Forest fires are a caused by many factors like dry rainfall conditions, climate change, high temperatures, lightning strikes. Forest Wildfire are major problem that causes harm to people, the environment, and the economy. To reduce the damage, its essential to detect wildfires early. Traditional methods like CNN and SVM have been used, but they have limitations such as low accuracy and requiring large amounts of data. To overcome these challenges, this project proposes using a deep learning approach called Reduced VGGNet for Wildfire Classification, which can automatically extract features from images and improve detection accuracy. The project aims to design a system that can detect wildfires and wildfire regions efficiently and effectively. The Vibe algorithm is used and bounding boxes are generated to detect the Wildfire region. Our proposed solution aims to provide a more efficient, scalable and cost-effective approach to addressing the problem. Our goal is to provide a reliable and efficient solution that can make a meaningful impact in addressing the problem and we’re confident that our proposed framework can achieve this.
Introduction
Wildfires are a global threat, causing serious harm to the environment, economy, and human lives. Early and accurate detection is crucial to prevent their spread. Traditional image-based wildfire detection methods are limited by poor accuracy, false alarms, and slow processing. To address this, the project proposes an automated wildfire detection system using deep learning, specifically a Reduced VGG16 model.
Project Objectives:
Accurately identify wildfire images in real-time.
Minimize false alarms.
Enhance detection efficiency in remote areas.
Improve computational performance for real-time use.
Send SMS alerts upon fire detection.
Key Contributions:
A Reduced VGG16 model is proposed for wildfire image classification, achieving 98.42% accuracy.
A wildfire region detection module using the Vibe algorithm identifies fire areas, calculates fire percentage, and sends SMS alerts.
Literature Review Highlights:
Machine Learning models (e.g., SVM, Decision Trees) are used for wildfire prediction based on environmental data.
Deep learning methods (e.g., VGGNet, ResNet, U-Net) show improved performance for fire image classification and segmentation.
Despite progress, challenges like false positives and computational cost remain.
Implementation Overview:
Dataset: 2350 images from the FLAME VISION dataset.
Preprocessing: Images resized to 224x224x3 and features normalized.
Models used:
SVM: Basic classifier, 65.3% accuracy.
Full VGG16: Deep CNN, 74.4% accuracy.
Reduced VGG16: Optimized model with transfer learning, 98.42% accuracy.
Region Detection: Fire areas are annotated using Vibe, bounding boxes are drawn, and alerts are sent with fire percentage.
Results:
Reduced VGG16 significantly outperformed SVM and Full VGG16 in accuracy, precision, and recall.
The system successfully distinguishes wildfire from non-wildfire images.
SMS alerts and visual detection make it practical for emergency services.
Conclusion
The wildfire detection system developed in this study successfully classifies wildfire and non-wildfire images and detects fire regions using advanced machine learning and computer vision techniques. The classification module, trained using a dataset of 1900 images, demonstrated high accuracy, with the Reduced-VGG16 model achieving 98.42% accuracy, significantly outperforming SVM and
Full-VGG16 models. The wildfire region detection module employed the Vibe algorithm to extract fire masks and detect fire regions, further validated using annotations. The
system efficiently draws bounding boxes, extracts fire coordinates, and integrates Twilio API for real-time alerts.
The comparative analysis of different models highlighted that deep learning-based approaches, particularly the Reduced-VGG16 model, provide significantly better classification accuracy, precision, and recall compared to traditional machine learning models like SVM. The integration of fire percentage calculation allows for fire severity assessment, making the system more informative and useful for real-world applications. Overall, the proposed approach provides an accurate, efficient, and real-time wildfire detection system, capable of sending timely alerts for wildfire management and mitigation.
References
[1] Adithi M. Shrouthy, Syed Matheen Pasha, Navya Shree S., Yamini S. R., and Lisha U., \"Forest fire prediction using ML and AI,\" [Journal], 2021.
[2] PerumalaJyothiSwaroop, \"Deep Learning Approaches for Early Detection of Forest Fires,\" International Journal of Research Publication and Reviews, Vol. 5, No. 11, pp. 7385 7390, November 2024.
[3] Lidong Wang, Huixi Zhang, Yin Zhang, Keyong Hu, and Kang, \"A Deep Learning Based Experiment on Forest Wildfire Detection in Machine Vision Course,\" IEEE Journal, 2023.
[4] SiddhantPatil, VidishWorah, Harsh Humne, DivyanshuAwari, and Dr. Rahila Sheikh, \"Forest Fire Prediction Using Machine Learning,\" International Journal for Research in Applied Science & Engineering Technology (IJRASET), December 2023
[5] Virupaksha Gouda R, Anoop R, Joshi Sameerna, ArifBasha, and Sahana Gali, \"Forest Fire Prediction Using Machine Learning,\" International Journal for Research in Applied Science & Engineering Technology (IJRASET), May 2023.
[6] MimounYandouzi, Mohammed Boukabous, Kamal Ghoumid, MounirGrari, Omar Moussaoui, IdrissIdrissi, MostafaAzizi, and AissaKerkourElmiad, \"Forest Fires Detection using Deep Transfer Learning,\" International Journal of Advanced Computer Science and Applications (IJACSA), Vol. 13, No. 8, 2022.
[7] Rakesh Reddy P. and Dr. K. Badari Nath, \"Forest Fire Detection Using Convolution Neural Networks,\" IJCSPUB, Vol. 12, Issue 3, July 2022, ISSN: 2250-1770.
[8] DevendharGogula, Pranav Kumar, and Rakesh Sharma, \"Deep Learning Framework for Forest Fire Detection Using Optical Images,\" Journal of Autonomous Intelligence, Vol. 7, Issue 5, 2024, doi: 10.32629/jai.v7i5.1522.
[9] Muhammad Safeer Khan, RajvardhanPatil, and Syed Ali Haider, \"Application of Convolutional Neural Networks for Wildfire Detection,\" IEEE SoutheastCon, 2020.
[10] GeorgieVadakkadathuRajan and Sinumol Paul, \"Forest Fire Detection Using Machine Learning,\" [Conference], 2022