Water, as an important part of ecosystems, is also an important topic in the field of remote sensing. Shadows and dense vegetation negatively affect most traditional methods used to extract water body information from remotely sensed images. As a result, extracting water body information with high precision from a wide range of remote sensing images which contain complex ground-based objects has proved difficult. In the present study, a method used for extracting water body information from remote sensing imagery considers the greenness and wetness of ground-based objects. Ground objects with varied water content and vegetation coverage have different characteristics in their greenness and wetness components obtained by the Tasseled Cap transformation (TCT). Multispectral information can be output as brightness, greenness, and wetness by Tasseled Cap transformation, which is widely used in satellite remote sensing images. Hence, a model used to extract water body information was constructed to weaken the influence of shadows and dense vegetation. Jiangsu and Anhui provinces are located along the Yangtze River, China, and were selected as the research area. The experiment used the wide-field-of-view (WFV) sensor onboard the Gaofen-1 satellite to acquire remotely sensed photos. The results showed that the contours and spatial extent of the water bodies extracted by the proposed method are highly consistent, and the influence of shadow and buildings is minimized; the method has a high Kappa coefficient (0.89), overall accuracy (92.72%), and user accuracy (88.04%). Thus, the method is useful in updating a geographical database of water bodies and in water resource management.

The main challenge in extracting coastal aquaculture ponds is how to weaken the influence of the “same-spectrum foreign objects” effect and how to improve the definition of the boundary and accuracy of the extraction results of coastal aquaculture ponds. In this study, a recognition model based on the U2-Net deep learning model using remote sensing images for extracting coastal aquaculture ponds has been constructed. Firstly, image preprocessing is performed to amplify the spectral features. Second, samples are produced by visual interpretation. Third, the U2-Net deep learning model is used to train and extract aquaculture ponds along the coastal region. Finally, post-processing is performed to optimize the extraction results of the model. This method was validated in experiments in the Zhoushan Archipelago, China. The experimental results show that the average F-measure of the method in the study for the four study cases reaches 0.93, and the average precision and average recall rate are 92.21% and 93.79%, which is suitable for extraction applications in aquaculture ponds along the coastal region. This study can quickly and accurately carry out the mapping of coastal aquaculture ponds and can provide technical support for marine resource management and sustainable development.

No abstract