Sparse Signal Processing Technologies For HyperSpectral Systems
  1. Application Scenarios
  2. Design of hyperspectral sensors and camera systems
  3. Spectral Mosaic Data Recovery
  4. Spectral Mosaic Compression
  5. Robust blind source separation
  6. Hyperspectral image (HSI) unmixing
  7. Hyperspectral image clustering
  8. Multi-label Land Cover Classification
  9. Demonstration

Design of hyperspectral sensors and camera systems

Many different techniques have been developed to enable spectral imaging. Typically a scanning-based solution is used, where at each point in time a subset of the spectral cube is sensed, requiring scanning over the remaining dimensions to fully acquire the cube. This is a major distinction compared to snapshot spectral cameras where the entire cube is acquired at one distinct point in time.
Three prototype systems (based on Ximea cameras) were developed, each one equipped with a different hyperspectral sensor. IMEC’s snapshot spectral cameras use multiple Fabry-Pérot filters to extract spectral information. The system is based on spectral filters monolithically integrated directly on top of the imaging sensor, thereby significantly avoiding the optical complexity of systems built around dispersive optics. It enables improved glare performance, a compact, low-weight and robust system and compatibility with low cost and high volume manufacturing.

The Fabry-Pérot filter concept

One linescan sensor (600-1000nm) and two different mosaic sensors (4x4 470-730nm; 5x5 600-1000nm) were developed for the design of the cameras. The linescan filter layout has a wedge design: the filters in the linescan layout are organized in small bands of a fixed height over the full width of the active area. Typically the width of the active area equals the width of the sensor. Because of the organization of the filters, in bands over the whole width of the sensor, this layout is best suited for linescan applications.
The snapshot mosaic filter layout has a per-pixel design: the filters in the snapshot mosaic layout are organized in a mosaic pattern (4x4 or 5x5). This pattern is repeated over the width and the height of the active area of the sensor. For this reason, one mosaic pattern is also called one macro pixel or sub pixel. However, this implies that the spatial resolution of an image is 4x4 or 5x5 times smaller than the full resolution of the image sensor, or that a higher spectral resolution results in a lower spatial resolution for snapshot mosaic systems. This motivates the work in exploiting the redundancy in snapshot spectral cubes to regain the higher spatial resolution, while simultaneously keeping the spectral resolution.

The sensors were calibrated to generate accurate spectral/spatial information for each of the spectral bands, and were later on integrated into cameras. Fully operational prototype systems were then developed based on these cameras, including an illumination setup, optical components for cameras and also software for hyperspectral data acquisition. These 3 different systems were made available to the partners, to perform various experiments during the remainder of the project. A training session was organized to demonstrate the usage of the camera systems and to provide essential user training.


  • P. Agrawal, K. Tack, B. Geelen, B. Masschelein, P. Aranda Moran, A. Lambrechts, and M. Jayapala, “Characterization of VNIR hyperspectral sensors with monolithically integrated optical filters,” Image Sensors and Imaging Systems 2016