Smoke and Flame Detection Method in Multiple Scenarios

Nowadays, global fire outbreaks occur frequently, and timely detection before a fire outbreak is imminent to reduce its harmful effects. Addressing the current issues of poor effectiveness, high false alarm rate, slow detection speed, and limited detection environment in smoke and flame detection, this paper proposes an improved approach based on the YOLOv7 algorithm. Employing EIOU as the boundary loss function enhances the model's convergence speed and regression accuracy. Furthermore, in the neck section of the structure, a one-time aggregation method is utilized to design the cross-level sub- network (GSCSP) module VoV-GSCSP, which reduces the com- plexity and computational burden of the network structure while maintaining sufficient accuracy. Finally, the Channel Attention (CA) mechanism is integrated into the backbone to enhance the extraction capability of features related to the edges and dense distribution positions of smoke and flames. The improved algorithm achieves a 2.3% increase in [email protected], reduces the overall network model parameters by 5.17MB, and improves detection speed to meet the real-time requirements of smoke and flame detection in multiple scenarios.