Biological invasion by Spartina alterniflora is widespread in coastal wetlands in China, threatening native vegetation. Obtaining the distribution of native and invasive vegetation at the small community scale is important to understand the invasion mechanism of S. alterniflora and the protection of wetland ecosystems. Remote sensing images can quickly and effectively identify vegetation at a large scale, which can have a considerable influence on the protection of biological diversity. Although Landsat-like space borne sensors have been applied in the monitoring of invasive species in wetlands, their spatial resolutions are often coarser than those of small communities of invasive species. To identify small-scale native (Phragmites australis and Scirpus mariqueter) and invasive (S. alterniflora) vegetation communities around Chongming Island in eastern China, we first investigated the performance of four classifiers based on spectral features. Then, textural features of Gaofen-1 images under an appropriate textural window size were added. Furthermore, a comprehensive classification (CC) method generated by multiple classification features was proposed to obtain a reliable classification, and CC was further improved by considering the properties of neighbouring pixels. The results demonstrated that (i) the random forest (RF) classifier reached a higher overall accuracy (OA) than all the other classifiers; (ii) adding textural features slightly increased the OA, but the production accuracy (PA) for P. australis and the user accuracy (UA) for S. alterniflora decreased; (iii) the CC method improved OA by up to 4.12 %, and accounting for the neighbouring pixels improved the OA to 93.96 %. The comprehensive method effectively improved not only OA but also PA and UA for the vegetation communities, and it can be used in other wetlands.