Answers are grounded in this paper and cite their sources.
Fire safety can be significantly improved using deep learning (DL) models, which reduce false alarms and enable reliable real-time detection. However, existing DL-based object detectors often struggle with achieving consistent accuracy and efficient multi-class detection of indoor fire and smoke due to complex environmental dynamics. This study introduces the FireNet-D5 model for effective indoor fire and smoke detection to address these challenges. FireNet-D5 incorporates a Squeeze-and-Excitation (SE) attention module in the backbone to strengthen feature representation and a Bi-directional Feature Pyramid Network (BiFPN) in the neck for multi-scale feature fusion. The model is further optimized using a genetic algorithm to determine optimal hyperparameters, while Grad-CAM is applied to ensure transparency of model predictions in safety-critical settings. To evaluate performance, a new indoor fire and smoke dataset (ISFire-4) is created, containing four classes: Blue Fire, Black Smoke, Yellow-Orange Fire, and White-Gray Smoke. FireNet-D5 and six benchmark object detection models are re-implemented and trained on ISFire-4. Statistical analysis using Cohen’s d validates practical significance, showing FireNet-D5 improves mAP@0.5 by 3.0%, reduces parameters by 3.3%, and provides a 32% comparative difference in effect size relative to its baseline configuration. Five-seed evaluation confirms stable performance, and cross-dataset testing shows improved generalization with respect to the existing methods. These results highlight that the proposed model enables intelligent fire safety management while improving detection reliability, emergency response, and decision-making.
No takes yet. Share an insight, caveat, or question.
Sozol et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: