TL;DR
This paper presents a feasibility study for SISSI, a novel Earth observation imager for the medium infrared (MIR) spectrum. By combining compressive sensing (CS) with super-resolution techniques, SISSI aims to achieve higher spatial resolution and better performance in monitoring high-temperature events, while also reducing payload mass and power consumption.
What Problem Does This Paper Address?
Monitoring Earth's surface in the Mid-Infrared (MIR) spectrum is crucial for various applications, including agriculture, meteorology, and hazard assessment (like volcanic activity). However, existing MIR imagers often struggle with achieving high spatial resolution, which limits their ability to accurately characterize phenomena like high-temperature events (HTEs). Furthermore, HTEs can cause saturation and blooming effects in detectors, further degrading image quality. The SISSI project tackles these challenges by proposing a new payload architecture that leverages advanced imaging techniques to overcome these limitations.
Methods
The core of the SISSI instrument is a Spatial Light Modulator (SLM), which is an array of electronically actuated micromirrors. This SLM acts as a spatial encoder. The instrument operates on the principle of Compressive Sensing (CS), which allows for acquiring data with fewer measurements than traditional methods. In essence, the SLM applies different binary patterns to modulate the incoming light, and a single-pixel detector captures the integrated signal. This "compressed acquisition" eliminates the need for a separate onboard compression board, saving mass, memory, and energy.
The "super-resolution" aspect comes from the fact that each detector pixel effectively integrates light from a larger area on the ground, which is further subdivided by the micromirrors on the SLM. The instrument aims for a 4x4 super-resolution factor, meaning the final reconstructed image has four times the resolution in each dimension compared to the detector's native pixel grid.
For reconstructing the super-resolved images from the compressed measurements, the study explored two categories of algorithms:
- Traditional: Total Variation (TV) pseudo-norm.
- Deep Learning: A convolutional neural network called ISTA-Net (and its improved version, ISTA-Net+).
Datasets
The paper mentions that reconstruction algorithms were tested using an image acquired by the ASI-PRISMA hyperspectral instrument over Trasimeno Lake (Italy). This serves as a benchmark for evaluating the reconstruction performance of the proposed CS approach. No new datasets were introduced in this paper.
Key Insights and Ideas
- CS for MIR Imaging: The SISSI project pioneers the application of CS and super-resolution techniques specifically for the MIR spectral range, which presents unique challenges.
- SLM as a Key Component: The use of an SLM as a programmable spatial encoder is central to the CS approach, enabling flexible and efficient data acquisition.
- Payload Simplification: By integrating compression into the acquisition process via CS, the payload design is simplified, leading to reduced mass, memory, and power requirements. This is a significant advantage for space-based instruments.
- Deep Learning for Reconstruction: The study highlights the potential of deep learning algorithms (like ISTA-Net) to achieve better reconstruction results compared to traditional CS methods, although it notes the need for suitable training data.
- Improved HTE Monitoring: The combination of super-resolution and CS is expected to significantly improve the monitoring of high-temperature events by providing higher spatial detail and mitigating saturation issues.
Key Results
The paper presents a feasibility study and demonstrates the concept through image reconstruction tests. A key qualitative result is shown in Figure 3, where an image acquired by ASI-PRISMA is reconstructed using the SISSI approach. The reconstruction of a macropixel (4x4 micromirrors) shows good quality, and a further increase in resolution (32x32 micropixels) is demonstrated. While quantitative metrics like accuracy percentages are not explicitly detailed for the reconstruction, the visual results suggest the feasibility of achieving super-resolution. The paper also lists the main characteristics of the SISSI payload, including its operational spectral range (3-5 µm), number of spectral bands (5), and a target super-resolution factor of 4x4.
Limitations and Future Work
- Deep Learning Training Data: A potential drawback of deep learning reconstruction methods is the requirement for a training dataset that closely matches the characteristics of the images to be acquired by SISSI. This suggests a need for further work in generating or acquiring relevant training data.
- Ground Segment Complexity: While the onboard payload is simplified, the computational burden for image reconstruction is shifted to the ground segment, which requires more processing power.
- Future Applications: The authors suggest that the SISSI payload could contribute significantly to applications within the ESA Living Planet Program, particularly for monitoring atmospheric gases (CH4, N2O, CO2) and high-temperature events. Further development and validation of the reconstruction algorithms and the full payload system are implied future steps.
Summary updated