Application of Deep Learning to Crack Segmentation in Neutron CT Images of Ancient Writing Knife

摘要: Neutron CT imaging offers unique advantages in metal defect detection and cultural relic analysis, particularly for exploring internal structures of ancient artifacts like writing knives, due to its high penetrability and hydrogen sensitivity. Accurate segmentation of its images is critical for defect detection, with crack segmentation in writing knife images being key to understanding craftsmanship and preservation. However, complex structures in these images—neutron scattering noise, blurred multi-material interfaces and overlapping gray scales hinder precise crack segmentation. Traditional algorithms, reliant on manual tuning and single-feature extraction, lack accuracy: they roughly distinguish macroscopic structures but fail to segment fine cracks in blade edges. This study addresses this by applying deep learning to crack segmentation in writing knife neutron CT images, using BSEResU-Net (a residual U-Net with SE attention). Trained on a small manually annotated dataset of The Western Han writing knife from China Spallation Neutron Source (CSNS), the model was validated on full-knife crack segmentation. Results show its superiority, it obtained an Area Under the ROC Curve (AUC) of 0.9793 and an F1 score of 0.9089 on the dataset, accurately capturing fine cracks. Compared with threshold segmentation, about 70% more cracks can be segmented. This framework resolves neutron data scarcity, provides an innovative solution for cultural heritage defect detection, and advances deep learning in multimodal penetrating imaging.