Development of Lampung Script Characters Recognition Model using TensorFlow

Abstract

In the face of cultural erosion, particularly the dwindling proficiency in deciphering Lampung characters, this research pioneers an innovative approach to cultural preservation. The Lampung character recognition model was developed using TensorFlow, a robust computer vision and machine learning framework. Convolutional Neural Networks (CNN) are integrated to enhance the image processing capabilities. The research employs the Design Science Research methodology, emphasizing problem identification, solution objectives, design and development, demonstration, evaluation, and communication. The dataset, comprising 3900 instances, is meticulously collected and features diverse Lampung script writing. Through preprocessing and classification, the model undergoes training with an 80:10:10 split for training, validation, and test data. The architecture includes CNN layers with ReLu activation functions, and transfer learning is employed using the MobileNet V2 network model. Demonstrating commendable performance, the model achieves an accuracy spectrum of 0.652 to 0.998. The research not only underscores the viability of the TensorFlow model but also establishes a foundation for future explorations in preserving Lampung cultural heritage. This intersection of advanced machine learning and cultural preservation signifies a promising synergy, ensuring the enduring legacy of Lampung characters amid societal and technological transformations.

Keywords

• lampung script
• cnn
• machine learning recognition
• tensorflow