Date of Publication :15th March 2025

Abstract: In the era of digital transformation, the footwear industry has experienced a significant shift toward e -commerce, creating a demand for automated image classification systems. This research presents a memory efficient, generalization -focused transfer learning model for footwear classification designed to handle unseen data while ensuring scalability. The approach leverages pre-trained Convolutional Neural Networks (CNNs) such as ResNet152V2, DenseNet201, NASNetMobile, and InceptionResNetV2, along with data augmentation and dropout techniques to address classification challenges. Transfer learning fine-tunes the features of these models, improving accuracy with limited labeled data. Among the models tested, InceptionResNetV2 achieved the highest accuracy of 98.78%. Applications include automated category labeling, improved recommendation systems, and enhanced search results on e-commerce platforms, ultimately making the online shopping experience more efficient and accurate.

Reference :