Fire detection with fine tuned YOLOv11

• Fire detection with fine tuned YOLOv11
  • Data
    • Preprocessing
  • Model
  • Fine Tuning
    • Run summary:
    • Curves
  • Detections
  • Dependencies
  • Project Setup
  • Limitations
  • Conclusions
  • Acknowledgements

Data

The Fire Dataset is used to fine tune the model. The dataset contains 751 total images with a training-validation-test splitting of 526, 150 and 75 images respectively.

Preprocessing

• Auto-Orient: Applied
• Resize: Stretch to 640x640

Model

The yolo11n version of the model is used to fine-tune on the dataset. The model was trained for 100 epochs with batch size 16.

Note: The .ipy notebook is not uploaded due to privacy issues.

Fine Tuning

YOLO11n summary (fused): 238 layers, 2,582,347 parameters, 0 gradients, 6.3 GFLOPs

Run summary:

Parameter / Metric	Value
lr/pg0	4e-05
lr/pg1	4e-05
lr/pg2	4e-05
metrics/mAP50(B)	0.99359
metrics/mAP50-95(B)	0.96755
metrics/precision(B)	0.99529
metrics/recall(B)	0.97778
model/GFLOPs	6.441
model/parameters	2590035
model/speed_PyTorch(ms)	2.36
train/box_loss	0.20238
train/cls_loss	0.17491
train/dfl_loss	0.80457
val/box_loss	0.23068
val/cls_loss	0.18552
val/dfl_loss	0.77294

Curves

Detections

Dependencies

• python 3.x
• opencv_contrib_python
• opencv_python
• ultralytics

Project Setup

1. Make a virtual environment using the following command:

   python3 -m venv myenv

   Replace myenv with the name you want for your virtual environment. This will create a folder named myenv in your current directory containing the virtual environment files.

2. Activate the virtual environment:

   source myenv/bin/activate

   Remember to replace myenv with the actual name of the environment created in the previous step.

3. Clone the repository:

   git clone https://github.com/bhaskrr/fire-detection-using-yolov11.git

4. Navigate to the root directory of the project:

   cd path/to/the/project

5. Install dependencies:

   pip install -r requirements.txt

Limitations

The model can be further improved with a more diverse dataset and more classes to detect, e.g. smoke.

Conclusions

This project demonstrates how a fine-tuned YOLOv11 model can be used for detecting fire.

Here are a few use cases for this project:

1. Wildfire Monitoring: Early detection of wildfires in forests or remote areas through real-time video feeds or drone footage, helping in quick response and minimizing damage.

2. Smart Security Systems: Integration with surveillance cameras in residential, commercial, or industrial properties to detect fire, triggering alarms or notifications automatically.

3. Industrial Safety: Monitoring areas in factories or warehouses where fire hazards are present, especially around chemical storage or flammable materials.

4. Autonomous Firefighting Drones: Fire detection systems could guide drones to automatically detect and respond to fires in hazardous or hard-to-reach areas.

5. Transport Safety: Real-time monitoring in public transport systems (trains, buses, or even airplanes) to detect fire risks and prevent accidents.

6. Fire Safety in Smart Homes: Integration into smart home systems to provide immediate alerts and notifications to homeowners and emergency services when fire is detected.

Acknowledgements

The images used to test the model are taken from kaggle.