Abstract: Based on L_EIOU loss function in YOLOv4 algorithm, a high-precision meter optical image recognition algorithm based on weight coefficient improvement of YOLOv4 loss function is studied in this paper, aiming at the self-localization and recognition targets of meter optical images under the background of substation intelligent operation and maintenance. The calculation principle of loss function is analyzed theoretically, and weight coefficient is used to improve the overlapping area of loss function, Euclidean distance of central position, horizontal and vertical ratio and confidence degree. The algorithm is applied to a substation in Hengshui, Hebei Province. The results show that the meter can be correctly located and read. After several iterations of weight coefficient optimization, compared with L_DIOU、L_CIOU、L_EIOU algorithms, L_WIOU algorithm shows better loss function convergence effect, and compared with L_EIOU, which has the best effect at present, the accuracy rate, recall rate and average accuracy of the solution are increased by 1.06%, 0.64% and 1.25% respectively, which has certain engineering significance in the field of substation meter monitoring.