<p>This study focuses on key substation instrument recognition technology, which is critical for smart grid construction. To address challenges such as large model sizes, high computational demands, poor real-time performance, and deployment difficulties on resource-constrained edge devices, we propose a lightweight object detector, YOLOv11-PSP, based on an improved YOLOv11n network. Additionally, a dedicated substation instrument dataset is created for training. First, we integrate the C3k2 module in YOLOv11n with the PConv from FasterNet, resulting in the C3k2_Faster_PConv module. Next, we introduce a slicing operation to the SimAM module, creating the SimAM With Slicing Attention (SWS) module. This module is then integrated with the convolution in YOLOv11n to form the Conv_SWS module. Finally, we reduce the number of detection heads from three to two. Experimental results show that, compared to the original model, the improved model achieves a 30.8% reduction in the number of parameters and a 28.8% reduction in model size. At the same time, the mean average precision (mAP) increases to 97.9%. The proposed model significantly improves lightweight design and computational efficiency while maintaining performance, making it suitable for deployment in resource-constrained environments. These improvements provide critical technical support for real-time monitoring and analysis of substation instruments, advancing the development of smart grids.</p>