Abstract: Real-time cross-sectional state verification of low-voltage equipment in an energy management system (EMS) is difficult in the maintenance of power equipment. Based on infrared images and deep learning, a real-time cross-sectional state verification method for low-voltage EMS equipment was proposed. The seed region growth method segments the red and green components of the real-time cross-section of the EMS low-voltage equipment in the infrared focal plane image to determine the abnormal area of the EMS low-voltage equipment. The invariant moment of the target shape in the abnormal area is extracted, and the difference in the active power of each branch before and after the disconnection of the EMS low-voltage equipment branch is used as the feature vector of the real-time section state check. A DC power flow model based on a deep neural network is constructed. The feature vector was input into the model, and the real-time section-state power flow calculation results were the output. The real-time section state power flow in abnormal areas was analyzed to determine if it was outside the limit, and a real-time section state check was completed. Experiments show that this method can determine the abnormal area of EMS low-voltage equipment. In addition, this method can accurately check the power flow of the real-time section state of the equipment to ensure the safety of equipment maintenance. Applying this method can improve the line qualification rate and voltage qualification rate.