A Review on Smoke Detection Using Dual Convolutional Networks from Infrared Frames

Computer Science and Engineering

This review looks at the latest achievements in smoke detection for infrared based on Dual Convolutional Neural Networks (Dual-CNNs). Traditional photoelectric and ionization sensors are slow or unreliable at night, in confined spaces or under heavy dust loads. Infrared (IR) imaging provides a means of illumination which is independent from lighting, and deep learning increases recognition accuracy. The architecture under review, proposed by Deng et al. (2024), uses two parallel CNNs: one extracts spatial grainy aspects from IR frames for the other extracts temporal information motion cues extracted from video sequences. The fused feature set provides a robust early-warning capability with fewer false alarms. This paper provides a summary of related work (2015-2025), describes the dual-CNN approach, summarizes comparison metrics and provides discussion around IoT-based deployment. In terms of research results, we demonstrate precision ≈ 94%, recall ≈ 92%, and greater adaptability than either YOLOv5 or ResNet. Finally, we summarize current and future limitations presented by dataset extension, model compression, and multimodal fusion.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.