Due to the wide distribution of grid meters, they are located in complex environments. It's difficult to identify them under different angles and lighting conditions. Their data and operational status need to be checked regularly However, there are many problems such as high labor intensity, low productivity and poor reliability. In this paper, we propose a deep-learning target detection algorithm, termed as YOLO-MDM, to identify grid meters. The algorithm is based on the yolov4 algorithm with the introduction of MobileNet. The final result is an efficient and accurate identification of grid meters.