This data collection encompasses 1,387 classified LVIS georeferenced images, which include four classes: Ice, Melt Pond, Open Water, and Shadow. The original LVIS images were acquired in July 2022 using a PhaseOne medium-format camera during the ICESat-2 2022 Arctic Summer calibration campaign, with spatial resolution ranging between 0.39 m and 0.5 m. An image screening was conducted prior to the image classification to remove cloudy images from the collection. The image classification was based on the Random Forest algorithm. An accuracy assessment using 20 randomly selected classified images indicated that the classified imagery has an overall accuracy of more than 85%. The classified images can be used for tracking sea ice dynamics over time and for providing reference for the interpretation of altimetry data.