Deep Learning with Geospatial Data

Deep Learning with Geospatial Data is a quote-based AI and machine-learning service from GeoWGS84 Corp., a geospatial company headquartered in Parker, Colorado, for organisations that need deep-learning models and analysis built for their geospatial imagery. GeoWGS84 collects and labels training and test data, trains deep-learning models, analyses the customer's data with them and generates analysis reports. The listed tasks are object detection with accuracy assessment, image classification and labelling, pixel classification, semantic and instance segmentation, and change detection.

The provider describes land cover and land use classification "based on features extracted from satellite imagery or other types of geospatial data", object detection and tracking (ships, cars, fleets, delivery vans or trucks), spatial prediction, geocoding and map visualisation. Named applications include locating impervious surfaces, agriculture, road networks, construction monitoring, damaged-roof detection for real estate, building maintenance, and wind-turbine and solar-farm inspection; AI can also support 3D modelling and DTM/DSM mapping. For training-data collection the team uses infrared, high-definition RGB, multispectral and thermal sensors. A company blog post walks through object detection with a pretrained deep-learning model in Esri ArcGIS Pro; its screenshots show car detection on a 3 cm sample orthomosaic.

The service is separate from GeoWGS84.ai, the company's self-serve GeoAI platform. Projects are priced by quote.