Abstract: Aiming at the problem of irregular features,insufficient texture information,and low recognition accuracy in detecting small targets within infrared images of electrical equipment metal fittings in substations,an improved method based on the YOLOv8 model is proposed in this paper,which creates a Dynamic Contextual Information Extraction Module (DCIEM) suitable for small target detection in infrared images by integrating Multi-Scale Dilated Attention (MSDA) and Deformable Convolution (DC),so as to effectively obtain the global features and local details of the target and adapt to the morphological changes of small infrared targets.Then with embedded Coordinate Attention (CA) mechanism,the ability of detail information extraction and small target detection is enhanced to improve the accuracy of target regression.Meanwhile,in order to solve the problem of loss of detailed features in the process of network convolution,by introducing weighted bidirectional feature pyramid network (BiFPN) mode,a weighted bidirectional path aggregation network (Bi-PANet) is proposed to optimize multi-scale feature fusion and prevent the loss of original features.The experimental results show that the improved YOLOv8 model proposed in this paper,improves the mean detection accuracy (mAP) with increase of 5.2%,and the recall rate (Recall) with increase of 6.7%,and precision rate with increase of 4.1%.