Forest-Fire-Detection

The Forest Fire Detection system is designed to identify and localize fire hotspots in images using a combination of a classifier and an object detection model. This system integrates data collection, model training, and deployment strategies to ensure effective wildfire management.

Table of Contents

1. Overview
2. Data Collection
3. Model Selection and Training
4. Model Features

Overview

The Forest Fire Detector consists of two main components:

• Classifier: A RESNET-50 model that filters images to identify those likely containing smoke or fire.
• Object Detection Model: A YOLOv5 or YOLOv8 model that locates fire and smoke within the filtered images.

Data Collection

Types of Data Required

• Aerial Images: Captured by drones or satellites showing forested areas with and without fire.
• Ground-Level Images: Photos or videos taken from ground surveillance cameras in forest areas.

Data Sources

Analaysied, filtered and merged data from the following sources.

• Kaggle: The Wildfire Dataset
• Roboflow:
  • Forest Fire Dataset (KZDLm)
  • Fire Detection Dataset (TIYPl)
  • Forest Fire Preprocessing

Data Annotation

• Tools: Use Roboflow for labeling and annotating fire regions in images.
• Annotations: Create bounding boxes around fire hotspots for training object detection models.

Model Selection and Training

Classifier

• Model Type: RESNET-50
• Training Data: Approximately 5,000 images
• Purpose: Filter images to pass only those with potential smoke or fire to the object detection model.

Object Detection Model

• Model Type: YOLOv5 or YOLOv8
• Purpose: Detect and localize fire hotspots within images.

Optional: Segmentation Model

• Model Type: U-Net
• Purpose: Provide detailed segmentation of fire regions at the pixel level.

Libraries Required

• TensorFlow and PyTorch: For building, training, and fine-tuning models.
• OpenCV: For image preprocessing, data augmentation, and integrating vision systems.
• Roboflow: For dataset management and preprocessing.

Training Process

1. Train Classifier:
   • Train the RESNET-50 model to filter out non-fire images.
2. Train Object Detection Model:
   • Train YOLOv5 or YOLOv8 to detect and localize fire hotspots using annotated data.

Model Integration

• Use the classifier to filter images, then pass the filtered images to the object detection model for precise localization. If a segmentation model is used, it provides additional detail about fire regions.

Model Features

• Real-time Detection: Detect fire hotspots in real-time from both aerial and ground-level images.
• Scalability: Handle various resolutions and image sources.
• Multimodal Analysis (Optional): Incorporate thermal data to enhance detection accuracy.