Welcome to the Wildfire Detection Research project repository! This repository hosts the code and resources related to our research on leveraging computer vision for fire detection. Our aim is to contribute to wildfire prevention efforts by developing and training an object detection model to accurately identify instances of fire and smoke in images.

• Trained YOLOv8 model using the D-Fire dataset for accurate fire and smoke detection.
• Explored different model sizes and their performance metrics.
• Developed a Streamlit app for practical demonstration of the model's capabilities.
• Investigated future implications of using computer vision in wildfire prevention strategies.

We utilized the D-Fire dataset, a curated collection of 21,000 labeled images, each annotated in YOLO format. The dataset focuses on fire and smoke instances, while also encompassing diverse visual cues, including non-fire images that resemble fire-like patterns.

We have explored many different datasets. Here is the summary:

We trained the YOLOv8 model by Ultralytics on the D-Fire dataset to achieve accurate fire and smoke detection. Our research not only focuses on achieving high accuracy but also on optimizing model parameters and hyperparameters to ensure efficiency and speed.

The trained YOLOv8 model demonstrated impressive performance on the D-Fire test dataset, with mAP@50 scores and inference time across different model sizes as follows. Evaluation was done using NVIDIA A100-SXM4-40. Resolution of input images was 640x640.

This research underscores the potential of computer vision in addressing real-world challenges, such as wildfire detection. As technology evolves, integrating machine learning tools into wildfire prevention and emergency response strategies could significantly enhance our ability to detect and mitigate wildfires effectively.

For a practical demonstration of our research, you can interact with our Wildfire Detection App, powered by the YOLOv8 model. This app allows you to upload images and observe the model's detection capabilities in action.

Please note that while our Streamlit app demonstrates the capabilities of our model, it is intended for demonstration purposes and may not be suitable for critical wildfire detection applications.

I would like to acknowledge the Institute of Smart Systems and Artificial Intelligence (ISSAI) at Nazarbayev University for fostering an environment of innovation and research excellence. The support I received from ISSAI has been integral to the successful completion of this endeavor.

I extend my heartfelt appreciation to my supervisor, Askat Kuzdeuov, at ISSAI, whose guidance and mentorship were indispensable to the success of this research. His expertise and support have been invaluable in shaping the direction and quality of this work.

I would also like to extend my thanks to the creators of the D-Fire dataset for providing a valuable resource that underpins the foundation of this research. Additionally, the Ultralytics team's contribution to the YOLOv8 model has been instrumental in enabling accurate and efficient fire detection.

This project is licensed under the MIT License.