Page 65 - GIS for Science Technology Showcases

REVEALING SUNKEN
SHIPS WITH GEOAI Deep learning detection extracts shipwreck locations
Madhu Hosuru, Vinay Viswambharan, Guneet Matreja, and Craig Greene, Esri
Geospatial artificial intelligence (GeoAI) is the combination of using location intelligence and GIS with the deep learning capabilities of AI. While GeoAI is often used for object and image classification, it can also be used for object detection, such as finding shipwrecks at the bottom of the ocean. The benefits of locating thousands of shipwrecks include:
• Updating navigational hazards for maritime mapping and charting
• Preserving maritime heritage, such as the National Oceanic and Atmospheric Administration database of shipwrecks
• Supporting programs such as the Remediation of Underwater Legacy Environmental Threats project, which identifies shipwreck locations and response efforts
• Introducing shipwrecks as potential artificial reefs to benefit ocean ecosystems
ArcGIS Pro deep learning–based object detection models were used to extract shipwreck information from bathymetric survey data and automate the task of detecting shipwrecks. Deep learning is a type of machine learning that relies on multiple layers of nonlinear processing for feature identification and pattern recognition described in a model.
The survey data was in Bathymetric Attributed Grid (BAG) format. BAG data is two-band imagery in which one band represents elevation and the other band represents uncertainty in the elevation values.
The model was trained on a ResNet-50 backbone containing more than a million images, 50 layers deep, and set for 80 epochs, the number of times the dataset is passed forward and backward through the neural network. Model training achieved an average precision score of 0.94, meaning the model was well trained.
The resulting trained model and model definition file were used to perform inferencing. In this case, the inference process searched for and identified shipwrecks in the BAG dataset.
Finally, the ArcGIS shaded relief function was applied, as shown in this map and inserts, to provide a 3D representation of the terrain and differentiate shipwrecks from the background.
Technology Showcase 199