Classification of Engraved Pottery Sherds Mixing Deep-Learning Features by Compact Bilinear Pooling

Abstract The ARCADIA project aims at using pattern recognition and machine learning to promote a systematic analysis of the large corpus of archaeological pottery fragments excavated in Saran (France). Dating from the High Middle Ages, these sherds have been engraved with repeated patterns using a carved wooden wheel. The study of these engraved patterns allows archaeologists to better understand the diffusion of ceramic productions. In this paper, we present a method that classifies patterns of ceramic sherds by combining deep learning-based features extracted from some pre-trained Convolutional Neural Network (CNN) models. A dataset composed of 888 digital patterns extracted from 3D scans of pottery sherds was used to evaluate our approach. The classification capacity of each CNN model was first assessed individually. Then, several combinations of common pooling methods using different classifiers were tested. The best result was obtained when features of the VGG19 and ResNet50 models were combined using Compact Bilinear Pooling (CBP) with a high classification rate of 95.23%.