Investigation of Transfer Learning for Image Classification and Impact on Training Sample Size for Image Classification Tasks

Recent developments in deep learning have brought huge breakthroughs in the image processing area, which triggered numerous successful applications and positively impacted the current big data context of Industry 4.0. On the other hand, it is widely known that large amounts of training data are required to train a deep learning model with millions of parameters from scratch, which limits its application in many industrial applications where sufficient labeled data are often lacking. Transfer learning is one of the practical solutions to reduce the data required for training, which tries to reuse learned knowledge for similar tasks. Nevertheless, many technical details of transfer learning implementation are not well documented.

In this talk, different transfer learning approaches and implementation details for high-performance model building under the constraint of limited available training data are investigated using two different industrial use-cases. Various methods of transfer learning are compared and important technical details are discussed. Through this study, the minimum number of training samples can be estimated and practical guidelines for the development of image classification models with limited data resources are summarized.