Abstract: To address the issues of easy confusion and low classification accuracy in wetland land use information extraction, this study proposed a wetland land use classification model based on the fusion of genetic algorithm and support vector machine preprocessed by random forest algorithm (RF-GA-SVM).Based on the data of Landsat 8 remote sensing imagery (bands 1~7), land use information extraction was conducted in Poyang Lake Nanji wetland.The results extracted by the RF-GA-SVM model were compared with the results by the fusion models based on traditional support vector machine (SVM), particle swarm optimization-optimized SVM (PSO-SVM), and gray wolf algorithm optimization-optimized SVM (GWO-SVM).The results showed that the RF-GA-SVM model achieved an overall accuracy of 99.2% in land use classification for the Nanji wetland of Poyang Lake.The feature optimization using the random forest algorithm effectively reduced feature redundancy, and the GA-SVM model classification further enhanced accuracy.Compared to other methods, the proposed model achieved higher classification accuracy while improving computational efficiency, enabling effective extraction of wetland land use information.The research findings can provide a reference for wetland landform identification and classification.