This research introduces AH-YOLO, a lightweight and highly accurate deep learning model optimized with infrared thermal imaging for real-time fire detection in aircraft hangars, outperforming existing methods in both speed and precision.
When it comes to aircraft hangars, even the smallest fire can lead to massive damage and potential disasters. Traditional smoke detectors or temperature sensors often react too late—or not at all. But thanks to a powerful new AI model called AH-YOLO, we now have a fast, accurate, and lightweight solution for real-time fire detection using infrared vision.
Let’s break down what AH-YOLO is, why it matters, and how it’s changing fire safety forever.
Imagine a giant hangar filled with planes, fuel, and composite materials. One small spark could turn the whole space into an inferno within minutes. Current fire safety systems often use:
While these work decently in homes or offices, they’re just not fast or smart enough for complex places like hangars. Why?
So what’s the fix? Artificial Intelligence + Infrared Imaging.
The researchers behind this study designed AH-YOLO, an improved and lightweight version of the famous YOLOv8 model (You Only Look Once) that detects fires using infrared thermal images. That means it “sees” heat, not just flames or smoke.
It’s not just another fire detector. AH-YOLO is built for speed, accuracy, and efficiency using three smart tools:
To train AH-YOLO properly, the researchers didn’t rely on simulations—they lit actual fires (safely!) in a real aircraft hangar in China.
Materials used for burning:
Using a dual-lens infrared thermal camera, they recorded flame behavior under different conditions:
In total, they created a custom dataset of 1500 infrared images, including both fire and non-fire situations. This made sure the AI model learned what real fire looks like—and what it doesn’t.
Let’s geek out a little. Here’s what powers AH-YOLO under the hood:
A modular neural network that cuts down on heavy computation. It's fast, efficient, and perfect for edge devices (like drones or security cameras). Think of it as the model’s super-slim engine.
This acts like the AI’s attention span. It tells the model where to “look” in the image—focusing on the hottest spots or flame patterns.
Enhances the model's ability to handle scale and location variation. Fires don’t always look the same—some are far, some small. DyHead helps adapt to these changes.
These modules were perfectly placed into YOLOv8’s architecture to maximize performance while keeping it light.
So how did AH-YOLO perform?
| Metric | YOLOv8n (original) | AH-YOLO (new model) |
| mAP@0.5 (accuracy) | 90.2% | 93.8% |
| Parameters | 3.01M | 2.54M |
| FPS (Frames/sec) | 142 | 171 |
It’s:
That’s a serious upgrade!
Compared to other models like YOLOv5n, EfficientNetV2, and MobileNetV3, AH-YOLO outperforms them in both speed and accuracy—making it ideal for real-time applications in hangars.
AH-YOLO isn’t just another research paper—it’s a practical, deployable solution for enhancing fire safety in critical infrastructure like:
By combining infrared imaging with AI, it works even in low light or smoky conditions—places where traditional detectors fail.
And since it’s lightweight, it doesn’t need powerful GPUs to run—making it perfect for edge computing.
While AH-YOLO is already a game-changer, there’s more to come:
Thanks to AH-YOLO, we’re a step closer to safer skies and smarter buildings. It shows how engineering and AI can work hand-in-hand to solve real-world problems—literally saving lives and infrastructure.
Fire Detection - The tech used to sense and spot fire early. Think of it as a smart “sniffer” that knows when something is burning—even before your smoke alarm goes off.
Aircraft Hangar - A large building where airplanes are stored or repaired. It’s like a garage for planes—but way bigger and more fire-prone due to fuel and materials inside.
Infrared Thermal Imaging - A camera technique that sees heat instead of light. Hot objects like flames glow brightly—even in total darkness or thick smoke. - More about this concept in the article "Beating the Heat: How Cool Roof Coatings Can Save Cities from the Urban Heat Island Effect".
YOLO (You Only Look Once) - A fast deep learning model that finds objects in images. It’s like giving your camera instant object recognition superpowers—detecting things in a single glance. - More about this concept in the article "Smart Bees, Smarter Tech | How Deep Learning is Changing Hive Monitoring Forever!".
AH-YOLO - A smarter, faster version of YOLO, made for fire detection using infrared. It’s specially designed to spot fires in tricky places like hangars—accurate, quick, and lightweight.
CNN (Convolutional Neural Network) - A type of AI that learns to understand images. It’s how your phone knows your face—and how AH-YOLO knows fire. - More about this concept in the article "Quantum-Inspired Algorithm Tackles Urban Noise Pollution: A Breakthrough for Smart Cities".
MobileOne - A lightweight building block for neural networks. It helps AI run faster and smoother—great for mobile or real-time use.
CBAM (Convolutional Block Attention Module) - A tool that tells the AI where to focus in the image. It’s like a highlighter for important spots, helping the model “pay attention” to flames.
DyHead (Dynamic Head) - An AI module that adapts to fire size, shape, and distance. Flames come in all forms—DyHead helps the model adjust to each one.
mAP (Mean Average Precision) - A score that shows how good the model is at detecting things. The closer to 1.0 (or 100%), the better. Think of it as the model’s report card. - More about this concept in the article "Smart Drones for Tiny Creatures: How AI is Revolutionizing Insect Monitoring".
FPS (Frames Per Second) - How many images the model can analyze each second. Higher FPS = faster detection = safer buildings. - More about this concept in the article "Revolutionizing Drone Detection: The RTSOD-YOLO Breakthrough".
Source: Deng, L.; Wang, Z.; Liu, Q. AH-YOLO: An Improved YOLOv8-Based Lightweight Model for Fire Detection in Aircraft Hangars. Fire 2025, 8, 199. https://doi.org/10.3390/fire8050199
From: Civil Aviation Flight University of China; Sichuan All-Electric Aviation Aircraft Key Technology Engineering Research Center.
