Abstract: The coastal wetland ecosystem is facing a severe threat from the secondary invasion of Spartina alterniflora. In order to achieve high-resolution and refined dynamic monitoring, a precise remote sensing classification method for Spartina alterniflora invasion based on drone multidimensional feature fusion is proposed-Multi-fea-ture Fusion Classification (MFC). This method first utilizes the DJI Mavic 3M drone to obtain multi temporal unmanned aerial vehicle remote sensing images with a resolution of 5 cm, including multispectral images and visible light images. Secondly, based on the Gini index feature selection strategy, 15 key features are selected from multidimensional features such as vegetation index, spectrum, and texture. Then, a Random Forest (RF) classification model was trained to classify the multi temporal unmanned aerial vehicle remote sensing images. Taking the Binhai Wetland in Nanhui New Town, Shanghai as the research area, based on the MFC method, the spatiotemporal dynamic changes of the secondary invasion of Spartina alterniflora were systematically analyzed. Replace the RF model in the MFC method with Artificial Neural Network (ANN) and Support Vector Machine (SVM) models respectively, and conduct comparative experiments. The results showed that the highest feature importance in spring, summer, and autumn was the mean of the first principal component (PCA MEA_1, 6.69%), Gaussian high pass filtering in the near-infrared band (HIGH_NIR, 5.78%), and the mean of the second principal component (PCA MEA_2, 7.82%), respectively. The key features that dominated the recognition accuracy of Spartina alterniflora varied in different seasons. The average overall accuracy (OA) of the RF model in different seasons is 96.47%, significantly better than the Support Vector Machine (SVM) model and the Artificial Neural Network (ANN) mo-del. The proportion of Spartina alterniflora area increased from 55.23% in spring to 62.13 % in summer, and then decreased to 49.48% in autumn, showing typical spatiotemporal characteristics of "local outbreak and marginal diffusion". Its secondary invasion is significantly correlated with phenological changes, tides, and human intervention. Research has confirmed that the MFC method based on multidimensional feature fusion dynamic adaptation mechanism can effectively improve the classification robustness in complex habitats, providing high-resolution technical support for accurate monitoring and patch management of invasive species in coastal wetlands.