Lightweight Deep Learning Model Optimization for Apple Disease Classification Application

International Conference on Applied Artificial Intelligence, Data Engineering and Sciences 2025 · Indonesia · 2025

M. Aldiki Febriantono, Nasywa Hilmi Cahyono, Brian Marcellino

Compared lightweight architectures (MobileNetV1, MobileNetV2, MobileNetV3, MnasNet, TinyNAS) on a custom Poncokusumo apple-orchard dataset (6 classes, ~10,668 images after augmentation). Reports accuracy, precision, recall, F1, MIoU, and training time, then validates the best model in a mobile application.

lightweight networksMobileNetMnasNetTinyNASapple diseasemobile application