A TRUSTED DISRUPTOR FOR INNOVATIVE GEOINT
What are Autonomous Intelligence Solutions?
L3Harris world-class engineering services and solutions delivery platform provides trusted, warfighter-ready intelligence at scale that our customers use to discover, predict and respond to events around the world. We enable accelerated insights and help our customers progress toward mission autonomy.
WHO WE ARE
WHAT WE DO
For over 40 years, L3Harris Autonomous Intelligence Solutions (AutoIntel) has been a trusted innovator for developing and managing advanced GEOINT solutions that collect, process, analyze and deliver remotely sensed data and information to solve our customer’s toughest mission challenges.
L3Harris Autonomous Intelligence Solutions is engineering innovative solutions that accelerate operationalizing artificial intelligence/machine learning (AI/ML) into the enterprise. We deliver simplified, scalable enterprise solutions and transform customers’ tactical edge across all domains.
Privacy Policy
| Terms of Use
AutoIntel Engineering Services
L3Harris’ advanced engineering solutions provide critical data and analytics that deliver actionable information to the warfighter.
DATA ASSURANCE
DATA CONDITIONING
ADVANCED ANALYTICS
SENSEMAKING
MODELING AND SIMULATION
We invest in data integrity and assurance capabilities that detect and defend warfighters against deep fakes, disinformation and other attacks to deliver more robust end-to-end AI/ML systems.
Our services improve the routing, optimization and protection of critical data through data preparation and structured observation management harvesting. An automated workflow using natural language processing (NLP) harnesses the power of historical, human-derived intelligence reports as an alternative to the expensive, traditional manual process for labeling data. This capability can reduce timelines for labeling datasets by a factor of 10.
• • •
Automated target recovery – Automatically constructs 3D models of targets, then scans the volumetric model forward in time to identify candidate objects matching the target All weather persistent surveillance – Blends EO and SAR data to use either modality (as available) to continuously monitor a site of interest
High-precision 3D volumetric analysis – Enables high-resolution mensuration of volumetric model of an area of interest High-precision volumetric time-series (4D) analysis – Enables measurements of change in a volumetric model over time Real-time Earth observation (EO) and synthetic aperture radar (SAR) change detection – Generates a predicted image matching the collection condition and performs a difference as soon as the new image is available
• •
L3Harris is a leader in the application of knowledge graphs to model decision support using real multi-intelligence data sources. Our models operate at multiple levels so analysts and decision makers can understand complex real-world data at the following levels:
Discovering and describing objects and relationships Monitoring the states of objects and relationships over time Characterizing activities, interpreting the meaning of state changes with context Inferencing about activities, enabling probabilistic forecasting and hypothesis testing
• • • •
L3Harris has been the trusted industry partner for high-fidelity imagery simulation that form the basis for many critical activities:
Design analysis Mission analysis Performance prediction Test dataset generation
Data calibration System checkout Anomaly resolution
We provide high-fidelity end-to-end spectrally and radiometrically-accurate electro-optical image chain simulation that accounts for all key components of the imaging chain from illumination all the way through image product formation.
GET TRUSTED AI, FAST.
IntelliEarth™ Integrator-G is a GOTS AI/ML mission solutions delivery platform that modernizes enterprise workflows by accelerating operationalization of AI/ML models. By eliminating software licensing fees, our proven capability is the most cost-effective way for government agencies to deploy AI/ML algorithms and models into and across the enterprise.
• • • • • •
Supports full-spectrum data sources Includes validated synthetic training data capabilities Enables algorithm lifecycle management integrated governance Establishes trust in AI/ML algorithms with robust test and evaluation components Natively captures provenance of any observable object generated by the system Allows for analyst feedback for retraining and continuous refinement
IntelliEarth™ Integrator-G:
SOLVE KEY CHALLENGES
IntelliEarth™ Integrator-G contains three key customizable components needed for machine learning operations:
Data management
Algorithm governance
Production environment
Production environment: IntelliEarth™ Integrator-G closely ties the highly iterative process of AI/ML algorithm development to its operational use and provides a mechanism for routine feedback and continuous improvement.
Data management: Store data within a local network or in the cloud. Use files at rest, prevent multiple copies of the same dataset and ensure the pedigree of the data used within a workflow.
Algorithm governance: Customizable lifecycle tracking of machine learning analytics from development to operations ensures user confidence in the results generated from algorithms.
RETURN TO TRUSTED AI, FAST.
SYNTHETIC TRAINING DATA
Our world-class training data improves quality and increases reliability of AI/ML models and algorithms so analysts and warfighters can be confident when making decisions.
Scene
System
Processing
Scene Preparation
Atmospheric Modeling
· Image-based simulation:
· MODTRAN:
Geolocation Flight path Orbital parameters Pointing stability
Optical Modeling
Optical defects: Stray light, Focus
Sensor Modeling Characteristics
· Pixel number and size · Dark current · Read noise · Fill factor
Collection Geometry
Power Spectral Density
· Compression · Calibration · Registration · Pan Sharpening
· Image Stitching · Geolocation · Enhancements · Orthorectification
Throughput
WFE and Pupil
· ·
Quantum Efficiency
Jitter Smear
· Raytracing-based simulation:
Generate 2.5D / 3D scene, Apply material properties, Perform activity modeling
Back-out source characteristics, Map image to reflectance
· · ·
Visibility Atmosphere Type Custom .tp5 files
Sun Geometry
Location
Texture Variants
Occlusions
Confusers
Worldview copyright notice
· · · ·
Optical Prescription: Focal Length, Obscuration, Vignetting, Aperture Diameter
· Physics-based, radiometrically accurate · Custom service tailored to meet data needs · Cross-phenomenology (PAN, MSI, Thermal, and SAR) · GEOINT subject matter experts · Support up to TS/SCI data generation
Synthetic Data Services
· Enterprise level, GOTS, high TRL solution · Customizable to meet specific data generation needs · Easy to use (“point and click”) UI available · Manages end-to-end data generation process · Includes data store with search & discovery features
Synthetic Data Tools
OFFERINGS
Satellite Image © 2023 Maxar Technologies
L3Harris has been a trusted provided of GEOINT remote sensing synthetic data for over 50 years.
Why L3Harris?
Significant effort has gone into optimizing our synthetic data approach to best support AI/ML training and achieve performance.
Leveraged by the NRO, NGA, Air Force and commercial customers to support trade studies and AI/ML algorithm training.
Capability is tested at the simulation component level and generated imagery is validated against operational systems.
Remains focused on producing physics-based, radiometrically accurate simulations that represent end-to-end collection system.
Data generation capabilities were first established in the 1970s to support image chain analysis for remote sensing systems.
Established
Accurate
Validated
Trusted
Optimized
CAD Model: Background: Texture Variant: Sun Geometry: Camera Zenith Angle:
Xian H-6
L3Harris' Synthetic Training Data Services allow users to generate specific data needs using a customizable end-to-end data generation process. Below is a sample of the variables available for selection.
Sample DATA
Boeing 737
Location 1
Location 2
Location 3
Default
Variant
0° Zenith, 45° Azimuth
60° Zenith, 120° Azimuth
30° Zenith, 270° Azimuth
15°
40°
55°
RESULT 1
Satellite Images © 2023 Maxar Technologies
1 of 5
2 of 5
3 of 5
4 of 5
5 of 5
RESULT 2
RESULT 3
RESULT 4
RESULT 5
The U.S. intelligence community needed AI/ML algorithm technology to quickly and accurately assess Worldview image chips and identify whether a Xian H-6 bomber or semi-truck was present.
USE Case
OBJECTIVE
CNN Model: ResNet34
Semi-Trucks
Real Data: 546 H6, 546 semi-trucks, and 1092 backgrounds
1 of 3
1,278
2 of 3
SYNTHETIC DATA
different backgrounds
81
texture variants
3,834
synthetic images generated
51
4
400
99% Accuracy
20
100
3 of 3
RESULTS
When all available real data has been used, adding synthetic data boosted performance and achieved
The synthetic data only classifier achieved
73% Accuracy
across all classes without the need for any real image data.
1000
REAL DATA
Real Only
Synthetic + Real
ACCURACY
80
60
40
0
# OF REAL IMAGES USED
Synthetic Data Only
Combining synthetic data with 100 real samples achieved
97% Accuracy
surpassing performance achieved when using all available real data.
Look Ahead
High Fidelity 3d Scenes
MOVINT
Closed Loop Synthetic Data
Classifier
Generator
(Syn Real)
Discriminator
(Raw Real vs Fake Real)
Raw Synthetic Data
Raw Real Data
NLL Loss
CrossEntropy-Loss
CrossEntropy Loss (Real Labels) & CrossEntropy Loss (Fake Labels)
Gradient Path
Data Path
Both
3D Scene © 2023 PLW Modelworks
Closing the loop between data generation and model training will allow for optimization of data generation driven by model performance.
Ability to insert dynamic objects into the scene will allow for data generation of complex, activity-based scenarios.
Ability to leverage 3D scenes will allow for data generation that includes backgrounds with higher levels of realism.
Accuracy Counts, No Programming Required
L3Harris’ commercial off-the-shelf deep-learning technology is specifically designed to work with remotely sensed imagery and solve geospatial problems.
ENVI® Deep Learning at Work:
Urban Growth: The ENVI® Deep Learning module makes it easy to explore and assess the environment. The module was used to generate the landcover classification image shown here. When another image was generated the following year, traditional change detection workflows in ENVI® were used to approximate the human impact on the environment and detect objects like new buildings and well pads.
Agriculture: Using the ENVI® Deep Learning module, the locations of current and past lava flows in Hawaii were identified. This information was used in ENVI® to understand the environmental impact that the volcanic gasses had on local crops, which gave farmers insights for insurance claims and an ability to understand if crops are safe for human consumption.
Disaster Response: When disasters strike, response time is very important. The ENVI® Deep Learning module has been tuned so you don’t need thousands of samples to create models for finding features. After a recent hurricane, the module was used to quickly characterize different types of damage to buildings throughout the region ranging from partial to full destruction. First, a handful of small areas were labeled according to the extent of the damage. When the model was applied to the scene, the damaged buildings were automatically classified according to the extent of the damage they sustained.
