Curbing the continuous abandonment of large areas of farmland is important for meeting the global food demand and promoting agricultural and rural development. Accurate identification is the key to the effective management and utilization of abandoned farmland. The identification of abandoned land based on a long time series of remote sensing data has become rapid and effective. Therefore, a set of training and test datasets generated from invariant samples and reference sample sets is established in this paper. On this basis, the Google Earth Engine (GEE) is used to classify Landsat and Sentinel high-precision long-term remote sensing images from 2000 to 2022. In addition, a change detector based on the sliding window algorithm is proposed to extract abandoned farmland in the Huangshui Basin from 2002 to 2020, and the intensity, trend, frequency, reclamation rate and utilization efficiency are analyzed. The results revealed that the OA of land use classification in the Huangshui Basin from 2000 to 2022 was between 0.852 and 0.91, and the kappa coefficient was between 0.822 and 0.89, indicating a good classification effect. From 2002 to 2020, the accumulated abandoned farmland area in the Huangshui Basin continued to increase, showing a fluctuating upward trend, and the phenomenon of farmland abandonment and reclamation occurs repeatedly in some areas. From the overall distribution, the abandoned area gradually increased from the central region to the southeast. With the passage of time, the amount of abandoned farmland in the valley increased gradually, and the abandoned area was transferred from the high mountains to the valley area. The average annual abandonment rate of supplementary farmland was 50.45%, which was much greater than that of basic farmland. Most of the supplementary farmland could not be effectively and judiciously used, and the utilization efficiency was low. The research results provide data support for the reuse of abandoned farmland in ecologically fragile plateau areas, the formulation of targeted strategies, the implementation of timely adjustments, and the establishment of new ideas and methods for the accurate identification of abandoned farmland. • Developed a set of training and test data sets generated by invariant and reference samples. • Proposed a sliding window algorithm based on the change detector to plot long time series abandoned farmland. • Spatiotemporal distribution, trend, and recultivation of abandoned were analyzed. • Abandonment rate of supplementary farmland is higher than original farmland. [ABSTRACT FROM AUTHOR]