Comparative Classification of Fire Smoke and Neutral Images Using Frozen Features
Keywords:
Transfer learning, Image classification, Feature extraction, Fire recognition, Smoke recognitionAbstract
Image classification can support visual fire monitoring, but overall accuracy alone does not describe differences in recognizing fire, smoke, and neutral scenes. This study compares MobileNetV2 and EfficientNetB0 using frozen ImageNet feature extractors on a balanced subset of the publicly accessible DeepQuestAI dataset. The experiment uses 1,500 images, comprising 960 training, 240 validation, and 300 test images. Both models use a 224 × 224 input, training-only augmentation, an identical classifier head, and a maximum of ten epochs. MobileNetV2 achieves 94.00% test accuracy, a macro F1-score of 0.9401, and a macro one-versus-rest area under the receiver operating characteristic curve of 0.9853. EfficientNetB0 achieves 90.67%, 0.9070, and 0.9845, respectively. Although EfficientNetB0 has slightly higher smoke recall, 95% versus 94%, its smoke precision is lower, 85.59% versus 93.07%. The results provide a configuration-specific baseline and demonstrate why class-wise evaluation complements aggregate scores. The conclusions are limited to one training run and the selected test set; deployment performance and robustness under distribution shifts are not established.
Downloads
References
[1] A. Elhanashi, S. Essahraui, P. Dini, and S. Saponara, “Early fire and smoke detection using deep learning: A comprehensive review of models, datasets, and challenges,” Applied Sciences, vol. 15, no. 18, Art. no. 10255, 2025, doi: 10.3390/app151810255.
[2] K. Muhammad, J. Ahmad, I. Mehmood, S. Rho, and S. W. Baik, “Convolutional neural networks based fire detection in surveillance videos,” IEEE Access, vol. 6, pp. 18174–18183, 2018, doi: 10.1109/ACCESS.2018.2812835.
[3] A. J. Dunnings and T. P. Breckon, “Experimentally defined convolutional neural network architecture variants for non-temporal real-time fire detection,” in Proc. 25th IEEE International Conference on Image Processing (ICIP), 2018, pp. 1558–1562, doi: 10.1109/ICIP.2018.8451657.
[4] J. Roh, S.-H. Min, and M.-S. Kong, “Analysis of fire prediction performance of image classification models based on convolutional neural network,” Fire Science and Engineering, vol. 36, no. 6, pp. 70–77, 2022, doi: 10.7731/KIFSE.9e906e7a.
[5] U. Ejaz, M. A. Hamza, and H.-c. Kim, “Channel attention for fire and smoke detection: Impact of augmentation, color spaces, and adversarial attacks,” Sensors, vol. 25, no. 4, Art. no. 1140, 2025, doi: 10.3390/s25041140.
[6] D. Sharobiddinov, H. U. R. Siddiqui, A. A. Saleem, G. Mendez Mezquita, D. L. Ramírez Vargas, and I. de la Torre Díez, “Edge-based autonomous fire and smoke detection using MobileNetV2,” Sensors, vol. 25, no. 20, Art. no. 6419, 2025, doi: 10.3390/s25206419.
[7] S. J. Pan and Q. Yang, “A survey on transfer learning,” IEEE Transactions on Knowledge and Data Engineering, vol. 22, no. 10, pp. 1345–1359, 2010, doi: 10.1109/TKDE.2009.191.
[8] J. Yosinski, J. Clune, Y. Bengio, and H. Lipson, “How transferable are features in deep neural networks?” in Advances in Neural Information Processing Systems, vol. 27, 2014. [Online]. Available: https://proceedings.neurips.cc/paper_files/paper/2014/hash/532a2f85b6977104bc93f8580abbb330-Abstract.html
[9] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “MobileNetV2: Inverted residuals and linear bottlenecks,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 4510–4520, doi: 10.1109/CVPR.2018.00474.
[10] M. Tan and Q. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” in Proc. 36th International Conference on Machine Learning, vol. 97, 2019, pp. 6105–6114. [Online]. Available: https://proceedings.mlr.press/v97/tan19a.html
[11] DeepQuestAI, “Fire-Smoke-Dataset,” GitHub repository. Accessed: Oct. 1, 2026. [Online]. Available: https://github.com/DeepQuestAI/Fire-Smoke-Dataset
[12] O. Russakovsky et al., “ImageNet large scale visual recognition challenge,” International Journal of Computer Vision, vol. 115, no. 3, pp. 211–252, 2015, doi: 10.1007/s11263-015-0816-y.
[13] C. Shorten and T. M. Khoshgoftaar, “A survey on image data augmentation for deep learning,” Journal of Big Data, vol. 6, Art. no. 60, 2019, doi: 10.1186/s40537-019-0197-0.
[14] N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, “Dropout: A simple way to prevent neural networks from overfitting,” Journal of Machine Learning Research, vol. 15, no. 56, pp. 1929–1958, 2014. [Online]. Available: https://www.jmlr.org/papers/v15/srivastava14a.html
[15] D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” in Proc. International Conference on Learning Representations (ICLR), 2015. [Online]. Available: https://arxiv.org/abs/1412.6980
[16] M. Abadi et al., “TensorFlow: A system for large-scale machine learning,” in Proc. 12th USENIX Symposium on Operating Systems Design and Implementation (OSDI), 2016, pp. 265–283. [Online]. Available: https://www.usenix.org/conference/osdi16/technical-sessions/presentation/abadi
[17] F. Pedregosa et al., “Scikit-learn: Machine learning in Python,” Journal of Machine Learning Research, vol. 12, no. 85, pp. 2825–2830, 2011. [Online]. Available: https://www.jmlr.org/papers/v12/pedregosa11a.html
[18] M. Sokolova and G. Lapalme, “A systematic analysis of performance measures for classification tasks,” Information Processing & Management, vol. 45, no. 4, pp. 427–437, 2009, doi: 10.1016/j.ipm.2009.03.002.
[19] T. Fawcett, “An introduction to ROC analysis,” Pattern Recognition Letters, vol. 27, no. 8, pp. 861–874, 2006, doi: 10.1016/j.patrec.2005.10.010.
[20] S. Kapoor and A. Narayanan, “Leakage and the reproducibility crisis in machine-learning-based science,” Patterns, vol. 4, no. 9, Art. no. 100804, 2023, doi: 10.1016/j.patter.2023.100804.
