A Hybrid IoT Edge-Cloud Based Lightweight Deep Learning Framework for Automated Date Palm Leaf Disease Diagnosis — Research Artifacts

PalmNet Research Artifacts This repository provides the complete research artifacts associated with PalmNet, a hybrid IoT edge–cloud framework for automated date palm leaf disease diagnosis based on knowledge distillation, confidence calibration, and threshold-optimized inference routing. Repository Contents Trained deep learning models, including: ConvNeXt-Tiny teacher model MobileNetV3, EfficientNet-Lite0, and ShuffleNetV2 baseline student models Knowledge-distilled ShuffleNetV2 student model Deployment-ready model exports: ONNX models for Raspberry Pi Zero 2 W deployment TensorFlow Lite (FP16 and FP32) models for Android deployment Complete training, validation, and evaluation source code (Python scripts and Jupyter notebooks) Raspberry Pi edge-node implementation featuring: Hybrid local/cloud inference Confidence-based routing GPIO button triggering GPS-based geotagging FastAPI cloud inference service with Docker containerization Full Android Studio project (Kotlin and Jetpack Compose) including: Viewer interface Expert review interface Experimental results and supplementary materials: Classification performance evaluation Confidence calibration analysis Threshold sensitivity analysis Grad-CAM visualizations Thesis methodology and results figures Diagnostic Classes The framework supports nine diagnostic classes: Potassium Deficiency Manganese Deficiency Magnesium Deficiency Black Scorch Leaf Spots Fusarium Wilt Rachis Blight Parlatoria blanchardi Healthy Sample Experimental Configuration Teacher Model: ConvNeXt-Tiny (Weighted Cross-Entropy) Student Model: ShuffleNetV2 ×1.0 Knowledge Distillation: Standard KD (Temperature = 2, α = 0.7) Calibration Temperature: T_cal = 1.3976 Routing Threshold: τ = 0.93 Dataset Availability The original image dataset is not redistributed in this repository because it is a third-party published benchmark dataset. Researchers should obtain the dataset from the original source: Namoun et al., Data in Brief, 2024. DOI: 10.1016/j.dib.2024.110933. Citation If you use these research artifacts, please cite the associated thesis, publication, and this Zenodo record.