Pokepedia: Pokemon Image Classification Using Transfer Learning

Abstract

Identifying images of various objects, living creatures, food, etc., and classifying them using machine learning has become a common task in computer vision. Humans may not identify every object they see, here comes machine learning that eases the life of human beings by identifying the object for the human. Poké mon is a cartoon that is widely watched by the majority of the younger generation around the world. The aim of this work to predict and classify Poké mon images using pre-trained models. In the proposed work, seven pre-trained models namely MobileNetV2, EfficientNetB7, EfficientNetV2L, DenseNet201, ResNet101, VGG19 and VGG16 were utilised to classify ten Poké mon characters which includes Pikachu, Raichu, Charmander, Bulbasaur, Squirtle, Eevee, Piplup, Snorlax, Jigglypuff, and Psyduck. The performance of the pre-trained models were evaluated on a dataset collected from the internet. The ResNet101 pre-trained model produces the highest accuracy of 95.60% when compared with the other models.