.png){kind=icon}
.png){kind=icon}
Return
CenterNet-GCA: An improved deep learning model for automated identification of gravel clusters in gravel-bed rivers
.png){kind=icon}
PRE
AI
Q
D
W
DOI：10.1016/j.geomorph.2026.110299.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Gravel clusters, as representative structures on gravel-bed surfaces, positively influence overall bed stability and benthic aquatic ecosystems. To investigate their effects on flow and sediment transport processes, quantitative characterization of these structures is essential. A direct quantification approach involves identifying individual gravel clusters on gravel beds. However, existing identification methods still rely to some extent on empirical judgment, and some require high-quality topographic data, limiting quantitative studies of cluster characteristics. Building upon the CenterNet deep learning object detection framework, this study introduces an enhanced model tailored for gravel cluster identification. First, to address the irregular shapes of gravel clusters, the improved model employs an anisotropic Gaussian distribution to construct center point heatmaps. Second, to fully leverage multi-scale contextual information, the model introduces a pyramid structure to fuse feature maps of different scales, thus preserving both deep semantic information and shallow spatial details. This enhances the model's discriminative capability when objects and backgrounds are highly similar (e.g., in gravel beds). Validation results show that the improved model enables fast and accurate identification of gravel clusters from images. Application to a new dataset demonstrates that the model trained on field riverbed data exhibits strong generalization ability. The proposed model provides a powerful tool for quantitative studies of gravel clusters, facilitating the revelation of quantitative relationships between cluster development and sediment initiation and transport processes.
Keywords:
Gravel beds
Object detection
Machine learning
DEMs
Photographs
Journal
IF:
3.3
Papers:
8.5K
Citations:
2.8W
Organization
Cited Papers