“Perennial soil erosion poses a serious threat to the black soil region in Northeast China, with erosion gullies being one of the main manifestations. At present, remote sensing technology has been widely used for monitoring and protecting erosion gullies, and has accumulated a large amount of annotated historical survey data. However, how to effectively use historical data to reliably extract gully information from the latest data taken by different sensors at different times is still a technical problem to be solved. To this end, relevant experts have proposed a cyclic self training framework (CSTF). This framework utilizes an object level pseudo label generation strategy to provide high-quality pseudo labels for the latest data during each self training process, and introduces a loss function based on the pseudo label trustworthiness factor to effectively alleviate the negative impact of pseudo label noise. To verify the effectiveness of CSTF, experts conducted a detailed comparative analysis with other advanced methods on two datasets in Huachuan County, Heilongjiang Province. The results indicate that CSTF has significant advantages in extracting erosion gully temporal phases, fully demonstrating its important potential and application value in promoting land monitoring and protection in the black soil region of Northeast China.”