Automated riverbed composition analysis using deep learning on underwater images
The sediment of alluvial riverbeds plays a significant role in river systems both in engineering and natural processes. However, the sediment composition can show high spatial and temporal heterogeneity, even on river-reach scale, making it difficult to representatively sample and assess. Convention...
Gespeichert in:
Veröffentlicht in: | Earth surface dynamics 2023-11, Vol.11 (6), p.1061-1095 |
---|---|
Hauptverfasser: | , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | The sediment of alluvial riverbeds plays a significant role in river systems both in engineering and natural processes. However, the sediment composition can show high spatial and temporal heterogeneity, even on river-reach scale, making it difficult to representatively sample and assess. Conventional sampling methods are inadequate and time-consuming for effectively capturing the variability of bed surface texture in these situations. In this study, we overcome this issue by adopting an image-based deep-learning (DL) algorithm. The algorithm was trained to recognise the main sediment classes in videos that were taken along cross sections underwater in the Danube. A total of 27 riverbed samples were collected and analysed for validation. The introduced DL-based method is fast, i.e. the videos of 300–400 m long sections can be analysed within minutes with continuous spatial sampling distribution (i.e. the whole riverbed along the path is mapped with images in ca. 0.3–1 m2 overlapping windows). The quality of the trained algorithm was evaluated (i) mathematically by dividing the annotated images into test and validation sets and also via (ii) intercomparison with other direct (sieving of physical samples) and indirect sampling methods (wavelet-based image processing of the riverbed images), focusing on the percentages of the detected sediment fractions. For the final evaluation, the sieving analysis of the collected physical samples were considered the ground truth. After correcting for samples affected by bed armouring, comparison of the DL approach with 14 physical samples yielded a mean classification error of 4.5 %. In addition, based upon the visual evaluation of the footage, the spatial trend in the fraction changes was also well captured along the cross sections. Suggestions for performing proper field measurements are also given; furthermore, possibilities for combining the algorithm with other techniques are highlighted, briefly showcasing the multi-purpose nature of underwater videos for hydromorphological assessment. |
---|---|
ISSN: | 2196-632X 2196-6311 2196-632X |
DOI: | 10.5194/esurf-11-1061-2023 |