Pest Detection Analysis - Setup Guide

📁 Dataset Structure

archive (2)/
├── train/
│   ├── ants/
│   ├── bees/
│   ├── beetle/
│   ├── catterpillar/
│   ├── earthworms/
│   ├── earwig/
│   ├── grasshopper/
│   ├── moth/
│   ├── slug/
│   ├── snail/
│   ├── wasp/
│   └── weevil/
└── test/
    ├── ants/
    ├── bees/
    ├── beetle/
    └── ... (same structure)

Total Classes: 12 pest categories

🚀 How to Run

Step 1: Install Required Libraries

Open a terminal in VS Code and run:

pip install tensorflow numpy pandas matplotlib seaborn pillow opencv-python scikit-learn

Step 2: Run the Notebook

1. Open pest_detection_analysis.ipynb
2. Run cells sequentially (Shift + Enter)
3. The notebook will automatically:
   • Load your pest images from the train/test folders
   • Count images in each class
   • Display sample images
   • Train 3 different models (LeNet-5, VGG16, MobileNetV2)
   • Compare their performance

📊 What's Updated

✅ Changes Made:

1. Dataset Path: Updated to your actual path C:\Users\Soumil\Downloads\archive (2)
2. Number of Classes: Automatically detects 12 pest classes
3. Data Loading: Uses ImageDataGenerator.flow_from_directory() to load images
4. Image Visualization: Added cell to display sample pest images
5. Training: All models now train on your real pest dataset
6. Validation: Uses your test folder for validation

🎯 Key Features:

• Automatic class detection from folder structure
• Real-time image counting for train/test sets
• Data augmentation for better model performance
• Proper train/test split using your folder structure

📈 Expected Output

When you run the notebook, you'll see:

Cell Output Example:

✓ Training directory found: C:\Users\Soumil\Downloads\archive (2)\train
✓ Number of pest classes: 12
✓ Pest categories: ants, bees, beetle, catterpillar, earthworms, earwig, grasshopper, moth, slug, snail, wasp, weevil

📊 TRAINING SET:
   Total images: XXXX
   - ants: XX images
   - bees: XX images
   - beetle: XX images
   ... (and so on)

📊 TESTING SET:
   Total images: XXXX
   - ants: XX images
   - bees: XX images
   ... (and so on)

🎨 Visualizations Included

1. Sample pest images from training set
2. Training/validation accuracy curves for all 3 models
3. Energy consumption comparison
4. Model complexity vs accuracy trade-off
5. Comprehensive comparison dashboard

⚙️ Model Configurations

Modified LeNet-5

• Image size: 128x128
• Most energy-efficient (123.2J per cycle)
• Best for edge deployment

VGG16

• Image size: 128x128
• Highest accuracy
• More computational cost (181.2J per cycle)

MobileNetV2

• Image size: 128x128
• Mobile-optimized
• Good balance (200.1J per cycle)

🔧 Configuration Parameters

You can adjust these in the notebook:

IMG_SIZE = 128        # Image resolution
BATCH_SIZE = 32       # Batch size for training
EPOCHS = 50           # Number of training epochs

📝 Notes

• The notebook uses data augmentation (rotation, flipping, zoom) to improve model robustness
• Training time depends on:
  • Number of images in your dataset
  • GPU availability (much faster with GPU)
  • Number of epochs
• Models are saved in saved_models/ folder after training

🎯 Next Steps

1. Run the notebook cells sequentially
2. Wait for models to train (this may take 20-60 minutes depending on hardware)
3. Review the comparison results
4. Use the saved models for deployment

⚠️ Troubleshooting

If you get "directory not found" error:

• Check that the path C:\Users\Soumil\Downloads\archive (2) exists
• Make sure train and test folders are inside it
• Verify folder names match exactly (case-sensitive)

If training is too slow:

• Reduce EPOCHS to 20-30
• Reduce IMG_SIZE to 64 or 96
• Reduce BATCH_SIZE to 16

If you run out of memory:

• Reduce BATCH_SIZE to 16 or 8
• Close other applications
• Restart the kernel

---

🎉 You're all set! Start running the cells and let the models train on your real pest dataset!