Wetland classification based on improved BiSeNet using Sentinel-1 and Sentinel-2 data

Wetland classification using remote sensing data is significant for ecological environmental protection. In this paper, a high-precision wetland classification method was proposed to address the problem that wetland categories are easily misclassified. To improve the application of BiSeNet in wetland classification, the feature attention refined module (FARM), feature fusion module (FFM) and Dense connected Atrous Spatial Pyramid Pooling (DenseASPP) were designed to form Dense-BiSeNet. In addition, we adopted Short-Term Dense Concatenate network (STDC network) as backbone network. In order to validate the effectiveness of the model, we performed comparison on the Yellow River Delta and Dafeng Nature Reserve dataset between Dense-BiSeNet and other advanced wetland classification methods. The results show that the model has high accuracy and has objective potential for application.