Determination of River Hydromorphological Features in Low-Land Rivers from Aerial Imagery and Direct Measurements Using Machine Learning Algorithms

Hydromorphology of rivers assessed through direct measurements is a time-consuming and relatively expensive procedure. The rapid development unmanned aerial vehicles machine learning (ML) technologies enables the usage images to determine hydromorphological units (HMUs) automatically. application various indirect data sources their combinations for determination river HMUs from was main aim this research. Aerial with without Sobel filter, layer boulders identified using Yolov5x6, depth streamflow velocity were used as sources. Three ML models constructed cases if one, two, or three used. HMU segmentation MobileNetV2 pre-trained on ImageNet feature extraction part U-net part. stratified K-fold cross-validation five folds carried out evaluate performance model due limited dataset. analysis results showed that measured metrics close ones trained only combination boulder filter. obtained demonstrated potential applied approach images, provided basis further increase its accuracy.