Utilizing Transfer Learning for Deep Learning Based Image Classification

Abstract

Since the inception of the ImageNet Challenge in 2010, numerous innovative deep CNNs (D-CNNs) have been developed. These models typically require extensive datasets for training, making their application to smaller datasets infrequent due to overfitting risks. This paper introduces an adapted deep neural network model designed to effectively fit a small dataset. The main contribution of this work lies in demonstrating the effective use of transfer learning and fine-tuning techniques to adapt pre-trained deep learning models, such as VGG16, VGG19, Inception V3, and Inception-ResNetV2, for small-scale image classification tasks. By incorporating data augmentation, dropout regularization, and selective layer freezing, the study achieves high classification accuracy while minimizing overfitting. This approach provides a practical solution for leveraging advanced deep learning architectures in scenarios with limited labeled data.