Use of Ontology to Facilitate the Creation of Synthetic Imagery of Industrial Facilities

Transcription

1 LA-UR Use of Ontology to Facilitate the Creation of Synthetic Imagery of Industrial Facilities Paul Pope Los Alamos National Laboratory Los Alamos, NM, USA Randy Roberts Lawrence Livermore National Laboratory Livermore, CA, USA SEMANTIC TECHNOLOGY FOR INTELLIGENCE, DEFENSE, AND SECURITY STIDS 2011 Center of Excellence in Command, Control, Communications, Computing, and Intelligence C4I George Mason University Fairfax, Virginia November, 2011

2 Acknowledgements Ian Burns (LANL) synthetic imagery of industrial facilities Luci Gaines (LANL) hierarchical graph of industrial facility types Raju Vatsavai (ORNL) industrial facility ontology illustration Shaun Gleason (ORNL) & Alexei Skurikhin (LANL) Geospatial Verification & Validation Assessment team Ontology Assessment and V&V teams 2 anon STIDS2011 reviewers DOE/NA22 Simulations, Algorithms, & Modeling (SAM)

3 A picture is worth a thousand words. (click for link) If a picture is worth a thousand words......then give me the words!

4 Image-to-Words (I2W): example 1 The surrounding area is rural. There is a facility located in the center of the image. There are three buildings. There is one tank farm with eight cylindrical tanks. The facility is a chemical processing facility.

5 Image-to-Words (I2W): example 2 The surrounding area is rural. There is a facility located in the center of the image. There are three buildings. There is one tank farm with eight cylindrical tanks. The facility is a chemical processing facility.

6 Image-to-Words (I2W): example 3 The surrounding area is rural. There is a facility located in the center of the image. There are three buildings. There is one tank farm with eight cylindrical tanks. The facility is a chemical processing facility.

7 Auto-Annotation Rapid increase in RS imagery => automation required Very difficult problem => computer vision!= human vision Gold Standard: Imagery Analyst (IA)

8 Goal Create suites of synoptic ( birds eye ) imagery of industrial facilities which can be used to conduct verification & validation of auto-annotation algorithms. Real For example: Geographic Object-Based Image Analysis (GEOBIA) Composite Synthetic

9 Geospatial Auto-Annotation V&V START (user input, e.g., XML file) Adequate?

10 Benchmark Imagery Requirements Synoptic Imagery => Current benchmark imagery datasets are unsuitable Caltech 101 PASCAL Large number of images => synthetic imagery ORIDS Generic synthesis framework => wide range of industries Realism => * number, size, and type of objects consistent with a particular industry & prod. capacity * size consistent with structural engineering & product availability * layout consistent with process flow * photorealism => computer graphics * consistent with SME

11 Proposed System START FINISH Other Outputs: Textual Descriptions Object Segmentations Object Annotations

12 Current Capabilities & Future Work ~ ~ ~ Legend We have already substantially realized this capability We have made some progress already on this capability We have not had an opportunity to make substantial progress on this capability Other Outputs: Textual Descriptions Object Segmentations Object Annotations

13 Hierarchical Graph of Industrial Facility Types T. Chisnell and G. Cole, Industrial Components: A Photo Interpretation Key on Industry, Photogrammetric Engineering, vol. 24, pp , March 1958.

14 Industrial Facility Ontology is-a has-part

15 Why do we need ontology? Because we need an explicit, formal representation of knowledge about a domain of application...[that] includes: Types of entities that exist in the domain Properties of those entities Relationships among entities Processes and events that happen with those entities where the term entity refers to any concept (real or fictitious, concrete or abstract) that can be described and reasoned about within the domain of application. -Paulo Costa, STIDS2011 workshop

16 Why do we need ontology? Because we need... Consistency in vocabulary, definitions, & relations Translate between a variety of databases Inherent alignment with industrial facility SME (domain) Connection to other efforts (e.g. remote sensing, GIS) (upper-level ontology, geo-ontology; span domains) Part-Whole relations (layout, flow) => Mereotopology Insights into new auto-annotation algorithm design

17 Discussion LA-UR Use of Ontology to Facilitate the Creation of Synthetic Imagery of Industrial Facilities Paul Pope Los Alamos National Laboratory Los Alamos, NM, USA Randy Roberts Lawrence Livermore National Laboratory Livermore, CA, USA SEMANTIC TECHNOLOGY FOR INTELLIGENCE, DEFENSE, AND SECURITY STIDS 2011 Center of Excellence in Command, Control, Communications, Computing, and Intelligence C4I George Mason University Fairfax, Virginia November, 2011

18 Backup Slides

19 GSV Conceptual Model System Response Quantity (SRQ)

20 Intrinsic & Extrinsic Factors Full-factorial design experiment = #samples*levels^factors = 3*3^12 = 1,594,323 images.