Tegra gra Go Goes s Ind ndustry: ustry: Emb mbedded edded Hyp ypersp erspectral ectral Imaging maging GPU Technology Conference Munich, 11 October 2017 Michael Schoeffmann Session Id: 23161 www.perception-park.com
Embedded Hyperspectral Imaging Introduction What is Hyperspectral Imaging? Hyperspectral Environment / Workflow Application Examples Embedded Hyperspectral Imaging www.perception-park.com 2
Introductory Example Rice Processing: Customers Expect Pure Rice www.perception-park.com 3
Introductory Example Rice Processing: Customers Expect Pure Rice In Production Rice Is Contaminated www.perception-park.com 4
Introductory Example Rice Processing: Customers Expect Pure Rice In Production Rice Is Contaminated Undesired, e.g. Wood www.perception-park.com 5
Introductory Example Rice Processing: Customers Expect Pure Rice In Production Rice Is Contaminated Undesired, e.g. Wood Forbidden, e.g. Maggots Indistinguishable in Visible Light www.perception-park.com 6
Introductory Example Rice Processing: Customers Expect Pure Rice In Production Rice Is Contaminated Undesired, e.g. Wood Forbidden, e.g. Maggots Indistinguishable in Visible Light Clearly visible in Extended Light Range www.perception-park.com 7
What is Light? Many object properties detectable in visible range, what if we extend this by the invisible range? www.perception-park.com 8
What is Light? Many object properties detectable in visible range, what if we extend this by the invisible range? Common ranges: Visible, Infrared Also used: Terahertz, UV www.perception-park.com 9
Light Measurement Molecules interact with light Light Source Object A certain wavelength may be consisting of Molecules reflected, transmitted, absorbed or Transmitted altered to some extent Light Transmitted and reflected light Reflected can be measured directly Light Measurement provides a Absorbed Light spectral fingerprint of a substance www.perception-park.com 10
Hyperspectral Data vs RGB Hyperspectral cubes can have any number of channels instead of only RGB Depending on the acquisition technology, currently up to 1000+ wavelength positions www.perception-park.com 11
Hyperspectral Data Models Hyperspectral models extract desired information Hyperspectral models project information to e.g. RGB or grayscale Hyperspectral models provide information in interpretable images www.perception-park.com 12
Hyperspectral Imaging – What we heard so far Light is not limited to the visual range Reflected or transmitted light provides a spectral fingerprint of examined objects Hyperspectral cubes contain much more information than RGB images Hyperspectral models provide desired information in interpretable images www.perception-park.com 13
Hyperspectral Environment - Motivation www.perception-park.com 14
Hyperspectral Imaging – Two Main Challenges Find the application relevant information. Apply the application in the field. www.perception-park.com 15
Hyperspectral Environment – Two-tier approach Deploy Perception STUDIO Perception CORE Acquire and work with hyperspectral data GPU accelerated processing engine for configurations • • WYSIWYG Editor for hyperspectral models Camera abstraction, standardization and calibration • • Compile configurations for real-time Standard machine vision interfaces • • processing Ready for Deployment in Industry • www.perception-park.com 16
Perception Studio – Develop an Application Customer needs no experience in Spectroscopy Image processing Acquisition technology Simple world of images Workflow Setup camera Acquire data cube Describe data cube Select method See output Generate model Apply to Core www.perception-park.com 17
Perception Core – Deployed Application Industrial PC Processing Box Chip Module Performance, Applicability cal.pac device RAM user ... ... Abstraction, Standardization k memory GEV GEV Compatibility streaming outp. camera input Application Camera Machine Vision UDP UDP Appl. ... ... CamLink CamLink configuration Eth. GPU PU line scan Usability, Selectivity area scan (snapshot) wavelength scan (MEMS) CI – Chemical Imaging CCI – Chemical Color Imaging C – Colorimetry IF – Interferometry 18 www.perception-park.com
Application Example: Pharmaceutical in Blister www.perception-park.com 19
Application Example: Mining www.perception-park.com 20
Application Example: Polymer Recycling www.perception-park.com 21
Application Example: API Distribution www.perception-park.com 22
Embedded Hyperspectral Imaging: Why Tegra? Performance Size, Weight & Power Consumption Efficient Code Development No Development Processing Power moving parts 23 www.perception-park.com
Embedded Hyperspectral Imaging: Special Purpose Device Blood Vessel Camera Object Output Vein Model Moisture Camera Object Output Moisture Model 24 www.perception-park.com
Distribution of Hyperspectral Systems www.perception-park.com 25
Thank you for your attention! www.perception-park.com
Recommend
More recommend