Signal Processing, Inc. A High Performance Algorithm to Improve the Spatial Resolution of HyspIRI Images 2012 NASA HyspIRI Science and Applications Workshop October 16-18, 2012 Signal Processing, Inc. Jin Zhou, Hua-mei Chen, Bulent Ayhan, and Chiman Kwan SBIR/STTR DATA RIGHTS Contract #: NNX11CI20P Topic #: S6.03 (2011) Contractor Name: Signal Processing, Inc. Contractor Address: 9700 Great Seneca Highway, Rockville, MD 20850 Expiration of SBIR/STTR Data Rights Period: August 18, 2017 The Government's rights to use, modify, reproduce, release, perform, display, or disclose technical data or computer software marked with this legend are restricted during the period shown as provided in paragraph (b)(4) of the Rights in Noncommercial Technical Data and Computer Software--Small Business Innovative Research (SBIR) Program clause contained in the above identified contract. No restrictions apply after the expiration date shown above. Any reproduction of technical data, computer software, or portions thereof marked with this legend must also reproduce the marking. Proprietary Information - SPI 1 1. Research Motivations • Signal Processing, Inc. Can we use low resolution (60 m) images to get material classification accuracy close to that of 15 m resolution? The goal is to improve HyspIRI imager performance by using advanced software algorithms. Mineral distribution (AVIRIS-15m res) Proprietary Information - SPI Mineral distribution (HyspIRI-60m res) 2 2. Phase 1 Technical Approach Signal Processing, Inc. LR hyperspectral image HR color or multi-spectral image Super-Resolution Algorithm Change Detection/ Material Classification Algorithms HR hyperspectral image • • Images need very accurate registration Possible applications: damage assessment, material classification, change detection, military surveillance and reconnaissance Proprietary Information - SPI 3 2. Phase 1 Technical Approach Signal Processing, Inc. Approach 1: 1 LR hyperspectral image + 1 HR color X  Tx where X is one (or more) hyperspectral pixels and x is one (or more) color pixels Proprietary Information - SPI 4 3. Phase 1 Results Signal Processing, Inc. Images 267x342x124 461nm to 901nm AF Wright Patterson Lab Proprietary Information - SPI 300x300x213 380nm to 2500nm Naval Postgraduate School (F.A. Kruse) 5 Signal Processing, Inc. 3. Phase 1 Results Hybrid color mapping vs. variational pan-sharpening (UCLA) • Our hybrid approach generates more accurate SR image UCLA method interpolation Proprietary Information - SPI Our Ours 6 Signal Processing, Inc. 3. Phase 1 Results Hybrid color mapping vs. variational pan-sharpening (UCLA) • Our hybrid approach generates more accurate SR image interpolation UCLA method Proprietary Information - SPI Ours 7 3. Phase 1 Results Signal Processing, Inc. Classification Results Proprietary Information - SPI Bicubic 8 3. Phase 1 Results Signal Processing, Inc. Classification Results Proprietary Information - SPI Our algorithm 9 3. Phase 1 Results Signal Processing, Inc. Classification Results Proprietary Information - SPI Ground truth 10 3. Phase 1 Results Signal Processing, Inc. Classification Results Ours Proprietary Information - SPI 11 4. Conclusions and Future Research Signal Processing, Inc. With more HR multispectral bands, we can perform even better Proprietary Information - SPI 3 color bands + 1 multispectral band 12 4. Conclusions and Future Research Signal Processing, Inc. With more HR multispectral bands, we can perform even better Proprietary Information - SPI 3 color bands + 3 multispectral bands 13 5. Future Research Signal Processing, Inc. • Task 1: Combine LANDSAT with HyspIRI images or Hyperion with ALI images. - Obtain actual images of the same scene collected at different times - Obtain signatures of certain materials. • Task 2: Combine Geoeye with HyspIRI images. – Customize algorithms – Improve speed of algorithms • Task 3: Develop integrated tools - Registration - HR hyperspectral generation - Change detection/target detection algorithm • Task 4: Real-time hardware prototype (HyperX) Each Hx3100 has 8 input ports and each port can accept 250 Mega 16-bit words of data, which can be processed in real-time, allowing more than 40 tetra bytes of incoming data for a 6 hour mission. • Task 5: Integration with NASA’s IPM Proprietary Information - SPI 14