面向河湖水域的AI半自动标注方法及软件设计OA

Currently,mainstream open-source annotation tools often require manual outlining and delineation of water conservancy targets during semantic annotation,which is time-consuming,labor-intensive,and prone to inaccuracies.To address these issues,a semi-automatic annotation method for river and lake water bodies was proposed,combining MAFs(Morphological Attribute Filtering)and SAM(Segment Anything Model).Interactive prompt points were input into the SAM prompt encoder to obtain initial segmentation masks.MAFs was then applied to the masks to eliminate noise such as"isolated islands"and"speckles"in the initial SAM results,ultimately producing high-quality masks.Experiments and software development were conducted using more than 12 000 UAV-acquired water body images.The results showed that,compared with traditional annotation tools,the average annotation time was reduced from 20～30 seconds per image to 2～3 seconds per image.In addition,noise interference such as"isolated islands"and"burrs"was reduced,resulting in higher-quality annotated images.These findings indicate that the proposed method achieves high annotation efficiency and quality,thereby shortening sample preparation time and improving sample quality in semantic model training.