Abstract: The ionospheric influence is one of the major challenges in low-frequency radio astronomy data processing. Effects such as absorption, scattering, and Faraday rotation caused by the ionosphere can lead to positional shifts of target sources in images obtained by ground-based telescopes, with the impact being particularly significant at low frequencies. This paper first designs a fully connected network based on an Autoencoder (AE) model to classify telescope observation images into "good" and "bad" categories. Subsequently, an unsupervised learning-based algorithm for solar position identification, tracking, and correction is developed. Through a target tracking and correction network combined with an iterative validation model, this algorithm performs position correction on video images where the target source position is shifted due to ionospheric effects. Experiments conducted on LOFAR solar image sequences (105 videos, 23,292 images; frequency range 19.9–78.5 MHz) demonstrate that the proposed algorithm effectively corrects ionospheric positional deviations. It converges within an average of 7.17 iterations at a processing speed of 6.18 frames per second (fps).