Fire Detection Approach for 10kV Distribution Room Based on Spatial Perception and Hybrid Attention Mechanism

Fire is one of the most significant perils in the distribution room. The approach of manually discovering fire through video surveillance fails to respond promptly to fire anomalies in the distribution room. From the perspective of safety inspection of 10kV distribution room and targeting fire anomaly detection, a 10kV distribution room flame detection method based on spatial perception and hybrid attention mechanism is proposed in conjunction with deep learning algorithms. Firstly, a backbone network based on the hybrid attention mechanism is put forward, integrating the hybrid attention mechanism and BiFormer module to simultaneously take into account both the local feature information and global feature information of the flame. Then, a spatial perception feature pyramid network is devised, employing the high-frequency perception module and spatial dependency perception module to enrich the context information of the features of minute targets. Finally, an attention transparent detection head is established to guarantee that higher values contribute more significantly to the overall training and confer importance on the maximum value. The experimental outcomes demonstrate that the proposed method achieves an accuracy rate of 95.6% and a recall rate of 92.5% in detecting 10kV distribution room flames, indicating excellent performance in this regard.