Improvement of Performance Related to Cross Dataset Handwritten Recognition Based on Transfer Learning

DOI

10.2991/978-94-6463-300-9_68How to use a DOI?

Keywords

Handwritten Recognition; Convolutional Neural Network; Transfer Learning

Abstract

Given the escalating diversity and intricacy of handwritten samples, it remains challenging to enhance the accuracy and robustness of recognition algorithms. This study proposed a solution to this problem by optimizing a pre-existing Convolutional Neural Network (CNN) model for handwritten recognition using a new dataset. Initially, the model was trained on the MNIST dataset. It was then fine-tuned using a self-collected dataset with diverse handwritten styles, employing transfer learning to capitalize on existing knowledge and alleviate the training load. This approach involved freezing certain layers in the pre-trained model to prevent overfitting and encourage more generalized feature extraction. The model exhibited high accuracy on the MNIST dataset, with a training accuracy of 99.07% and a testing accuracy of 99.29%. However, when tested on the self-collected dataset, the accuracy dropped to 11.89%. After applying transfer learning, the model achieved an improved testing accuracy of 38.46% on the self-collected dataset. Despite this improvement, the significant gap between training and testing accuracy indicated overfitting, suggesting the need for additional strategies to enhance model generalization. This study establishes a foundation for future work on improving CNN model performance on diverse handwritten digit datasets.

Copyright

© 2023 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Kaiyuan Chen
PY  - 2023
DA  - 2023/11/27
TI  - Improvement of Performance Related to Cross Dataset Handwritten Recognition Based on Transfer Learning
BT  - Proceedings of the 2023 International Conference on Image, Algorithms and Artificial Intelligence (ICIAAI 2023)
PB  - Atlantis Press
SP  - 656
EP  - 662
SN  - 2352-538X
UR  - https://doi.org/10.2991/978-94-6463-300-9_68
DO  - 10.2991/978-94-6463-300-9_68
ID  - Chen2023
ER  -