Optimization and Application of Improved Fire Detection Algorithm in Inspection Robot Field

Aiming at the deficiency of the inspection robot in the detection accuracy and speed of flame and smoke in the complex scene, a fire detection algorithm based on improved YOLOv5s is proposed. Based on YOLOv5s, the model adds Coordinate Attention (CA) mechanism to the backbone network to improve the learning ability of the model on features. By adding large-scale Detect layer to improve the multi-scale detection mechanism, and improve the location loss function, the convergence behavior of the training process is optimized. The test results show that: through the collaborative design of “interval C3CA attention mechanism + newly added 160×160 small target detection layer +SIoU loss function,” the mAP of the improved model is improved by 1.62% compared with the original YOLOv5s, which is much higher than the improvement range of adding any component separately. The fire detection performance in complex scenes is improved by superposition, providing more effective technical support for the inspection robot’s early fire warning.