Matching up Stone Tools and Storage Bag Using Image Identification with CNN

Ancient stone tools excavated from ruins are important for archaeological research. Joining materials must be frequently assembled and disassembled to reveal the manufacturing process of stone tools. Each joining part is stored in a bag with a piece of paper writing its identification number and PEAKIT images. However, it is difficult to identify each joining part of a stone tool during assembly or disassembly, which makes stone tool storage and management relatively laborious and time-consuming. This study proposes an image-based identification method to match each stone tool with its corresponding storage bag. For each stone tool, two commonly used stable postures were defined for image acquisition. The collected image dataset was divided into training, validation, and testing sets. The training set provides known information to build a database. An untrained CNN model and seven pre-trained models were used to learn the distinguishing features. The experimental results on the test set showed better accuracy and processing speed than those of previous research.