Abstract

The Yellow River Delta (YRD) is a typical coastal wetland in China, dominated by coastal salt marsh vegetation, the YRD wetlands are continuously converted by inland sediment and impact of waves and storm surges, so the accurate classification of the coastal wetlands in the YRD is important for the development and protection of wetland resources. In this study, we adopted deep learning methods and integrated remote sensing information to accurately identify the coastal wetland types of the YRD. At first, to construct a classification system for considering the features of the YRD, and then the labelled data are annotated based on the Sentinel-2 satellite data by using the measured data as a reference basis. Secondly, the CNN, ResNet-50 and MobileNetV1 were selected for fine classification of coastal wetland. In order to improve the classification accuracy, this paper introduces the coordinate attention mechanism in the ResNet-50, and the new network named CAResNet-50 was proposed. The results show that, compared with ResNet-50, the kappa coefficient of the classification result of CAResNet-50 is improved by 0.03, and the overall accuracy is increased by 2.68%. After the comprehensive analysis of several evaluation indexes, the CAResNet-50 model has the best classification performance at the coastal wetland classification in the YRD. The results indicated that the CAResNet-50 model, which introduces the coordinate attention mechanism, provides a new method for the accurate classification of coastal wetland in the region scales.