Oasis agriculture in arid regions often faces a fundamental tension between limited water resources and socioeconomic development. Accurately characterizing the current status and temporal evolution of cropping structures is essential for formulating adaptive agricultural policies, optimizing spatial planting layouts, and ensuring regional food security. As a representative example of oasis agriculture in China's arid zones, the Turpan-Hami Basin serves as a major production base for grapes and Hami melons, earning the reputation of being the "Land of Fruits and Melons in China". In this study, long-term Landsat series imagery, in conjunction with crop phenological information, was utilized to develop a Temporal Extreme-Value and Visual Separation (EVS) method for extracting sensitive remote sensing features. A random forest-based crop identification model was subsequently constructed to generate high-precision datasets of cropping structures and to analyze their spatiotemporal dynamics. The results demonstrate that the proposed model effectively extracted agricultural planting information, achieving overall accuracies exceeding 85 % and Kappa coefficients above 0.79 across all years. From 1990 to 2023, the agricultural planting area in the Turpan-Hami Basin expanded from 104.95 x 103 ha to 191.90 x 103 ha. Moreover, the findings reveal a distinct transformation in the regional cropping structure, shifting from traditional grain cultivation toward high-value economic crops such as melons, grapes, and cotton. This research provides a robust data foundation and methodological reference for the scientific management and optimization of cropping structures in arid oasis agricultural systems.