Abstract
Spartina alterniflora is one of the most important invasive alien plants in coastal wetlands in China, and its rapid expansion has caused serious negative ecological impacts. Current researches mainly have focused on the monitoring of S. alterniflora using moderate-resolution satellite images at the regional or landscape scale, while the monitoring capability of early invasion at the patch scale is limited, and it is difficult to obtain satellite images stably due to cloud and tidal interference. In this paper, we propose a semi-automatic monitoring framework for patch-scale monitoring of S. alterniflora based on deep learning techniques with unmanned aerial vehicle (UAV) RGB images. First, the spectral characteristics of S. alterniflora in the visible light band were analyzed, and Visible-band Difference Vegetation Index (VDVI) with the strongest stability for this feature type was selected for fusion. Second, a dataset of S. alterniflora on both sides of the Yellow River estuary was constructed and trained using the improved U-Net++ deep learning network. Finally, prediction and accuracy verification were carried out for some dense distribution areas of S. alterniflora. The experimental results show that the overall accuracy (OA) of the prediction results obtained by this method is 98.1%, and the kappa coefficient is 0.960, which is better than the traditional classification methods, and has certain significance for the patch scale monitoring of S. alterniflora.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call