Research data on DKDP crystal laser damage detection based on machine learning and image processing

Deuterated Potassium Dihydrogen Phosphate (DKDP) is currently the only nonlinear optical crystal material used in Inertial Confinement Fusion (ICF) laser-driven systems. Variations in image detection techniques during DKDP crystal laser-induced damage testing often introduce errors in image processing. To address this challenge, a novel detection method integrating machine learning and image processing is proposed, enabling rapid identification of diverse damage patterns and precise quantification of critical parameters (e.g., damage point counts and densities) in DKDP crystal images. By leveraging local two-dimensional pattern features extracted from crystal damage images, a Support Vector Machine (SVM) is employed for damage-type classification. For heterogeneous damage images, advanced image processing techniques—including image differencing, high-pass filtering, and high-boost filtering—are applied to significantly enhance the local contrast of subtle damage features. Thresholding and regional image segmentation methods are further utilized to extract edge pixels of damage points, enabling systematic classification, labeling, and statistical analysis. Experimental results demonstrate that the proposed method achieves a classification accuracy of 97.00%, with a damage point counting accuracy of 92.30% for DKDP crystals. The system processes each image within 0.9 s, providing a robust technical foundation for automated damage detection in DKDP crystals. This research contributes to quality control in high-energy laser applications by improving detection efficiency and accuracy.