Author(s): Liekai Cao; Danxun Li; Chao Qin; Xiaohu Guo
Linked Author(s):
Keywords: No Keywords
Abstract: Accurate and efficient identification of shorelines is a key technology to carry out movable bed model tests and provide basic data for riverbed evolution analysis. Addressing the imperative for rapid measurement of wide-range shorelines during movable bed model tests, this study proposes a shoreline measurement method based on the semantic segmentation model, DeepLabV3+. Firstly, the Deeplab v3+ network for shoreline detection is trained by applying the shoreline images captured in movable bed model tests. Then, the method adopts multi-camera overhead shooting, synchronously collects test images and stitches them into panoramic ortho-images. Subsequent uniform cropping of the ortho-images facilitates segmentation via the trained Deeplab v3+ network for shoreline recognition, enabling the extraction of boundary coordinates from local images. Finally, integration of these coordinates yields instantaneous shoreline coordinates and proof images. Application of this method in movable bed model tests, encompassing both straight and curved channels, reveals a recognition accuracy of 2mm, providing valuable insights into the continuous processes of shoreline alterations. The results demonstrate the capability of the proposed method to achieve automated real-time refinement of shoreline extraction over long distances and large-scale movable bed models.
Year: 2024