Fire and smoke detection results raise questions for smart cities

by AI & Deep LearningFeatured NewsFire DetectionResearch

Detection system combines two AI models

Researchers at Jouf University in Saudi Arabia have developed an artificial intelligence framework for early smoke and fire detection in smart city environments.

A study by Amr Abozeid and Rayan Alanazi describes a hybrid system that combines a Vision Transformer with the YOLOv8 object detection architecture.

The framework is designed to identify early-stage smoke and fire patterns in complex visual scenes and support faster response through smart city monitoring systems.

The study says conventional heat and smoke detectors often raise alerts only after smoke reaches the device.

It adds that existing video-based systems can also produce false alarms when visual conditions include dust, clouds or changes in lighting.

The Vision Transformer analyses global visual patterns in an image so the system can detect subtle smoke characteristics and spatial relationships across a scene.

The YOLOv8 module then performs real-time object detection to localise smoke and fire regions at speed.

Detection results and next steps

According to the study in Scientific Reports, the researchers trained and tested the model on more than 7,000 images from urban and rural datasets.

These images included smoke and fire scenes captured under different lighting conditions and in different environments.

The system achieved 99.2% accuracy, with precision of 98.5% and recall of 97.8%.

Its F1 score reached 98.1% and inference latency was about 45 milliseconds, which the study says enabled real-time detection at about 22 frames per second.

The researchers reported an accuracy increase of about 4.3% compared with conventional convolutional neural network approaches and other single-model detection systems.

The study also says the framework performed better than several existing detection models across precision, recall and localisation accuracy.

It states that the hybrid architecture can distinguish real smoke or fire from visually similar features such as printed flames or static images by analysing contextual and spatial characteristics rather than relying only on colour.

The authors say further validation in real-world settings is still needed, with future work set to examine thermal imaging and environmental sensor inputs for low-visibility conditions.

Firefighting UAV swarm study outlines AI inspection and cyber defence approach

by Aerial FirefightingFeatured NewsFire SuppressionMiddle East - Breaking NewsResearch

UAV swarm design for firefighting missions

A peer reviewed paper describes a six-drone UAV swarm framework for firefighting operations that aims to maintain mission continuity while reducing exposure to cyberattacks on inter-drone communications.

The study is titled “A Cyber-Resilient UAV Swarm Framework for Fire-Fighting with AI-Based In-Flight Defect Inspection”, authored by Ahad Alotaibi and Abdullah Alrasheedi of the Department of Advanced Technology, Canadian College of Kuwait, Al Jahra, Kuwait, and published in the Journal of Computer and Communications, Vol.14 No.1 (January 2026).

The proposed architecture uses five operational drones assigned mission roles such as thermal observation, environmental sensing, close-range visual assessment, payload support and communications extension.

It also adds one Inspector/Commander drone positioned to supervise, capture inspection imagery and act as a coordination node linking the swarm to the Ground Control Station (GCS).

The paper frames the approach around two risk areas in swarm missions, physical degradation in fire-ground environments and cyber threats exploiting wireless coordination traffic.

AI inspection and cyber-resilient communications approach

The framework includes a mobile application called Drone Inspector that manages pre-processing, cloud submission and alerting for in-flight defect inspection imagery.

The workflow described has the Inspector/Commander UAV capturing high-resolution images of neighbouring operational drones at defined intervals and sending them through the swarm communications layer to the Drone Inspector application.

The application then submits images to Amazon Rekognition Custom Labels via API and receives defect labels with confidence scores.

The paper describes defect categories including exposed wiring, landing gear damage, landing gear misalignment and deformation of landing components.

Inspection frequency is described as adaptive, with inspections every two minutes under nominal conditions and every 30 seconds in higher risk areas linked to gas sensor readings indicating proximity to an active fire zone.

In the implementation described, the application triggers a critical alert to the Ground Control Station when confidence exceeds a defined threshold, with the paper describing a threshold of 80%.

For communications security, the paper proposes subnet segmentation and Route Optimization for Autonomous Systems (ROAS) to reduce the feasibility of Man-in-the-Middle (MITM) and traffic manipulation attacks.

Subnet segmentation is described as dividing the swarm network into role-based subnetworks with routing policies controlling inter-subnet communication.

ROAS is described as a dynamic routing approach intended to adjust paths based on network conditions and topology changes to reduce persistent interception points.

Evaluation approach and reported results

The paper describes evaluation across physical deployment feasibility, AI inspection workflow performance and cybersecurity simulation.

It describes a six-UAV deployment consistent with the proposed architecture, with the Inspector/Commander UAV maintaining a supervisory position to capture imagery and support communication.

Example UAV platforms named include DJI Matrice series platforms and an Autel EVO Max model, with additional roles described for payload delivery and communications relay.

Live fire was not used for safety reasons.

The paper describes using manoeuvres, formation flight and environmental stressors such as wind variability to emulate operational challenges relevant to emergency response.

For AI defect detection, it describes Amazon Rekognition Custom Labels trained on a labelled dataset including normal conditions and representative defect scenarios, with reported classification metrics including precision, recall and F1-score.

For cybersecurity simulation, it describes a network emulation environment built in GNS3, using a Kali Linux attacker node and Ettercap for adversarial traffic injection, with a comparison between a baseline flat network and a secured configuration applying segmentation and ROAS.

The paper references EtherApe traffic visualisation and describes figures intended to show traffic concentration through the attacker node under baseline conditions, followed by more balanced traffic patterns after defences are applied.

Across these experiments, the authors report reduced attack success rates, early detection of defects and improved operational reliability within the combined inspection and network defence framework.

The paper presents the system as a hierarchical swarm design that links physical integrity monitoring and cybersecurity measures within a single operational framework.