Research on the optimization strategy of robot fire detection algorithm

This study addresses the challenge of balancing recognition accuracy and computational efficiency in fire detection algorithms by proposing an improved network model based on the YOLOv5s architecture. In the core architecture of the network, the conventional C3 module is replaced with a novel C3S module, which integrates the SimAM attention mechanism into the Bottleneck structure to enhance feature representation. In the network’s neck region, the standard conv block is substituted with the lightweight GsConv module to further optimize performance. Additionally, the WIoU loss function is employed, and the Soft-NMS technique is adopted to refine the prior box generation, collectively contributing to improved detection accuracy. Experimental results on a custom fire image dataset demonstrate that the enhanced YOLO-CGWS model achieves a detection precision of 94.5 percent. Relative to the baseline, the proposed model improves precision, recall, and mean average precision by 6.7, 5.6, and 3.7 percentage points, respectively, while reducing model parameters by 6.3 percentage points and significantly lowering both missed and false detection rates. Due to its strong generalization and efficiency, the optimized model shows considerable potential for deployment in surveillance systems, intelligent robotics, and other fire-monitoring applications.