在较大范围中更准确地提取小微湿地，对于进一步加强生态文明建设具有重要意义。本文结合物候特征与SAR影像特性建立了湿地的分类体系，并运用MultiRocket-RF时序分类模型基于地物在不同物候期对Sentinel-1雷达波束的散射差异实现湿地提取。此外，在以5hm2为面积阈值界定小微湿地的基础上，提出了一种缓解湿地斑块粘连问题的方法，以减少小微湿地被误分为非小微湿地。最后，以长三角生态绿色一体化发展示范区为例，分析与验证了小微湿地提取模型的有效性。结果表明：（1）SAR时序数据与MultiRocket-RF多元时序分类模型的结合能够较好地适应基于物候的湿地分类体系，对各类湿地的提取效果优良，总体精度达93.6%，Kappa系数达0.888，Macro-F1得分达0.804，尤其对于小微湿地中的淹水草本与木本植被、季节性淹没区的提取更具优势；（2）对湿地斑块粘连问题的缓解能够提升对小微湿地的提取精度，虽然Kappa系数没有显著变化，但Macro-F1得分由0.798提升至0.804，且混淆矩阵中各类小微湿地的提取准确率整体更优；（3）模型较适用于提取1hm2以上的小微湿地，而由于斑点噪声、几何失真等影响，对于1hm2以下小微湿地提取的效果欠佳。本文在拓展SAR数据应用领域的同时，也为小微湿地的提取提供了一种新的技术思路。

Accurately extracting small wetlands over larger areas is of great significance for enhancing the efficiency of wetland monitoring and conservation, as well as further promoting ecological civilization construction. From the perspective of phenology, this paper first established a wetland classification system, including permanent water bodies, flooded herbaceous vegetation, flooded woody vegetation, seasonal inundation areas and paddy fields, based on the characteristics of SAR images, and then constructed the time series of backscatter coefficients and coherence coefficients from Sentinel-1 VV and VH polarization modes. Subsequently, the MultiRocket-RF time series classification model was applied to extract wetlands based on the scattering differences of radar beams from various land cover types during different phenological periods. In addition, based on using 5hm2 as area threshold to divide small wetlands and large wetlands, this paper also proposed a morphological method to alleviate the problem of wetland patch adhesion and reduce the misclassification of small wetlands as large wetlands. Finally, considering that the natural attribute of the extensive wetlands and floodplains, as well as the social development characteristic of being committed to exploring new mechanisms for sustainable development, the Yangtze River Delta Ecological Green Integrated Development Demonstration Area was selected as the study area, the entire year of 2021 was chosen as the sample period. By comparing the performance of the models across various data sources and different sizes of small wetlands, this paper analyzed and verified the effectiveness of the small wetlands extraction method, which combines SAR time-series images with the MultiRocket-RF model. Results show that: (1) The combination of SAR time series data and MultiRocket-RF time series classification model could better adapt to the phenology-based wetland classification system, and had excellent performance in extracting various types of wetlands. The overall accuracy reached 93.6%, the Kappa coefficient reached 0.888 and the Macro-F1 score reached 0.804. Notably, the model was particularly advantageous for identifying flooded herbaceous vegetation, flooded woody vegetation and seasonal submerged areas in small wetlands. (2) By applying moderate morphological dilation and erosion to sever the small, narrow, connected wetland patches, the problem of wetland patch adhesion could be effectively alleviated, leading to an improvement in the accuracy of small wetland extraction. Although the Kappa coefficient did not change significantly, the Macro-F1 score increased from 0.798 to 0.804, and the extraction accuracy of various small wetlands in the confusion matrix was generally better. (3) The model was more suitable for extracting small wetlands larger than 1hm2 in this paper. However, due to the speckle noise, geometric distortion, and other negative effects caused by the side-looking characteristics and imaging principles of SAR satellites, the performance of extracting wetlands smaller than 1 hm2 was suboptimal, with a significant deviation from the true distribution. This paper not only expanded the application fields of SAR data, but also provided a new technical idea for the extraction of small wetlands, which contributes to further promoting the refined and holistic development of wetland ecological conservation and scientific management.