Advancing Fire and Smoke Detection Reliability: Integrating Generalized ELAN and Programmable Gradient Information within YOLOv9 to Discriminate Against False Positives

Abstract This paper presents YOLOv9, the advancement in the YOLO (You Only Look Once) series, innovating with the GELAN (Generalized Efficient Layer Aggregation Network) and PGI (Programmable Gradient Information) to enhance the precision of detection systems. We detail our empirical evaluation of YOLOv9 on robust and diverse datasets, highlighting its capability to maintain high detection accuracy while significantly reducing false positives, even with limited training data and in challenging scenarios. In this case, the approach has the potential to improve the accuracy of fire and smoke detection without being trained on pseudo-fire and pseudo-smoke scenes and without infrared lighting conditions. The findings highlight its potential as a reliable tool in this safety-critical application. Moreover, the proposed models demonstrate YOLOv9's superiority over its predecessors, with good performance, achieving high precision (P, mAP50, and mAP50-95) and a recall doing the inference in object early detection technologies.