Facial Recognition and Classification of Terracotta Warriors in the Mausoleum of the First Emperor Using Deep Learning

Abstract

Abstract. The facial features of the Terracotta Warriors unearthed from the Mausoleum of the First Emperor of Qin are authentic depictions of the appearance of soldiers from the same period. Recognizing facial features to classify the Terracotta Warriors is one of the crucial aspects of archaeological research. Due to limitations in the collection of facial samples from the Terracotta Warriors, an enhanced SqueezeNet model is proposed for deep learning facial recognition. The FaceNet backbone feature extraction network has been improved by replacing the initial 7×7 convolution kernel with three 3×3 convolution kernels. The model's feature extraction layer is composed of alternating convolution layers, pooling layers, Fire modules, and pooling layers, with the introduction of an exponential function to smooth the shape of the loss function. Finally, facial classification of 295 Terracotta Warriors is accomplished using Agglomerative Clustering. The model demonstrates a facial recognition accuracy of 95.6%, showing a respective improvement of 4.1% and 2.8% compared to the classical SqueezeNet and Inception_ResNetV1 models. This approach better meets the requirements for facial recognition and classification of Terracotta Warriors, providing intelligent and efficient technical support for technological archaeology.