AI-Driven Ground Points Extraction for Rugged Terrains in Coastal Landscape – A Case Study

Abstract. Coastal erosion is a great threat to the coastal ecosystem and is often quantified by retreating shoreline and the loss of sand from the coastal zone. To quantify the volumetric loss of sand, the ground and non-ground features should be separated, and erosion could be quantified using the extracted ground features. While there are several algorithms available for ground and non-ground classification, most tend to remove valid ground points that are crucial in accurate volumetric loss and gain estimates. This study proposes raster-based approach to extract topography of the ground with better reliability of rugged terrains where the surface is often over smoothened. In this approach 3D point cloud obtained from the UAS-SfM (Unmanned Aircraft System – Structure from Motion) technique is converted to raster with five bands including red, green, blue, elevation and slope with elevation and slope derived from the surface model. This approach uses a Random Forest Classifier (RFC), which utilizes all five bands to train the model. The classified ground points are transformed into a Digital Terrain Model (DTM) using the Inverse Distance Weight (IDW) interpolation technique. The DTM generated is cross validated with the orthomosaic and Digital Surface Model (DSM). The results shown that the DTM generated using this machine learning approach produced reliable result with RMSE 0.059m.