🐛 Jute Pest Classification System

An intelligent web application for identifying and classifying 17 different types of jute pests using an optimized TensorFlow Lite deep learning model. This system provides farmers and agricultural experts with an easy-to-use tool for rapid pest identification.

A basic jute pest type classifier based on pest image

🌐 Live Demo

Try it now jute-pest-classifier.streamlit.app

✨ Features

• 🎯 High Accuracy 95.5% test accuracy with optimized TFLite model
• ⚡ Real-time Classification Instant pest identification from uploaded images
• 📊 Confidence Scoring Detailed probability scores for all predictions
• 🏆 Top-3 Predictions Alternative possibilities with confidence levels
• 📱 User-friendly Interface Clean, intuitive web interface
• 🚀 Cloud Deployment Deployed on Streamlit Community Cloud
• 📈 Detailed Analytics Complete probability breakdown for all 17 pest classes

🚀 Quick Start

Prerequisites

• Python 3.8 or higher
• 4GB+ RAM (for model loading)
• Modern web browser

Installation

1. Clone the repository

git clone repository-url
cd Jute-Pest-Classification

2. Install dependencies

pip install -r requirements.txt

3. Run the application

streamlit run app.py

4. Access the app
   • Open your browser and navigate to http://localhost:8501
   • Or use the live demo https://jute-pest-classifier.streamlit.app/

📁 Project Structure

Jute-Pest-Classification
├── 📄 app.py                     # Main Streamlit application
├── 📓 JP.ipynb                   # Jupyter notebook for model training
├── 📋 requirements.txt           # Python dependencies
├── 🤖 jute_pest_model.tflite     # Optimized TFLite model (~42MB)
├── 📖 README.md                  # This documentation
└── 📖 LICENSE                  # LICENSE

🐛 Supported Pest Classifications

The system can accurately identify and classify 17 different jute pest species

Pest List

#	Pest Name	Category	Damage Type
1	Beet Armyworm	Lepidoptera	Leaf feeder
2	Black Hairy Caterpillar	Caterpillar	Defoliator
3	Cutworm	Lepidoptera	Rootstem cutter
4	Field Cricket	Orthoptera	Omnivorous
5	Jute Aphid	Hemiptera	Sap sucker
6	Jute Hairy Caterpillar	Caterpillar	Leaf feeder
7	Jute Red Mite	Acari	Sap sucker
8	Jute Semilooper	Lepidoptera	Defoliator
9	Jute Stem Girdler	Coleoptera	Stem borer
10	Jute Stem Weevil	Coleoptera	Stem borer
11	Leaf Beetle	Coleoptera	Leaf feeder
12	Mealybug	Hemiptera	Sap sucker
13	Pod Borer	Lepidoptera	Podseed feeder
14	Scopula Emissaria	Lepidoptera	Defoliator
15	Termite	Isoptera	Woodcellulose feeder
16	Termite odontotermes (Rambur)	Isoptera	Soilroot feeder
17	Yellow Mite	Acari	Sap sucker

📊 Model Performance & Specifications

Performance Metrics

• 🎯 Test Accuracy 95.5%
• ⚡ Inference Time 1-3 seconds per image
• 🧠 Model Architecture TensorFlow Lite (Optimized)
• 📐 Input Resolution 480×480 pixels
• 💾 Model Size ~42MB
• 🔄 Loading Time Instant loading

Technical Details

• Framework TensorFlow Lite
• Pre-training ImageNet-21k (original BiT model)
• Fine-tuning Custom jute pest dataset
• Optimization TFLite conversion with quantization
• Deployment Streamlit Community Cloud

🎯 How to Use

Step-by-Step Guide

1. 🚀 Launch the Application

   streamlit run app.py

2. ⚡ Instant Loading

   • Model loads instantly from optimized TFLite format
   • No waiting time required

3. 📤 Upload an Image

   • Click Browse files or drag & drop
   • Supported formats JPG, JPEG, PNG, BMP
   • Recommended Clear, well-lit images

4. 🔍 View Results

   • Primary Prediction Most likely pest type
   • Confidence Score Model certainty (0-100%)
   • Top 3 Predictions Alternative possibilities
   • Full Analysis Complete probability breakdown

📸 Image Guidelines

✅ Good Images	❌ Avoid
Clear, focused pest	Blurry or out-of-focus
Good lighting	Too dark / too bright
Pest fills frame	Pest too small
Single pest visible	Multiple pests
Natural colors	Heavy filters

⚡ Performance Optimization

Model Size Optimization

For faster loading times, you can create optimized model versions

1. TensorFlow Lite Conversion (Recommended)

   python convert_to_tflite.py

   • Reduces model size by 50-80%
   • Faster loading and inference
   • Maintains high accuracy

2. Model Compression

   python optimize_model.py

   • Creates optimized SavedModel
   • Better memory efficiency

🔧 Troubleshooting

Common Issues & Solutions

Issue Solution

Model loading fails Ensure my_saved_bit_model contains all files Slow first load Normal behavior - subsequent loads are cached Low accuracy Use clear, well-lit images with visible pests Memory errors Ensure 4GB+ RAM available Import errors Run pip install -r requirements.txt

Model File

The optimized model file

jute_pest_model.tflite    # 42MB TensorFlow Lite model

Model Conversion

The TFLite model was converted from the original BiT model for optimal deployment

1. Original training See JP.ipynb notebook
2. Model conversion Converted to TensorFlow Lite format
3. Optimization Quantized for smaller size and faster inference

API Integration

The core prediction function can be used independently

from app import load_model, preprocess_image, predict_pest

# Load model once
model = load_model()

# Predict on new image
image_array = preprocess_image(your_image)
pred_idx, confidence, probabilities = predict_pest(model, image_array)

📊 Dataset Information

• Total Images Varies by class
• Image Resolution 512×512 pixels
• Format RGB color images
• Split TrainValidationTest
• Augmentation Rotation, flip, zoom, brightness

🔬 Research & Citations

This project was made using dataset provided by University of California, Irvine (UCI) Machine Learning Repository

@misc{jute_pest_920,
  author       = {Islam, Muhammad Tanvirul},
  title        = {{Jute Pest}},
  year         = {2024},
  howpublished = {UCI Machine Learning Repository},
  note         = {{DOI}: https://doi.org/10.24432/C5289P}
}

This project is based on the BiT (Big Transfer) model architecture, optimized for deployment

@article{kolesnikov2020big,
  title={Big Transfer (BiT) General Visual Representation Learning},
  author={Kolesnikov, Alexander and Beyer, Lucas and Zhai, Xiaohua and Puigcerver, Joan and Yung, Jessica and Gelly, Sylvain and Houlsby, Neil},
  journal={ECCV},
  year={2020}
}

🤝 Contributing

We welcome contributions! Here's how to get started

1. 🍴 Fork the repository
2. 🌿 Create a feature branch

   git checkout -b featureamazing-feature

3. 💻 Make your changes
4. ✅ Test thoroughly
5. 📝 Commit your changes

   git commit -m Add amazing feature

6. 🚀 Push to the branch

   git push origin featureamazing-feature

7. 🔄 Open a Pull Request

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• TensorFlow Team for the excellent deep learning framework
• Streamlit Team for the amazing web app framework
• Google Research for the BiT model architecture
• Agricultural Research Community for pest classification datasets

📞 Support & Contact

• 📧 Email vanshoberoi462@gmail.com

---

⭐ Star this repository if it helped you!
Made with ❤️ for the agricultural community