Meteorological forecasting services and meteorological energy development require data with longer time series, higher spatial and temporal resolution, especially for hourly data. Meteorological data scanned from recording papers have problems such as stains, fading, blurring, and smearing, which cannot meet the requirements of archiving and servicing, and also makes the numerical extraction of images greatly difficult, and the accuracy of extraction results is not guaranteed. This paper proposes an image optimization algorithm based on K means, which can quickly identify and distinguish the image background and data recording curves, filter noise in images, and unify the color and thickness of data recording curves. After optimization, the contrast and sharpness of the images are obviously increased, and the volume is obviously reduced. In practice, it is found that the optimized images save storage resources and cost, and the recognition rate is obviously improved. The result shows that the optimization method based on K means improves the quality and application effect of meteorological digital files.