Analisis Efektifitas Inception-EfficientNet Architecture untuk Klasifikasi Identitas Kartu Tanda Mahasiswa dengan Pendekatan OCR dan Faster R-CNN

Abstract

Student Identity Card has an important role as an individual identification tool in the higher education environment. KTM is often the main tool in identity verification process to access campus facilities, such as library services. However, currently the access system to the library still relies on the manual method which involves staff entering data manually into a computer which is prone to typing errors. This study aims to design an identity recognition system on KTM by combining OCR method and involving the Inception-EfficientNet architecture on the Faster R-CNN model in classifying text and photo found on KTM. The Inception-EfficientNet architecture is designed with 5 convolution and 2 maxpooling layers, and involves RPN (Region Proposal Network) and ROI Pooling which researchers have designed as important elements in establishing the Faster R-CNN method. The data collected includes three classes, namely hafizh, richo, and vandy. The test results of the OCR method in recognizing identities based on characters show a high level of accuracy, which is equal to 98.35%. On the other hand, the Faster R-CNN method is able to classify photos with very good performance which achieves a success accuracy of 91.83%. Based on testing the entire system that combines the OCR method and the Faster R-CNN method, carried out on 10 different data samples, and managed to achieve a success rate of 90%. These findings emphasize the fact that a collaborative approach between the OCR method and Faster R-CNN model has the potential to increase accessibility and reliability KTM identity recognition.