Tsinghua bamboo slip scribe verification using Siamese networks

Abstract The exploration of bamboo slip manuscripts and their historical and cultural significance has become increasingly challenging. This paper proposes a deep learning method for scribe verification of Warring States bamboo slips, aiming to identify whether slips were transcribed by the same writer. Based on the Siamese network, we improved the original MobileNet_V3 to MobileNet_V3+, integrating a Squeeze-and-Excitation attention mechanism for feature extraction and weighting. Using a curated dataset from Tsinghua University’s collection, enhanced by data augmentation to increase the number of samples and balance between positive and negative sample pairs, the model achieves 90.2% verification accuracy with an area under the receiver operating characteristic curve of 0.96. Tests on slips with unclear attribution further confirm the model’s effectiveness, offering a new computational approach to historical manuscript analysis.