Nature inspired firefighter assistant by Unmanned Aerial Vehicle (UAV) data

Link to the paper:

• https://m.growingscience.com/beta/jfs/5885-nature-inspired-firefighter-assistant-by-unmanned-aerial-vehicle-uav-data.html
• DOI: http://dx.doi.org/10.5267/j.jfs.2023.1.004

Please cite:

• Mousavi, S., and A. Ilanloo. "Nature inspired firefighter assistant by unmanned aerial vehicle (UAV) data." Journal of Future Sustainability 3.3 (2023): 143-166.

Overview

Wildfires are a significant threat to forests, wildlife, and human safety. This research leverages Unmanned Aerial Vehicles (UAVs) equipped with color, thermal, and infrared cameras, along with nature-inspired algorithms, to detect and analyze wildfires efficiently.

Highlights:

• Fire Detection: Uses image segmentation and classification on color and thermal datasets.
• Smoke Analysis: Proposes a workflow for smoke detection using multi-color space techniques.
• Nature-Inspired Algorithms: Implements Chicken Swarm Algorithm (CSA), Bees Algorithm (BA), and Biogeography-Based Optimization (BBO) for enhanced accuracy.

---

Features

1. Fire Segmentation

• Techniques Used:
  • CSA Intensity Adjustment: Enhances image contrast.
  • DnCNN Denoising: Removes unwanted noise for clearer segmentation.
  • Bees Algorithm: Performs robust and fast image segmentation.
• Performance Metrics:
  • FLAME Dataset: Precision: 95.57%
  • DeepFire Dataset: Precision: 91.74%

2. Fire Classification

• Process:
  • Uses Local Phase Quantization (LPQ) for frequency-based feature extraction.
  • Selects features via Biogeography-Based Optimization (BBO).
  • Classifies fire/no-fire images using Artificial Neural Networks (ANN) optimized with Firefly Algorithm (FA).
• Performance Metrics:
  • FLAME Dataset: Accuracy: 91.33%
  • DeepFire Dataset: Accuracy: 96.88%

3. Smoke Detection

• Separates and enhances RGB channels using Histogram Equalization and Median Filtering.
• Converts images into HSV, Lab, and YCbCr color spaces for multi-channel analysis.
• Identifies smoke using morphology operations on combined channels.

---

Methodology

Datasets

• FLAME (2021): Includes color and thermal images of wildfires.
• DeepFire (2022): Latest dataset for UAV-based fire detection.

Algorithms

• Chicken Swarm Algorithm (CSA):
  • Improves image contrast in dense environments.
• Bees Algorithm (BA):
  • Combines global and local searches for precise segmentation.
• DnCNN:
  • Denoises images using a deep convolutional neural network.
• Firefly Algorithm (FA):
  • Optimizes ANN for better classification accuracy.
• Biogeography-Based Optimization (BBO):
  • Selects impactful features for classification.

---

Results

Metric	FLAME (Color)	DeepFire (Thermal)
Segmentation Precision	95.57%	91.74%
Classification Accuracy	91.33%	96.88%