MAPSS-DeepLearning

The MAPSS project developed and executed Deep Learning techniques using Supervised Learning functions in order to identify potential archaeological features such as funerary and habitation sites across Mongolia. These techniques automate remote site discovery by using ArcGIS Pro’s dedicated Deep Learning Toolset, which features a third-party Deep Learning Python API (TensorFlow, PyTorch, or Keras) to detect pixel clusters (‘objects’) in an image, deploying a specified Python raster function to process each object. By training the system using known datasets of Heritage Resources and ready-made Deep Learning Models such as MaskRCNN, YOLOv3, and Single Shot Detector, analysts run the selected model against an input raster image to produce results according to each object class (such as ‘burial mound’ or ‘winter camp habitation side’).