Dataset · Zenodo · May 12, 2026

Dataset for article Enhanced Fine-Tuning Optimization for Real-Time Fire and Smoke Detection Using YOLOv9 Architecture with Comprehensive Ablation Analysis (dataset annotated)"

Bekbol Sultankulov, Marat Nurtas, Daniker Chepashev, Yerbol Sultankulov, Nurgul Abenova, Aizhan Altaibek

Abstract

This dataset and model release accompanies the paper "Enhanced Fine-Tuning Optimization for Real-Time Fire and Smoke Detection Using YOLOv9 Architecture with Comprehensive Ablation Analysis" published in Engineering, Technology & Applied Science Research. The release contains the fine-tuned YOLOv9c model weights, annotation files in COCO format, and the configuration files used for training and evaluation. The training data combines 20,000 publicly available fire and smoke images sourced from the Roboflow platform with 15,000 images acquired from two operational manufacturing facilities in Kazakhstan (industrial portion not redistributed due to non-disclosure agreements; the public portion of annotations and the trained model are provided). The fine-tuned YOLOv9c model achieves 78.0% precision, 76.0% recall, 77.0% F1-score, and 68.7% IoU at 30.2 FPS on the test set, supporting real-time fire and smoke detection in surveillance applications. Intended use: research and reproducibility of the fine-tuning methodology, ablation analysis, and baseline comparisons reported in the paper.

Fire Detection and Safety SystemsAdvanced Neural Network ApplicationsIndustrial Vision Systems and Defect Detectionfire detectionsmoke detectionYOLOv9

Dataset for article Enhanced Fine-Tuning Optimization for Real-Time Fire and Smoke Detection Using YOLOv9 Architecture with Comprehensive Ablation Analysis (dataset annotated)"

Abstract

More by Bekbol Sultankulov