Nowadays, investigation into ionospheric conditions has found its applications in many fields such as short-wave communications, weather nowcasting, risk survey, real-time hazards prediction, and others (Trebing et al., 2021). The image data related to ionospheric electron density characteristics are generally produced by ionosondes based on vertical incidence, which can reach millions of pictures per year. As a result, traditional manual scaling manner could not be competent to huge ionograms scaling task in the short term. Thus, automatic scaling by developing computer programs has been considered (Fagre et al., 2021). The most-widely-used automatic scaling routine is automatic real-time ionogram scaling with true-height (ARTIST) (Galkin & Reinisch, 2008). Another common autoscaling method is Autoscala (Pezzopane & Scotto, 2007). Certainly, other various types of methods have been developed (Jiang et al., 2013;Tsai & Berkey, 2000). However, ionogram data inherently has the imbalance on the pixel points and on the obtained different layer traces, with the impacts from the man-made or other noises and the ionospheric phenomena such as spread F, it is still a challenge to reliably and accurately predict the E, F1, and F2 layers height trace for ionograms.