JoQert/argon-bilig-olive-models

Read in: 🇬🇧 English | 🇹🇷 Türkçe

🫒 Argon-Bilig: Gemlik Olive AI Diagnostic and Decision Support Engine

Argon-Bilig is a hybrid artificial intelligence diagnostic system specifically tailored for the Gemlik olive cultivar (Olea europaea cv. 'Gemlik'). It detects Olive Peacock Spot (Venturia oleaginea) and Olive Rust Mite (Aculus olearius) deformations at pixel-level precision, eliminates complex background environmental noise, and generates phenologically aligned phytosanitary prescriptions.

This research was developed within the scope of the TÜBİTAK 2204-A High School Research Projects Competition and the TEKNOFEST Agricultural Technologies Competition.

---

👥 Authors & Ownership

• Muhammed Emin DUYAR
• Ahmet Buğra GİRGİN
• Year: 2026

---

📦 Companion Benchmark Dataset

These weights were trained and benchmarked on the officially standardized Argon-Bilig reference dataset:

• 🔗 Dataset Hub: JoQert/argon-bilig-olive-dataset

---

📊 1. Model Weights & Real-time CPU Inference Benchmarks

Benchmarks were evaluated on a standard multi-threaded CPU environment in real-time:

Model Component	Filename	Architecture	File Size	Input Resolution	Avg. CPU Latency
Model 1 (Segmentation)	yolov8_zeytin_seg.onnx	YOLOv8s-seg	45.18 MB	640x640 px	387.3 ms
Model 2 (Classification)	efficientnet_classifier.onnx	EfficientNet-B0	15.79 MB	224x224 px	16.1 ms
Total Pipeline	Hybrid Sequential Flow	Two-Stage ONNX	60.97 MB	-	403.4 ms

---

🏷️ 2. Classes, Distribution & Class Weights

To resolve class imbalance during training, Weighted Cross-Entropy Loss was utilized:

Class ID	Class Label	Target Pathology / Organism	Train Split	Val Split	Loss Weight
0	Healthy	Healthy Olive Leaf (Olea europaea)	1,033 leaves	259 leaves	0.889
1	acerculus_olearius	Olive Rust Mite (Aculus olearius) *	398 leaves	100 leaves	2.308
2	olive_peacock_spot	Olive Peacock Spot (Venturia oleaginea)	1,325 leaves	332 leaves	0.693

📌 Taxonomy Note on Class Labels:
The directory and code label acerculus_olearius represents the pest Olive Rust Mite (Aculus olearius, Family: Eriophyidae). The key label is maintained for programmatic integrity with trained checkpoints.

---

📈 3. Validation Performance Scorecard

Empirical evaluation on the holdout validation set (691 samples):

Class	Precision	Recall	F1-Score	Support
Healthy Olive Leaf	99.23% (0.9923)	99.23% (0.9923)	99.23% (0.9923)	259
Olive Rust Mite	99.00% (0.9900)	99.00% (0.9900)	99.00% (0.9900)	100
Olive Peacock Spot	99.40% (0.9940)	99.40% (0.9940)	99.40% (0.9940)	332
Overall Accuracy	-	-	99.28%	691

Agri-Safety Engineering Note: In agricultural diagnostic systems, the most critical performance metric is Recall (Sensitivity) to prevent false negatives (failing to identify diseased foliage). Our hybrid architecture minimizes false negative risks across all pathology classes.

---

🧠 Hybrid Decision Pipeline Architecture

Raw Leaf Input (Field Image)
       │
       ▼
[Model 1: YOLOv8s-seg] ──(Lesion Area > 0.5%)──► [Peacock Spot Diagnosis & Severity Rating]
       │
 (No Peacock Spot)
       │
[Background Masking & Tight Crop]
       │
       ▼
[Model 2: EfficientNet-B0] ────────────────────► [Olive Rust Mite OR Healthy Leaf]

---

🎯 Target Pathologies & Phytosanitary Prescriptions

Class Label	Pathogen / Pest	Common Name	Agronomic Prescription & Control Strategy
olive_peacock_spot	Venturia oleaginea	Olive Peacock Spot	1.0% - 1.5% Bordeaux Mixture or registered copper-based fungicides before seasonal rains.
acerculus_olearius	Aculus olearius	Olive Rust Mite	Application of Wettable Sulfur (WP) or registered acaricides during vegetative shoot flushes.
Healthy	Olea europaea	Healthy Foliage	Routine canopy aeration pruning, balanced N-P-K fertigation, and drip irrigation.

---

💻 Python Quickstart Inference (ONNX Runtime)

import onnxruntime as ort
import numpy as np
import cv2

# Initialize ONNX inference session
session = ort.InferenceSession("efficientnet_classifier.onnx", providers=['CPUExecutionProvider'])
classes = ['Healthy', 'acerculus_olearius', 'olive_peacock_spot']

# Load and preprocess image (224x224 RGB)
img = cv2.imread("sample_leaf.jpg")
img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
resized = cv2.resize(img_rgb, (224, 224)).astype(np.float32) / 255.0

# Standard normalization
mean = np.array([0.485, 0.456, 0.406], dtype=np.float32)
std = np.array([0.229, 0.224, 0.225], dtype=np.float32)
normalized = (resized - mean) / std
input_tensor = np.transpose(normalized, (2, 0, 1))[np.newaxis, ...]

# Run inference
input_name = session.get_inputs()[0].name
logits = session.run(None, {input_name: input_tensor})[0][0]
probs = np.exp(logits) / np.sum(np.exp(logits))

pred_class = classes[np.argmax(probs)]
print(f"Diagnosis: {pred_class} (Confidence: {np.max(probs)*100:.2f}%)")

---

📖 Citation

If you use these models or weights in your research, please cite:

@misc{argonbilig_models2026,
  title={Argon-Bilig: Hybrid AI Diagnostic and Decision Support Engine for Gemlik Olive Diseases},
  author={Duyar, Muhammed Emin and Girgin, Ahmet Bu{\u{g}}ra},
  year={2026},
  howpublished={Hugging Face Hub: \url{https://huggingface.co/JoQert/argon-bilig-olive-models}},
  note={Pretrained ONNX weights and hybrid diagnostic pipeline}
}

---

🙏 Acknowledgements

We gratefully acknowledge the researchers and open datasets that supported preliminary model exploration:

• Olive Peacock Eye Dataset: Diker, A., Elen, A., Közkurt, C. et al. (2024). Eur Food Res Technol, 250, 287–299. DOI: 10.1007/s00217-023-04386-8
• Olive Leaf Disease Unified Dataset: Abeer Saleh (CC0: Public Domain)
• Olive Leaf Image Dataset: Habibul Basher (CC0: Public Domain)