EEM-YOLO: Real-Time Fire and Smoke Detection in Power Substation Environments

In response to the urgent demand for real-time detection and early warning of fires and smoke in substations, this study proposes an EEM-YOLO fire and smoke detection system based on the improvement of YOLO11s, aiming to reduce the false detection rate and missed detection rate of traditional detection networks under complex background interference conditions. First, the module of EIEStem is used instead of the shallow convolutional structure of the YOLO11s network to realize the refined extraction of flame smoke texture features by combining edge features with spatial information. Then, the lightweight EIEM module is introduced into the C3k2 module of the backbone network, which adopts the multiscale feature fusion mechanism to reduce the computation amount while fusing the edge and spatial information of the image. After experiments, it is shown that compared to the original YOLO11s, the model precision rate has increased by 1.5%, the model recall rate has increased by 0.6%, the mAP50 index is improved by 0. 7%, and the number of model parameters is reduced by 0.2 × 10⁶. The improved network is more powerful in detailed perception and spatial modeling while meeting real-time detection needs, and it is able to capture smoke and flame in substation scenes more accurately, which significantly improves detection results.