ugahost991/plant-disease-detector

🌱 Plant Disease Detection - Vector Database

This dataset contains pre-computed embeddings and FAISS index for plant disease detection using a RAG (Retrieval-Augmented Generation) approach.

📊 Dataset Overview

• Total Images: 54,304
• Disease Classes: 38
• Embedding Model: ResNet-50 (Microsoft)
• Embedding Dimension: 2048
• Index Type: FAISS IndexFlatIP (Cosine Similarity)

📁 Files

File	Size	Description
embeddings.npy	~440 MB	Pre-computed 2048-D feature vectors for all images
metadata.json	~8 MB	Image paths, disease classes, and split information
faiss_index.bin	~440 MB	FAISS similarity search index

🌿 Supported Plant Diseases

The dataset covers 38 disease classes across multiple crops:

Tomato (10 classes)

• Target Spot
• Septoria Leaf Spot
• Early Blight
• Late Blight
• Bacterial Spot
• Leaf Mold
• Spider Mites
• Yellow Leaf Curl Virus
• Mosaic Virus
• Healthy

Pepper/Bell Pepper (2 classes)

• Bacterial Spot
• Healthy

Corn/Maize (4 classes)

• Northern Leaf Blight
• Common Rust
• Cercospora Leaf Spot
• Healthy

Potato (3 classes)

• Early Blight
• Late Blight
• Healthy

Apple (4 classes)

• Apple Scab
• Black Rot
• Cedar Apple Rust
• Healthy

Grape (4 classes)

• Black Rot
• Esca (Black Measles)
• Leaf Blight
• Healthy

Cherry (2 classes)

• Powdery Mildew
• Healthy

Strawberry (2 classes)

• Leaf Scorch
• Healthy

Other Crops

• Peach, Raspberry, Soybean, Squash, and more

🚀 Usage

Load the Database

import numpy as np
import json
import faiss

# Load embeddings
embeddings = np.load('embeddings.npy')
print(f"Embeddings shape: {embeddings.shape}")  # (54304, 2048)

# Load metadata
with open('metadata.json', 'r') as f:
    metadata = json.load(f)
print(f"Total images: {len(metadata)}")

# Load FAISS index
index = faiss.read_index('faiss_index.bin')
print(f"Index size: {index.ntotal}")

Search for Similar Images

from transformers import AutoImageProcessor, AutoModel
from PIL import Image
import torch

# Load model
processor = AutoImageProcessor.from_pretrained("microsoft/resnet-50")
model = AutoModel.from_pretrained("microsoft/resnet-50")

# Process query image
image = Image.open("plant_leaf.jpg")
inputs = processor(images=image, return_tensors="pt")

# Extract features
with torch.no_grad():
    outputs = model(**inputs)
    query_embedding = outputs.pooler_output.numpy()

# Normalize
query_embedding = query_embedding / np.linalg.norm(query_embedding)

# Search
k = 5  # Top 5 similar images
distances, indices = index.search(query_embedding, k)

# Get results
for i, idx in enumerate(indices[0]):
    similar_image = metadata[idx]
    print(f"{i+1}. {similar_image['class']} - Similarity: {distances[0][i]:.4f}")

Use with FastAPI

from fastapi import FastAPI, File, UploadFile
import numpy as np
import faiss

app = FastAPI()

# Load database on startup
embeddings = np.load('embeddings.npy')
metadata = json.load(open('metadata.json'))
index = faiss.read_index('faiss_index.bin')

@app.post("/predict")
async def predict(file: UploadFile = File(...)):
    # Extract features from uploaded image
    # ... (feature extraction code)
    
    # Search similar images
    distances, indices = index.search(query_embedding, k=5)
    
    # Majority voting for disease class
    classes = [metadata[idx]['class'] for idx in indices[0]]
    predicted_class = max(set(classes), key=classes.count)
    
    return {"disease": predicted_class, "confidence": distances[0][0]}

🎯 Use Cases

1. Plant Disease Detection: Identify diseases from leaf images
2. Agricultural Monitoring: Track crop health
3. Research: Study plant disease patterns
4. Education: Learn about plant diseases
5. Mobile Apps: Build plant disease detection apps

🔬 Technical Details

Embedding Extraction

• Model: microsoft/resnet-50
• Input Size: 224x224 pixels
• Normalization: L2 normalization
• Batch Size: 64 (GPU)
• Processing Time: ~15 minutes on T4 GPU

FAISS Index

• Type: IndexFlatIP (Inner Product)
• Metric: Cosine Similarity
• Dimension: 2048
• Search Time: <10ms per query

📈 Performance

• Search Speed: Sub-10ms per query
• Accuracy: High (based on similarity)
• Scalability: Can handle millions of images
• Memory: ~1.8 GB total

🛠️ Building the Database

The database was built using:

# Extract features
from transformers import AutoImageProcessor, AutoModel
processor = AutoImageProcessor.from_pretrained("microsoft/resnet-50")
model = AutoModel.from_pretrained("microsoft/resnet-50")

# Process images
embeddings = []
for image_path in image_paths:
    image = Image.open(image_path)
    inputs = processor(images=image, return_tensors="pt")
    outputs = model(**inputs)
    embedding = outputs.pooler_output.numpy()
    embeddings.append(embedding)

# Build FAISS index
embeddings = np.vstack(embeddings)
faiss.normalize_L2(embeddings)
index = faiss.IndexFlatIP(2048)
index.add(embeddings)

📦 Integration

Download in Your App

from huggingface_hub import hf_hub_download

files = ['embeddings.npy', 'metadata.json', 'faiss_index.bin']

for filename in files:
    hf_hub_download(
        repo_id="ugahost991/plant-disease-detector",
        filename=filename,
        repo_type="dataset",
        local_dir="vector_store"
    )

📄 License

MIT License - Free to use for commercial and non-commercial purposes

🙏 Acknowledgments

• Dataset: PlantVillage Dataset
• Model: Microsoft ResNet-50
• Framework: FAISS (Facebook AI)
• Platform: Hugging Face Hub

📞 Contact

Gaston Software Solutions LLP

• 🌐 Website: www.gss-tec.com
• 📧 Email: info@gss-tec.com
• 💼 LinkedIn: Gaston Software Solutions LLP
• 🐛 Issues: Open an issue on the dataset repository

For business inquiries, partnerships, or custom solutions, please contact us via email or visit our website.

🔗 Related Resources

• Space: Plant Disease Detector API
• Model: microsoft/resnet-50
• Original Dataset: PlantVillage on Kaggle

📊 Statistics

{
  "total_images": 54304,
  "unique_classes": 38,
  "embedding_dimension": 2048,
  "index_type": "FAISS IndexFlatIP",
  "file_size_mb": 888,
  "build_time_minutes": 15,
  "gpu_used": "T4"
}

🎓 Citation

If you use this dataset, please cite:

@dataset{plant_disease_embeddings_2024,
  title={Plant Disease Detection Vector Database},
  author={ugahost991},
  year={2024},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/ugahost991/plant-disease-detector}}
}

---

Built with ❤️ using ResNet-50, FAISS, and Hugging Face 🌱🚀