Non-Destructive Taxonomic Identification of Cenomanian-Coniacian Ammonites in Colombia Using Artificial Intelligence and Deep Learning

Abstract

This paper introduces an innovative methodological framework for the automated and non-destructive taxonomic identification of Cenomanian-Coniacian ammonites from Upper Cretaceous formations in Colombia using artificial intelligence. The taxonomic classification of macrofossils—particularly ammonites—is traditionally a time-consuming and expertise-dependent task, often limited by preservation quality, morphological plasticity, and the scarcity of digitized reference collections. These constraints have hindered the full integration of ammonite-bearing units such as Hondita, Loma Gorda, and Frontera into 
regional biostratigraphic models. To address this gap, we propose an image classification model based on a convolutional neural network (ResNet-18 with transfer learning), trained on expert-labeled fossil images enriched with stratigraphic metadata, which yields promising results in classification accuracy and morphological interpretability. The approach builds on a systematic review of recent AI applications in paleontology, including fossil segmentation, synthetic dataset generation, and non-invasive imaging
techniques such as micro-CT and laser-stimulated fluorescence. Beyond classification, the framework enables the integration of geochemical and stratigraphic data to refine chronostratigraphy and reconstruct paleoenvironments, while offering a scalable solution for digitizing and curating fossil collections in resource-constrained settings. By embedding machine learning into the fossil identification workflow, this study contributes to the modernization of paleontological research and promotes the inclusion of
Colombia’s fossil heritage in global geoscientific databases.