Identifikasi Wajah Bermasker dan Bertopi Menggunakan Teknologi Deep Learning Berbasis MobileNetV2

The use of face coverings, such as masks and hats, has become common in public spaces, posing challenges for face recognition systems due to the partial occlusion of distinctive facial features. This study proposes an adaptive face identification system based on MobileNetV2 using a transfer learning approach to identify individuals under different face-covering conditions without requiring the removal of the coverings. The dataset consisted of eight subjects categorized into four conditions: uncovered faces, mask-only, hat-only, and combined mask-and-hat conditions. Face detection was performed using the Haar Cascade Classifier, and the detected facial images were resized to 224 × 224 × 3 pixels. The model was trained for 20 epochs using the Adam optimizer with ImageNet-pretrained weights. The results showed that the model achieved a peak validation accuracy of 96.97% during training, while evaluation on an independent test set of 48 images yielded an overall accuracy of 87.50%. Category-specific accuracy reached 93.75% for mask-only conditions, 87.50% for hat-only conditions, and 81.25% for combined mask-and-hat conditions. These findings indicate that MobileNetV2 has the potential to support face identification under partial occlusion. Its lightweight architecture also offers potential for deployment on edge devices. However, further evaluation using larger and more diverse datasets is needed to establish the model's generalizability.