This project focuses on classifying images into two categories: Smoke and Fire, using deep learning and transfer learning techniques. We use EfficientNetB0 as a pre-trained model, perform fine-tuning, and visualize performance through evaluation metrics and graphs.
- Problem: Binary classification of images into "Smoke" or "Fire"
- Dataset: Custom dataset organized into train/test folders
- Preprocessing:
- Image resizing to 224x224
- Normalization
- Label encoding
 
- Visualization:
- Class distribution
- Sample images
 
- Transfer Learning with EfficientNetB0
- Feature Extraction and Fine Tuning
- Partial Layer Freezing
- Data Augmentation (optional)
- Model Evaluation: Accuracy, Precision, Recall, F1-Score, Confusion Matrix
- Deployment Phase: Prediction on unseen images
- TensorFlow / Keras
- NumPy
- Matplotlib
- Seaborn
- scikit-learn
- OpenCV
- 
Train a deep learning model using EfficientNetB0. 
- 
Fine-tune pre-trained models on a custom Smoke/Fire dataset. 
- 
Evaluate performance using: 
- 
Accuracy 
- 
Precision 
- 
Recall 
- 
Confusion Matrix 
- 
Classification Report 
Metric Value
Test Accuracy ~76.0%
Precision (Fire) 1.0000
Recall (Fire) 0.5200
Precision (Smoke) 0.6757
Recall (Smoke) 1.0000
- 
Training and Validation Accuracy/Loss Curves 
- 
Confusion Matrix Visualization 
- 
Classification Report 
- 
Predictions on Unseen Images