Fire Event Detection Dataset: Acoustic Measurements

FIRE EVENT DETECTION DATASET: ACOUSTIC MEASUREMENTS This dataset contains acoustic recordings of fire events and non-fire events, collected as a proof-of-concept for smart fire detection using machine learning. The data was collected using a modified ISO 5660 cone calorimeter setup, recording acoustic emissions from various materials during heating, pyrolysis, and burning phases. This dataset is the official companion to the paper:"A Novel Method for Smart Fire Detection Using Acoustic Measurements and Machine Learning: Proof of Concept"John Martinsson, Marcus Runefors, Håkan Frantzich, Dag Glebe, Margaret McNamee, Olof Mogren.Fire Technology (2022). DOI: 10.1007/s10694-022-01307-1 DATASET CONTENTS The repository contains three primary data files: data.zip (Raw Acoustic Source Files) Contains the original .wav recordings of the experiments. Recording Specifications: 96 kHz / 24 bit. Materials included: Softwood (Spruce), Hardwood (Oak), Chipboard, Polymethylmethacrylate (PMMA), Polyurethane (PUR), and a PUR/fabric combination. Includes both "fire events" (materials exposed to radiation) and "non-fire events" (cone calorimeter background noise, ventilation fan noise). dataset_spruce_oak_pmma_pur_chipboard_sr_32000.hdf5 (Prepared Dataset) The pre-processed dataset ready for machine learning pipelines. The original .wav files have been down-sampled to 32,000 Hz and split into 5-second segments without overlap. experiments.zip (Model Weights and Splits) Contains the exact train/validation/test data splits used in the paper. Contains the pre-trained weights for the CNN14 baseline model. CODE AND USAGE The Python source code to reproduce the data splits, train the CNN14 model, and generate the figures from the paper is available on GitHub:https://github.com/johnmartinsson/fire-event-detection-dataset CITATION If you use this dataset, please cite the corresponding paper:Martinsson, J., Runefors, M., Frantzich, H., Glebe, D., McNamee, M., & Mogren, O. (2022). A Novel Method for Smart Fire Detection Using Acoustic Measurements and Machine Learning: Proof of Concept. Fire Technology, 58, 3385-3403. https://doi.org/10.1007/s10694-022-01307-1