Autentikasi Pilot Berbasis Pengenalan Wajah Menggunakan Convolutional Neural Network Kustom dengan Deployment TFLite

Most commercial UAV platforms still lack a mandatory pre-flight identity check, leaving ground control stations exposed to impersonation and unauthorized takeover. This paper proposes a facial recognition-based pilot authentication system using a custom four-block Convolutional Neural Network (CNN) to classify three identity classes: Pilot 1, Pilot 2, and Stranger. Trained on 2,125 facial images with data augmentation, Batch Normalization, MaxPooling, and Dropout regularization, the model achieves a test accuracy of 99.07% with a loss of 0.0613. The model is further converted into TensorFlow Lite (TFLite) format for efficient deployment on resource-constrained embedded devices, offering a computationally light alternative to conventional authentication methods as an embedded pre-flight security checkpoint for UAV ground control stations.