A lightweight enhanced convolution and fused attention mechanisms for fire risk detection in the power grid

Effective fire risk detection is crucial for ensuring power grid safety. However, existing algorithms face challenges in accurately detecting incipient flames and often suffer from high computational complexity, limiting their deployment on edge devices. To address these challenges, we propose a Dual Pooling Fusion Lightweight Attention (DPFLA) module to enhance small-target detection accuracy by effectively capturing fine-grained features. We then adopt a Linear Augmented Depthwise Separable Convolution (LADSC) module to further reduce the computational cost, making it suitable for edge deployment. Experimental evaluations on enriched public datasets demonstrate that the proposed algorithm achieves a detection accuracy of 95.6% for small targets at 70.6 frames per second, outperforming existing solutions and making it highly suitable for real-time deployment on edge devices. This suggests that our proposed approach offers a new path for fire detection in resource-constrained environments, ensuring the safety of the power grid.