoothing method is used for data equalization. Finally, a comprehensive size calibration strategy is proposed. The CNN size calibration is used for small size, and the pixel-level size calibration for large size. The size calibration accuracy of damage points is improved by about 40%, and for contaminant points is 30%. This work can be used for accurate size calibration of surface damage and contaminants on large aperture optics and provide technical support for its restoration.