nihal4/rice-disease-model

A fine-tuned ResNet50 model that classifies rice leaf diseases from photos, exported to ONNX for fast CPU inference. Built to power a Telegram bot that delivers treatment advice in Bangla for Bangladeshi farmers.

---

Model Details

Property	Value
Architecture	ResNet50 (fine-tuned)
Format	ONNX
Input	RGB image, 224×224
Output	4-class softmax
Best Val Accuracy	98.47%
Inference	CPU (ONNX Runtime)

---

Classes

Index	Class	Bangla
0	Bacterial_Leaf_Blight	ব্যাকটেরিয়াল লিফ ব্লাইট
1	Brown_Spot	ব্রাউন স্পট
2	Leaf_Blast	লিফ ব্লাস্ট
3	Tungro	টুংরো

---

Performance

Per-Class Accuracy on Test Set

Class	Accuracy
Bacterial Leaf Blight	98.5%
Brown Spot	97.0%
Leaf Blast	95.5%
Tungro	100.0%

Confusion Matrix (Normalized)

Training Curve

ResNet50 trained for 45 epochs with best validation accuracy of 98.47%.

---

Dataset

Merged from two publicly available sources:

Source	Link
Rice Leaf Bacterial and Fungal Disease Dataset (Hasan et al., 2023)	Mendeley Data
Rice Leaf Disease Image	Kaggle / Nirmal Sankalana

Class	Total Images
Bacterial_Leaf_Blight	1,308
Brown Spot	1,308
Leaf_Blast	1,308
Tungro	1,308
Total	5,232

Balanced: All classes were downsampled to 1,308 images (the size of the smallest class, Tungro) to prevent class imbalance during training.

Sample Images

---

Usage

import onnxruntime as ort
import numpy as np
from PIL import Image

# Load model
session = ort.InferenceSession("model.onnx")

class_names = [
    "Bacterial_Leaf_Blight",
    "Brown_Spot",
    "Leaf_Blast",
    "Tungro"
]

def preprocess(image_path):
    img = Image.open(image_path).convert("RGB").resize((224, 224))
    img = np.array(img, dtype=np.float32) / 255.0
    img = (img - [0.485, 0.456, 0.406]) / [0.229, 0.224, 0.225]
    return img.transpose(2, 0, 1)[np.newaxis].astype(np.float32)

def predict(image_path):
    input_tensor = preprocess(image_path)
    outputs = session.run(None, {"input": input_tensor})
    probs = outputs[0][0]
    idx = np.argmax(probs)
    return class_names[idx], float(probs[idx])

label, confidence = predict("rice_leaf.jpg")
print(f"{label} ({confidence:.1%} confidence)")

---

Telegram Bot

This model powers a Telegram bot that lets farmers send a photo of a diseased rice leaf and receive instant diagnosis and treatment advice in Bangla.

🔗 Try the bot →
🌐 Project demo page →
💻 Source code →

---

Citation

If you use this model, please cite:

@misc{rice-leaf-disease-classifier,
  author = {nihal4},
  title  = {Rice Leaf Disease Classifier},
  year   = {2026},
  url    = {https://huggingface.co/nihal4/rice-leaf-disease-classifier}
}

---

Made with ♥ for Bangladeshi rice farmers

Dataset Citations

If you use the training data, please cite the original sources:

Mendeley Data (Bangladesh dataset):

Hasan, Mehedi; Khatun, Sonia; Raihan, Md. Abu; Uddin, Abdul Hasib (2023),
"Rice Leaf Bacterial and Fungal Disease Dataset",
Mendeley Data, V2, doi: 10.17632/hx6f852hw4.2

Kaggle (Nirmal dataset):

Nirmal Sankalana, "Rice Leaf Disease Image", Kaggle. https://www.kaggle.com/datasets/nirmalsankalana/rice-leaf-disease-image