At a Glance
National Geospatial-Intelligence Agency
Develop a trainable deep learning feature identification system capable of detecting and localizing important security and combat-related features in 2D and 3D satellite & drone imagery.
Created pre-processing, training & inference system trained on a combination of open and secret datasets. Object labels have hundreds of classes and often 200k building examples per image. This system is an on-going project with dedicated UI, Flask/Docker App, batch processing, and introduction of new multi-class training approaches. The system is currently being deployed via AWS GovCloud for all terrains and training locations.
Expertise & Technology
It is challenging to transform raw source sensor data into corresponding geospatial information. Analyzing disparate sensor imagery and elevation and feature data is time-consuming and manpower intensive. Historically, automation of these tasks has been elusive or inefficient and has required a large amount of edge case modifications. However, recent advances in Artificial Intelligence (AI) and Machine Learning (ML) have created new opportunities to automate these tasks, and this project will apply such techniques to feature-to-image correspondence for electro-optical (EO), thermal (IR), synthetic aperture radar (SAR), and multispectral (MSI) imagery. With these advances, prove the technical feasibility of the deep-learning based system to accurately and quickly find key features quicker than a trained human with high accuracy.
SFL Scientific Solution
We can detect net new buildings and the time of change of a building footprint. This ability to automatically extract differences and changes in structures by streaming new images and data feeds, at any rate of data ingestion, can be integrated as a monitoring tool, with any number of statistics extracted and extended to LIDAR and other systems. It can also be used to infer economic, movement, and other regional patterns.
To improve capabilities, feature segmentation was developed. This segmentation capability is considered state of the art and rivals human performance on this scale to detect building differences. The solution is easily extendable to situations with more labels, such as roads, vehicles, or airports, and derived features such as average building size, occupied percentage of land, or presence of airplanes.
The developed deep learning system currently is able to process images of any building density and dimensions in under a minute using these modern detection methods with GPU acceleration, at over 95% accuracy.