Transfer Learning Model Evaluation on CNN Algorithm: Indonesian Sign Language System (SIBI)

In Indonesia as much as elsewhere, the deaf can communicate using sign language. The Indonesian Sign Language System (SIBI) is one of the sign language systems used in Indonesia. A model produced by the Convolutional Neural Network (CNN) method can be used in computer science for the recognition of sign language. By using the Transfer Learning paradigm, CNN's performance may be enhanced. However, not many researches have been conducted to assess the effectiveness of transfer learning on sign language models, particularly those that use the TensorFlow library. In fact, the evaluation results can influence the selection of the transfer learning model together with CNN. This study aims to evaluate the efficacy of using the CNN model for SIBI sign language through Transfer Learning. The data used are images of 24 SIBI alphabets and are processed through the TensorFlow library. The images will be recognized through the transfer learning performance of 6 models, namely VGG16, VGG19, Resnet50, Desenet121, Inception-V3 and MobileNet-V2. The results of the study found that through the TensorFlow library, Mobilenetv2 had the highest accuracy of 78% after 20 epochs.