Abstract: Wetland vegetation is a critical component of the Yellow River Delta ecosystem. However, its distribution is highly heterogeneous, with similar spectral characteristics among different vegetation types. Additionally, periodic submergence by water bodies can lead to significant spectral variability within the same vegetation type, posing challenges for remote sensing-based classification and recognition and reducing efficiency. Utilizing the Google Earth Engine (GEE) cloud computing platform and the Harmonic Analysis of Time Series (HANTS) algorithm, phenological characteristics were extracted to determine the optimal time phases for vegetation distribution mapping. Ground survey data and multi-source remote sensing imagery were employed to design four classification schemes, and machine learning algorithms were used to create multi-temporal vegetation classification maps for the years 2016–2023. Overall classification accuracies exceeded 93.38%, with Kappa coefficients higher than 0.92. Among the schemes, the object-oriented classification method combined with radar polarization features yielded the highest accuracy, particularly in distinguishing between Phragmites australis (common reed) and Spartina alterniflora, with accuracies exceeding 97%. The accuracy for Suaeda salsa was slightly lower, though still close to 92%. Analysis of multi-year spatiotemporal vegetation evolution revealed that the area of Spartina alterniflora increased from 35.6 km² to 52.2 km² before being sharply reduced to less than 10 km² due to cutting, while Phragmites australis displayed a decrease-then-increase trend, with its minimum area at 74.3 km². The area of Suaeda salsa showed a fluctuating decline. The distribution of Phragmites australis and Suaeda salsa was closely aligned with river courses, and cutting significantly curtailed the southward expansion of Spartina alterniflora into the protected zone, achieving an effective overall removal. The evolution of vegetation growth was found to be heavily influenced by human activities. This study demonstrates that the proposed method offers high accuracy, fast computation, and low memory consumption, making it well-suited for wetland vegetation monitoring in the Yellow River Delta. It provides theoretical support for ecological conservation and high-quality development in the region.