Ruzindana/cropsense-mobilenetv2

CropSense MobileNetV2 - Crop Disease Detection Model

Model Description

This is a MobileNetV2-based deep learning model trained to detect crop diseases in leaf images. The model is specifically designed for smallholder farmers in Rwanda to identify common crop diseases.

Model Type: Image Classification
Framework: Keras/TensorFlow
Architecture: MobileNetV2
Input: 224x224 RGB images
Output: Disease classification (Healthy, Powdery, Rust)

Model Details

• Model Size: ~14 MB
• Classes: 3 (Healthy, Powdery, Rust)
• Input Shape: (224, 224, 3)
• Optimized for: Mobile and edge devices

Usage

Using Python

from huggingface_hub import hf_hub_download
import tensorflow as tf
from PIL import Image
import numpy as np

# Download model from Hugging Face
model_path = hf_hub_download(
    repo_id="Ruzindana/cropsense-mobilenetv2",
    filename="best_MobileNetV2.keras"
)

# Load the model
model = tf.keras.models.load_model(model_path)

# Load class names
import json
with open(hf_hub_download(
    repo_id="Ruzindana/cropsense-mobilenetv2",
    filename="model_metadata.json"
)) as f:
    metadata = json.load(f)
    class_names = metadata.get("classes", ["Healthy", "Powdery", "Rust"])

# Preprocess image
def preprocess_image(image_path):
    img = Image.open(image_path).resize((224, 224))
    img_array = np.array(img) / 255.0
    img_array = np.expand_dims(img_array, axis=0)
    return img_array

# Make prediction
image = preprocess_image("path/to/leaf_image.jpg")
predictions = model.predict(image)
predicted_class_idx = np.argmax(predictions[0])
confidence = float(predictions[0][predicted_class_idx])
predicted_class = class_names[predicted_class_idx]

print(f"Prediction: {predicted_class} ({confidence*100:.2f}% confidence)")

Using FastAPI Backend

Set the environment variable:

export HUGGINGFACE_MODEL_ID="Ruzindana/cropsense-mobilenetv2"

The backend will automatically download and use the model.

Model Performance

• Accuracy: Optimized for mobile deployment
• Inference Speed: Fast inference on CPU and mobile devices
• Use Case: Real-time crop disease detection in field conditions

Training Data

The model was trained on a dataset of crop leaf images from Rwanda, focusing on:

• Healthy crop leaves
• Powdery mildew infected leaves
• Rust disease infected leaves

Limitations

• Trained specifically for certain crop types common in Rwanda
• Best results with clear, well-lit leaf images
• May require retraining for different geographic regions or crop varieties

Citation

If you use this model, please cite:

@model{cropsense-mobilenetv2,
  author = {Ruzindana, Diana},
  title = {CropSense MobileNetV2 - Crop Disease Detection Model},
  year = {2025},
  url = {https://huggingface.co/Ruzindana/cropsense-mobilenetv2}
}

License

MIT License - See LICENSE file for details

Contact

For questions or support, please contact the model maintainer.