This paper seeks to establish a standard deep learning-powered method for studying buccal dental microwear patterns on late prehistoric individuals in the northeastern Iberian Peninsula. The paper adopts a dataset consisting of 88 buccal surface micrographs of archeological materials, which is extended to 528 images and categorized into four diets: Agriculturalists, Gatherers, Hunters, and Fishermen. Image preprocessing is performed with resizing, normalization, and data augmentation. Although models like MobileNetV2, ResNet50 InceptionV3, and EfficientNetB0 have been used, transfer learning with the pretrained MobileNetV2 model is utilized for diet classification feature extraction. The MobileNetV2 model attains a validation accuracy of 80.49% and generalizes to all four diet classes. Deep learning is more efficient and precise than traditional microwear analysis methods in determining dietary trends. Despite the issues of small dataset size and inherent variability of microwear patterns, the approach provides a more objective, faster, and more scalable method of analyzing large datasets and drawing more complex subsistence strategies from ancient societies.