Deep Learning Image Classification for Surat Mangyan Script Preservation
Gerhard P. Tan, Aivor C. Padilla, Richmond Bryan A. Supleo, Dianalyn Espiritu, Romalyn Gomez, Marife A. Rosales · 2024
This paper presented an approach to Image Classification focusing on the Surat Mangyan script, an ancient writing system indigenous to the Philippines. Capitalizing on the Deep Learning techniques, mainly on the Convolutional Neural Networks (CNNs), the proposed system aims to accurately identify and classify Surat Mangyan characters. The research explores data augmentation methods to enhance model performance against variations in handwriting styles. The study investigates the impact of different architectural configurations on recognition accuracy, including the number of convolutional layers, kernel sizes, and activation functions. Experimental results demonstrate the effectiveness of the developed system in achieving a high classification accuracy, with a classification accuracy that accounts for 98.92%. The system also exhibits a high sensitivity (at a Macro-Averaged) of 98.98%, with an F1-score of 98.9% arriving at a precision of 98.98%.