Image Classification of Persian Handwritten Digits Using Simplified VGG16

This study aims to use a Convolutional Neural Network (CNN) with a simplified VGG16 architecture to classify Persian handwritten digits. The main challenge lies in recognizing the unique characteristics of Persian digits, which have complex and varied structures. Leveraging neural network technology, this project focuses on developing a model capable of accurately recognizing Persian digits. The dataset used consists of 150,000 Persian digit images, with 15,000 images per class. The results show that the difference between tests without Hyperparameter Tuning and with Hyperparameter Tuning is minimal, with loss and accuracy values of 0.0086 and 0.9972, respectively, for tests without tuning, and 0.0057 and 0.9979 for tests with tuning. These findings indicate that even with the base CNN model using the simplified VGG16 architecture, it is sufficient for effective Image classification in this major task.