Automated Identification and Classification of Teeth Fossils

This study explores the performance of modern, accurate deep-learning algorithms on the classification of fossil teeth in different types of animals such as sharks, elephants, hyrax, and Afradapis. Sharks teeth are typically the most common fossils found in Wadi Al-hitan and they often provide valuable information about the diets and trophic interactions of past organisms, which can be used to infer aspects of the environments they lived in. Taxonomic identification of fossils is usually performed manually in a time-consuming manner and requires considerable expertise due to the vast quantity and high diversity of teeth fossils. Using modern teeth with known categories, deep-learning algorithms can be trained to identify, classify, and extract information about teeth fossils such as determining whether the animal is herbivorous or carnivorous. It also can be applied in other niches such as comparative anatomy. In this paper, a method for fully automated teeth fossils identification and classification is introduced and validated. A pipeline is proposed that begins with the image-capturing process and ends with a classification result. To assess the method performance and dataset representability that was created in cooperation with Mansoura University vertebrae paleontology lab (MUVP). Extensive experiments were performed on our dataset of more than 50,000 images of the above-mentioned 4 classes. Compared with Resnet50 model 82% precision, 80% recall, and an F1 score of 80%, VGG16 model 79% precision, 78% recall, and an F1 score of 78%, the Mobilenetv2 model 70% precision, 66 % recall, and an F1 score of 66 %, and Convolutional Neural Networks (CNN) 60% precision, 57% recall, and an F1 score of 58%. The ResNet50 model had the highest accuracy (80 % ) on the experimental dataset. The system provides a reliable approach for the detection of fossil teeth from 2D images that can be applied in a range of morphology and paleontology contexts for research purposes.