• Personal Career Note

• $\textcolor{red}{\text{Ground Segment System Software Engineering - Satellite Camera Design, Calibration and Image Preprocessing}}$ ↗

  • Pushbroom Optical Hyperspectral Spectrometer Calibration and Pre-processing (L0-L1b)
    • 1. Introduction
    • 2. Calibration Objectives
      • Spectral Calibration
      • Radiometric Calibration
      • Geometric (Spatial) Calibration
    • 3. Methodology and Calibration Steps
      • Pre-Calibration Alignment
        • Mechanical Alignment
        • Dark Signal and Offset
      • Spectral Calibration
        • Wavelength Registration
        • Smile and Keystone
        • Stability Tests
      • Radiometric Calibration
        • Absolute Radiance Calibration
        • Linearity and Noise
        • Stray Light Assessment
      • Geometric Calibration
        • PSF Mapping
        • Field-of-View Verification
        • Smile and Keystone Validation
    • 4. Data Analysis and Correction
      • Key Data Products
      • Uncertainty Estimation
      • Validation
  • Pushbroom Multispectral High-Resolution Camera Calibration and Pre-processing (L0-L2)
    • Overview
      • Level 0 Data
      • Level 1 Radiometric Corrections
      • Band Co-registration
      • Georeferencing
      • Level 2 Processing
      • Additional Features
    • Requirements and Development Framework
    • Project Structure
  • On-Orbit Radiometric, Spatial, and Geometric Calibration
    • Non Uniformity Correction (NUC)
    • Bad Pixel Correction
    • Denoising
    • Image Restoration
      • MTF Calculation and PSF Sharpening
    • Band Registration
    • Georeferencing Process
    • Top of Atmospheric (TOA) Conversion
    • Atmospheric Correction
    • Mathematical Modeling and SNR Simulation

• $\textcolor{red}{\text{Remote Sensing Software Engineer}}$ ↗

  • $\textcolor{yellow}{\text{Tools and Skills}}$ ↗
  • Experience with ESA, USGS, and NASA Products
  • Remote Sensing Applications
  • Sentinel 5P TROPOMI ATBD
  • Tool Development on Qt Designer
  • Sentinel-2 and Landsat Constellation Processing
  • ASTER Image Processing
  • Image Processing with Deep Learning

• $\textcolor{red}{\text{Goofing Around Berlin}}$ 🍺 ↗

*I am a Satellite Image Processing and CalVal Engineer, as well as a Remote Sensing Engineer.

I specialize in developing and implementing algorithms for the calibration and validation of pushbroom multispectral and hyperspectral (SWIR) spaceborne / airborne optical instruments and preprocessing between Level 0 - Level 2 dataset.

I have experience with $\textcolor{yellow}{\text{cassegrain telescope}}$ and $\textcolor{yellow}{\text{hyperspectral diffractive telescope}}$, $\textcolor{yellow}{\text{hyperspectral (SWIR)}}$, $\textcolor{yellow}{\text{Thermal and multispectral sensors up to VNIR}}$ for remote sensing applications.

So far, I've participated in 11 satellite missions, focusing on multispectral and hyperspectral imaging systems, covering design, manufacturing, testing, launch, on-orbit calibration/validation, and ground segment software development.

My interdisciplinary experience extends to AOCS, Software Engineering, System Engineering, Optical Design, Optoelectronics, and Mission Planning.

• $\textcolor{red}{\text{Note:}}$ $\textcolor{red}{\text{This document has been prepared using publicly available information}}$ $\textcolor{red}{\text{from the literature and reflects concepts I have learned during my}}$ $\textcolor{red}{\text{professional experience. It does not include any proprietary or confidential}}$ $\textcolor{red}{\text{information.}}$

• I perform laboratory calibration of optical hyperspectral spectrometers for spaceborne applications. Following Baumgartner's calibration framework and utilizing DLR's Optical Calibration Laboratory facilities, I ensure the instrument meets its spectral, radiometric, and geometric fidelity requirements. Through my calibration campaign, I systematically address and characterize:
  • Detector-related effects
  • Spectral response characteristics
  • Radiometric accuracy parameters
  • Geometric alignment specifications
• This comprehensive calibration process allows me to validate that the instrument will perform according to mission specifications once deployed in space.

• Spectral Calibration

  • Determine the Instrument Spectral Response Function (ISRF) as a function of wavelength and pixel position.
  • Verify ISRF Full Width at Half Maximum (FWHM) meets the target.
  • Validate spectral oversampling and a spectral sampling interval.
  • Characterize and minimize smile and keystone effects to thresholds below acceptable levels.
  • Confirm pixel spectral linearity and ISRF stability under thermal and mechanical perturbations.
  • Measure and characterize spectral resolution across the full spectral range.
  • Determine wavelength calibration accuracy and precision.
  • Evaluate spectral sampling interval and spectral resolution stability.

• Radiometric Calibration

  • Establish a radiometric reference achieving high multiplicative accuracy using multiple radiance levels.
  • Verify overall optical transmission and detector response consistency.
  • Ensure radiometric stability and zero-level offset stability.
  • Validate detector linearity up to the saturation level and stray light contributions.

• Geometric (Spatial) Calibration

  • Characterize focal length, aperture, slit geometry, and alignment.
  • Confirm spatial sampling distance at specified orbital altitude.
  • Measure and correct spatial smile and keystone effects to below the specified threshold for design.

• Pre-Calibration Alignment

  • Mechanical Alignment

    • Mount the instrument on a vibration-isolated optical table.
    • Verify alignment of optical components using a collimated reference beam.
    • Record slit orientation and grating alignment.

  • Dark Signal and Offset

    • Operate the detector in a dark and thermally controlled environment.
    • Characterize dark current and offset for reserved dark pixels.

• Spectral Calibration

  • Wavelength Registration

    • Use a monochromator to supply narrow spectral lines across the spectrometer's bandwidth.
    • Measure ISRF profiles for each pixel and extract FWHM.

  • Smile and Keystone

    • Shift the illumination spot across the Field of View (FoV) and record spectral and spatial shifts.

  • Stability Tests

    • Simulate thermal and mechanical stresses and evaluate ISRF variation.

• Radiometric Calibration

  • Absolute Radiance Calibration

    • Illuminate the entrance slit with a calibrated integrating sphere.
    • Derive gain and offset coefficients for each pixel.

  • Linearity and Noise

    • Assess signal-to-noise ratio (SNR), dark current, and readout noise.
    • Vary integration time and radiance levels to confirm detector linearity.

  • Stray Light Assessment

    • Simulate high-contrast scenes and evaluate stray light suppression.

• Geometric Calibration

  • PSF Mapping

    • Use a collimated beam and pinhole mask to map the Point Spread Function (PSF).

  • Field-of-View Verification

    • Measure FoV and ensure alignment with design specifications.

  • Smile and Keystone Validation

    • Introduce distinct spectral lines across the FoV and measure detector output to quantify smile and keystone effects.

• Key Data Products

  • Spectral calibration files (ISRF maps, wavelength alignment).
  • Radiometric calibration files (gain, offset, linearity corrections).
  • Geometric calibration files (distortion maps, smile/keystone corrections).

• Uncertainty Estimation

  • Assess calibration uncertainties based on reference standards and environmental stability.
  • Document potential sources of error, including mechanical shifts and signal noise.

• Validation

  • Apply calibration files to a test dataset and verify compliance with requirements.

• Responsible for developing calibration, validation and preprocessing pipeline ground segment software starting from Level-0 (raw) to Level-2 (science ready dataset). The pipeline includes the following information and algorithms:

  • Multispectral Demo Satellite Preprocessing Pipeline

    A demonstration preprocessing pipeline for pushbroom multispectral optical instruments (under development).
    
    

    This project implements an image preprocessing pipeline for multispectral satellite imagery, with features including:

    • Level 0 corrections including:

      • decoding
      • missing package check, flag generation

    • Level 1 corrections including:

      • Non-uniformity correction (NUC):

        • Using key data (gain,offset)

      • Dark current correction

      • Denoising:

        • Using various filters or designing digital signal filters

      • Radiometric Conversion:

        • using key data from lab or on-orbit calibration campaigns.

      • MTF Compensation / PSF deconvolution

      • Band co-registration

        • Sensor image acquisition model or key point extraction and matching

      • Georeferencing

        • Using GDAL or rasterio
        • If possible, using central pixel coordinate from Metadata file to georeference image.
        • If GNSS data is missing, downloading sentinel-2 images using TLE of the satellite and google earth engine. Thereafter, image-to-image georeferencing applied by keypoint extraction and matching.

      • Calculation geolocation accuracy, CE95

      • Orthorectification

    • Level 2:

      • Atmospheric correction:

        • using Py6S

    • Additionally:

      • Pansharpening

        • Implemented various algorithms, including Simple Brovey, Gram-Schmidt, ESRI, and Brovey.

      • Image quality Report Generation:

        • To evaluate the quality of raw images and the effectiveness of the applied correction methods, metrics such as Peak Signal-to-Noise Ratio (PSNR), RMSE, SSIM, MSE, GIQE, CE95, radiometric accuracy are calculated and reported in PDF file.
    • 
      
      

    Requirements	Testing and Development Framework
    Python and Packages:	Development Environment:
    - Python 3.x	- Docker, Kubernetes
    - NumPy	- Dask for distributed computing
    - OpenCV	Testing:
    - GDAL	- Pytest for automated testing
    - rasterio	- Unit Testing coverage
    - scikit-image	- Comprehensive logging system
    - matplotlib
    - Earth Engine API

    Project Structure

    • 02_scripts
      • Core processing scripts
        • level_0.py - Level 0 processing
        • level_1.py - Level 1 processing and corrections
        • band_coreg.py - Band co-registration
        • georeferencing_v1.py - Georeferencing
        • metrics_ips.py - Quality metrics
        • pansharp.py - Pansharpening

    License

    This project is licensed under the GNU GPL v3 - see the LICENSE file for details.

    
    
    

• I send commands to capture images of pseudo-invariant sites such as the Mauritania Desert, Dome-C, or Antarctic for the flatfield image at different TDI stages and exposure times. I follow the USGS Test Sites Catalog.

• I use images taken at night during passes over the Atlantic Ocean, ensuring there are no clouds and no light, as darkfield images.

• Non Uniformity Correction (NUC)

  • Calculate the mean of each column for the flatfield and darkfield images. Call the results for each column flatfield_desired and darkfield_desired.

  • Calculate gain and offset as:

    $$ gain = \frac{\overline{flatfield_{desired}} - \overline{darkfield_{desired}}}{flatfield_{desired} - darkfield_{desired}} $$

    $$ offset = \overline{flatfield_{desired}} - gain \cdot flatfield_{desired} $$

  • Apply non-uniformity correction and flatfielding simultaneously (NUC). A dark_offset parameter is taken from laboratory results:

    $$ NUC_{frame} = {img \cdot gain} + {offset - dark_{offset}} $$

  • Store the gain and offset data in Calibration Key Data (CKD) container.

• Bad Pixel Correction

  • Calculate the global variance of the pixels as a threshold.
  • Identify pixels exceeding the threshold.
  • Replace bad pixels with the average value of neighboring pixels.

• Denoising

  • Implement a Butterworth low-pass filter with parameters chosen by trial and error.

• Image Restoration

  • In the worst-case scenario, relying on the satellite's internal clock for image capture may introduce a 1-second timing offset, leading to a positional deviation of up to 7 km. Therefore, I use structures like take-off runways and bridges as MTF targets if I cannot capture dedicated MTF targets like Baotou.

  • MTF Calculation and PSF Sharpening

    • Identify a suitable edge (close to main scan or cross-scan axes) with sufficient contrast and low noise.
    • Construct the Edge Spread Function (ESF) from the edge.
    • Derive the Line Spread Function (LSF) by differentiating the ESF.
    • Take the normalized Fourier transform of the LSF to obtain the MTF.
    • The PSF is derived from the MTF (Fourier Transform of the PSF).
    • If the PSF is noisy or inaccurate, consider simulating a PSF model.
    • Normalize the PSF kernel and convolve the image. Alternatively, use Wiener deconvolution.

• Band Registration

  • Because of subpixel alignment issues:
    
    
    1. Convert images to 8-bit and apply CLAHE to create dummy bands.
    2. Use the selected reference band to find keypoints and descriptors of the other bands with SIFT.
    3. Match keypoints with a FLANN-based matcher.
    4. Calculate the homography matrix.
    5. Warp the bands relative to the reference band using the homography matrix.

• Georeferencing Process

  • Use Sentinel-2 bands as a reference for image-to-image georeferencing.
  • Estimate coordinates of the image scene using TLE information.
  • Download Sentinel-2 images from Google Earth Engine for those coordinates.
  • Apply the same feature matching and warping method used in band registration.
  • Copy the corner coordinates, CRS, and transform from Sentinel-2 to the T2.1 bands.

• Top of Atmospheric (TOA) Conversion

  • The result of NUC is still in Digital Numbers. Convert to radiometric units using radiometric gain/offset for each band. The Sentinel-2 TOA Reflectance equation can be applied:

    $$ Radiance_{TOA} = (NUC_{frame} - radiance_{offset}) \cdot radiance_{gain} $$

• Atmospheric Correction

  • Using Py6S for atmospheric correction
  • Familiar with MODTRAN and LibRadTran atmospheric models.
  • Used FLAASH model on ENVI for Landsat 8 OLI dataset.
  • Because of MODTRAN licensing, follow the atmospheric correction algorithm indicated in Landsat 8-9 Calibration and Validation (Cal/Val) Algorithm Description Document (ADD), page 776.
  • Work in progress on atmospheric modeling.

• Mathematical Modeling and SNR Simulation

  • Expertise in mathematical modeling of the optics and sensor integrated system to calculate total photon collected by camera depending on the satellite's attitude (roll, pitch,yaw).

• Tools & Skills

  • Proficient in Python, C and MATLAB.
  • Experienced in cloud-based deployments, implementing CI/CD, Git, Pytest, Docker, and PEP 8 compliance for robust software engineering.
  • Utilize GPU Programming, Cloud Computing, Database Management and Container-based isolated Development
  • Skilled in software like ENVI, ArcMap, Global Mapper, QGIS, and PCI Geomatica.
  • Comfortable with Linux environments, shell scripting, git / github and containerization (Docker).
  • Leverage various Python libraries indicated in the skills section (e.g., NumPy, GDAL, etc.).
  • Comprehensive understanding of the entire data processing and mission planning lifecycle, from design to deployment, ensuring high-quality data products and efficient satellite operations.

• Experience with ESA, USGS, and NASA Products

  • Sentinel-1 (SAR)
  • Sentinel-2 A/B
  • Sentinel-5P
  • Landsat 7/8/9
  • ASTER
  • MRO CTX and HiRISE

• Remote Sensing Applications

  • Metallic mineral exploration
  • Fault line detection
  • Natural disaster analysis
  • NDVI, NDWI, NBR
  • Image segmentation
  • Deep learning approaches
  • Surface deformation detection

• Sentinel 5P TROPOMI ATBD

  • Deep knowledge in Sentinel 5P TROPOMI Algorithm Theoretical Basis Document (ATBD)
  • Deep knowledge in Sentinel 2 ATBD

• Tool Development on Qt Designer

  • Developed image processing pipeline tools.
  • Created image database search and download utilities.

• Sentinel-2 and Landsat Constellation Processing

  • Processed images for:
    • Normalized Difference Vegetation Index (NDVI)
    • Normalized Difference Water Index (NDWI)
    • Normalized Burn Ratio (NBR)
    • Lineament extraction

• ASTER Image Processing

  • Leveraged ASTER's VNIR, SWIR, and TIR channels.
  • Processed ASTER data to identify minerals and mineral groups for mining applications.

• Image Processing with Deep Learning (Python, OpenCV, Keras)

  • Purpose: To detect and identify target objects in satellite imagery using both onboard and ground segment software
  • Object detection using color segmentation, template matching, corner/edge/contour detection.
  • Implemented:
    • Feature matching
    • Watershed algorithm
    • Face and cat-face recognition
    • Pedestrian detection
    • Developed various object tracking algorithms.
    • Built CNNs for real-time digit classification and object detection.
    • Utilized pyramid representation, sliding window, non-maximum suppression, and region proposals.
  • Implemented R-CNN, YOLO, and SSD for object detection.

• I just moved to Berlin and I'm practically glued to my computer chair. If anyone's bored and wants to grab a coffee or beer to chat about satellites, space, rock music, Elon Musk's wife, astrophysics, parallel universes, or Laika the dog, feel free to get in touch! 😎