Abstract: The detection of wildfire in transmission line corridors under multiple background interference suffers from high false and missed detection rates, and the difficulty of detecting wildfire and smoke in small targets, etc. The YOLOv5s model has high real-time performance, but the accuracy of detecting wildfire in the actual power grid requires to be improved. Consequently, a flame and smoke detection method based on YOLOv5s-EBWS is proposed in this paper, after analyzing the multiple disturbances of wildfire pictures in transmission line corridors. Firstly, a dataset is constructed by collecting many live images of wildfire incidents along actual transmission lines. Secondly, the efficient multi-scale attention (EMA) mechanism and weighted bi-directional feature pyramid network (BiFPN) are introduced into the YOLOv5s model to efficiently extract and fully integrate multi-scale spatial information and important features of flame and smoke. Then, wise intersection over union (WIoU) loss function and soft non-maximum suppression (Soft NMS) method are introduced to avoid the missed detection of overlapping fire and smoke, ultimately to enhance the generalization of the model. Finally, example verification is carried out. The results show that the dataset in this paper improves the evaluation indexes of the model by 12%~15% compared with the public fire dataset. Compared with YOLOv5, the size of YOLOv5s-EBWS model is reduced by 0.4 MB, and the values of e_mAP@0.5, e_mAP@0.5:0.95 and F₁ for fire and smoke are increased by 2.3%, 3.8%, 4.9%, and 5.5%, respectively. The proposed method provides a new way to detect wildfire occurred in transmission line corridors.