LIGHTWEIGHT DETECTORS FOR EARLY AGRICULTURAL FIRE DETECTION: SSD-MOBILENETV2 VERSUS ENHANCED YOLOV5S
Keywords:
Keywords: agricultural fire detection; UAV; lightweight object detection; SSD; MobileNetV2; YOLOv5.Abstract
Abstract. Early detection of agricultural fires requires detectors that are accurate enough to notice faint smoke and small flames yet light enough to run on a UAV or an edge device. This review compares two recently published lightweight detectors, an SSD with a MobileNetV2 backbone and a YOLOv5s whose C3 block is deepened with DarknetBottleneck modules, using the accuracy, speed and computational cost reported by their authors. The SSD-MobileNetV2 model is the cheapest (5.0 GFLOPs, 45 FPS), while the modified YOLOv5s is faster (74 FPS), supported by a more rigorous data split, at about 17 GFLOPs. Because the datasets and metrics differ and several published figures are internally inconsistent, neither model can be declared superior; we therefore outline a unified evaluation protocol for fair benchmarking.
References
[1] A. Abdusalomov, S. Umirzakova, K. Tashev, N. Egamberdiev, G. Belalova, A. Meliboev, I. Atadjanov, Z. Temirov, and Y. I. Cho, “AI-Driven UAV Surveillance for Agricultural Fire Safety,” Fire, vol. 8, no. 4, p. 142, 2025. https://doi.org/10.3390/fire8040142
[2] S. N. Saydirasulovich, S. Umirzakova, A. N. Azamatovich, S. Mukhamadiev, Z. Temirov, A. Abdusalomov, and Y. I. Cho, “Lightweight YOLOv5s Model for Early Detection of Agricultural Fires,” Fire, vol. 8, no. 5, p. 187, 2025. https://doi.org/10.3390/fire8050187
[3] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, and A. C. Berg, “SSD: Single Shot MultiBox Detector,” in Proc. ECCV, 2016, pp. 21-37.
[4] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 39, no. 6, pp. 1137-1149, 2017.
[5] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You Only Look Once: Unified, Real-Time Object Detection,” in Proc. IEEE CVPR, 2016, pp. 779-788.
[6] G. Jocher et al., “ultralytics/yolov5: v7.0,” Zenodo, 2022. https://doi.org/10.5281/zenodo.7347926
[7] C.-Y. Wang, H.-Y. M. Liao, Y.-H. Wu, P.-Y. Chen, J.-W. Hsieh, and I.-H. Yeh, “CSPNet: A New Backbone that can Enhance Learning Capability of CNN,” in Proc. IEEE/CVF CVPR Workshops, 2020, pp. 390-391.
[8] S. Liu, L. Qi, H. Qin, J. Shi, and J. Jia, “Path Aggregation Network for Instance Segmentation,” in Proc. IEEE CVPR, 2018, pp. 8759-8768.
[9] Z. Zheng, P. Wang, W. Liu, J. Li, R. Ye, and D. Ren, “Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression,” in Proc. AAAI, vol. 34, no. 7, 2020, pp. 12993-13000.
[10] A. G. Howard et al., “MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications,” arXiv:1704.04861, 2017.
[11] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “MobileNetV2: Inverted Residuals and Linear Bottlenecks,” in Proc. IEEE CVPR, 2018, pp. 4510-4520.
[12] 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.
[13] S. Elfwing, E. Uchibe, and K. Doya, “Sigmoid-weighted linear units for neural network function approximation in reinforcement learning,” Neural Networks, vol. 107, pp. 3-11, 2018.
