In recent years, machine learning has been applied in various fields and applications, and its use continues to increase. Machine learning models discover patterns and relationships in sample inputs as well as in the responses expected from the data. The revealed relationships are used for a new unknown data set to automate the solution of intelligent problems in pattern recognition, prediction or decision making. Complex tasks require a lot of data and time to train, and the data needs to be labeled for supervised training. Transfer learning is a technique used to transfer knowledge gained from one task to solve another. This approach makes it possible to reduce the amount of initial data and training time or increase the accuracy of solving an intellectual problem. This paper presents the results of experimental studies on the static hand gesture recognition based on our proposed deep neural network model, with traditional full learning on all parameters, and a convolutional neural network of the VGG-16 architecture, pre-trained using the concept of transfer learning. Adam optimizer is used as a gradient descent algorithm when training a deep neural network. The back propagation algorithm is used to fine-tune the pre-trained neural network classifier to get updated weights and improve the accuracy of gesture recognition. Software implementation of the gesture recognition system was made using Python image processing libraries obtained from the image capture sensor. Neural networks are designed using TensorFlow and Keras deep learning frameworks. The performance of the proposed deep neural network model is compared with the model for the modified VGG-16 architecture and the transfer learning.
APPLYING OF THE TRANSFER LEARNING TO CONVOLUTIONAL NEURAL NETWORKS FOR IMAGE CLASSIFICATION
Published March 2023
246
105
Abstract
Language
Русский
How to Cite
[1]
Satybaldina Д., Kalymova К. and Sydykov Д. 2023. APPLYING OF THE TRANSFER LEARNING TO CONVOLUTIONAL NEURAL NETWORKS FOR IMAGE CLASSIFICATION. Bulletin of Abai KazNPU. Series of Physical and mathematical sciences. 81, 1 (Mar. 2023), 159–169. DOI:https://doi.org/10.51889/2959-5894.2023.81.1.018.