Evaluation of Improved MobileNetV3-YOLO Models for IoT Devices by Fuzzy Multi-Criteria Decision-Making
Shiyan Du, Jiacheng Li, Masato Noto · 2024
Fire detection is crucial in natural disaster prevention, but the limited computational resources of Internet of Things (IoT) devices impose strict requirements for lightweight and efficient models. This study aims to evaluate and compare the performance of various object detection models for fire detection to optimize model selection in resource-constrained environments. We selected the YOLOv8 series, YOLOv10 series, and our improved MobileNetV3-YOLOv8n and MobileNetV3-YOLOv10n models for evaluation. The results demonstrate that the MobileNetV3-YOLOv8n model achieves the best overall performance, followed by the YOLOv8n and MobileNetV3-YOLOv10n models. Sensitivity analysis confirms the stability of these results under minor variations in metric weights, particularly within the top three rankings. On the basis of these findings, we provide specific model selection recommendations for IoT devices with varying computational capabilities.