In the realm of automatic inspection within the power system industry, leveraging artificial intelligence algorithms for the automated detection of typical objects holds immense significance. This paper introduces a novel approach, an asymmetric convolutional neural network, designed for intelligent recognition of typical objects in the power system. Specifically, the proposed algorithm adopts the YOLOv7 network architecture as a baseline for object detection, incorporating a newly developed asymmetric convolution layer. This innovative layer aims to enhance the extraction of image features and improve robustness against target flipping. The effectiveness of the proposed method was validated through its application to a project at the Information & Telecommunication Company, State Grid Ningxia Electric Power Co., Ltd. The objective was to detect the presence of the person in charge at the operation scene. Experimental results demonstrated a notable improvement, with the detection mAP of the person in charge reaching 95.8%, a 6.9% increase compared to YOLOv7, effectively enhancing accuracy. Additionally, the model achieved a recall rate of 93.4%, making it applicable for power system operation detection tasks and elevating the industry’s level of automation.