Satellite image restoration aims to improve image quality by compensating for degradations (e.g., noise and blur) introduced by the imaging system and acquisition conditions. As a fundamental preprocessing step, restoration directly impacts both ground-based product generation and emerging onboard AI applications. Traditional restoration pipelines based on sequential physical models are computationally intensive and slow, making them unsuitable for onboard environments. In this paper, we introduce ConvBEERS: a Convolutional Board-ready Embedded and Efficient Restoration model for Space to investigate whether a light and non-generative residual convolutional network, trained on simulated satellite data, can match or surpass a traditional ground-processing restoration pipeline across multiple operating conditions. Experiments conducted on simulated datasets and real Pleiades-HR imagery demonstrate that the proposed approach achieves competitive image quality, with a +6.9dB PSNR improvement. Evaluation on a downstream object detection task demonstrates that restoration significantly improves performance, with up to +5.1% mAP@50. In addition, successful deployment on a Xilinx Versal VCK190 FPGA validates its practical feasibility for satellite onboard processing, with a ~41x reduction in latency compared to the traditional pipeline. These results demonstrate the relevance of using lightweight CNNs to achieve competitive restoration quality while addressing real-world constraints in spaceborne systems.
Figures & tables
Figure 1: ConvBEERS architecture. It is composed of 3x3 size-preserving convolution layers (stride=1, padding=1) and 16 residual blocks without batch normalisation for a total of 1.2M trainable parameters.
Dataset
MTF
SNR @ L0 / L1
# Patches
Sim-Degraded-Variable
3–7%
50 ± 40 / 110 ± 40
196 128 / 36
Sim-Degraded-Fixed
7%
50 / 110
196 128 / 36
Sim-Reference-Fixed
25%
80 / 170
196 128 / 36
Real-Raw-Pleiades
∼ 7%
∼ 50 / ∼ 110
– / 19
Table 1: Summary of simulated and real datasets used in this work.
Method
PSNR ↑
SSIM ↑
LPIPS ↓
DISTS ↓
Traditional
20.6
0.843
0.216
0.220
ConvBEERS
27.5
0.932
0.068
0.077
Table 2: Full-reference image quality metrics on OpenAerialDataset.
Method
MTF@Nyq
SNR @ L0/L1
Traditional
N/A
65.33 / 138.51
ConvBEERS
20%
66.27 / 156.07
Table 3: Physical quality metric on OpenAerialDataset.
Figure 2: Image restoration on a simulated degraded image with embedded mire.
Deg. Level
MTF (Deg.)
MTF (Rest.)
ΔMTF
MTF 7%
7.41%
20.17%
+0.1276
MTF 5%
5.58%
15.55%
+0.0997
MTF 3%
3.36%
13.83%
+0.1047
Table 4: MTF before and after restoration with ConvBEERS for different blur levels on simulated degraded data.
Figure 3: Comparison of raw Pléiades image and restored results using the traditional algorithm and ConvBEERS.
Figure 4: Object detection evaluation pipeline . The simulated degraded and restored datasets are created from the DIOR dataset using both our sensor simulation framework and ConvBEERS. Then, three YOLOv11n models are trained independently on (1) reference images, (2) degraded images, and (3) restored images using the training and validation sets. Their performance is compared on the test sets to evaluate the impact of restoration on a detection task.
Dataset
MTF
SNR @ L0 / L1
#Patches
Deg-Med
1–2%
30 ± 5 / 70 ± 5
18 000 / 2000 / 3463
Deg-High
0.1–0.5%
10 ± 5 / 50 ± 5
18 000 / 2000 / 3463
Table 5: Summary of simulated and real DIOR datasets used in this work.
Dataset
Set
mAP50
mAP90
Prec.
Rec.
Reference
N/A
75.90%
54.10%
86.60%
69.30%
Deg-Med
Deg.
73.33%
51.5%
85.00%
66.93%
Rest.
75.52%
53.88%
87.47%
68.32%
Deg-High
Deg.
67.30%
46.29%
63.80%
60.70%
Rest.
72.37%
50.57%
85.03%
66.36%
Table 6: Object detection comparison between degraded and restored DIOR datasets with Yolo11n on 256 x 256 pixel images.
Figure 5: Qualitative comparison of restored RGB images with ConvBEERS
Model
Prec.
Size (Mb)
PSNR (dB)
SSIM
Baseline
Float32
1.97
28.37
0.600
ConvBEERS
Float32
4.78
28.58
0.887
ConvBEERS
Int8
2.01
26.77
0.848
Table 7: Quantization performance for a 13000×9000 panchromatic swath image.
Method
Latency (s) ↓
FPS ↑
Power (W) ↓
Pxls/s/W ↑
Baseline (CPU)
1724
0.2
22.5
3161
ConvBEERS (DPU)
42
7.2
29.6
98965
Table 8: Benchmark on Versal AI Edge VCK190 FPGA for a 13600×9016 panchromatic swath image. FPS are given for 640x640 patches
Onboard satellites must restore a channel-degraded image on a few watts, using neuromorphic accelerators (e.g., BrainChip Akida, Intel Loihi-2) that support no softmax or attention. We ask which encoder restores best under that constraint and introduce LIMODENet (LinearMix-ODENet), a 0.69M softmax-/QKV-free backbone whose residual stages read as ODE discretizations and which is empirically information-preserving (probe accuracy rises 79.9% -> 98.4% from stem to head). At iso-parameters it restores 1 dB DVB-S2X-degraded EuroSAT better than a CNN autoencoder (+1.75 dB PSNR) and a skip-connection U-Net (+1.07 dB), three seeds, non-overlapping. Unconstrained modern restorers (NAFNet, Restormer) win on fidelity; we decompose that gap: spiking-legal additive skips recover about half, and the rest traces to attention and channel gating. LIMODENet then converts end-to-end to a spiking network with zero blocked operations, versus 22-24 for the competitors: not the best restorer available, but the best verified deployable within a real power budget.
Thanh-Dung Le, Vu Nguyen Ha, Ti Ti Nguyen +1
Texas A&M University - Corpus Christi, TX, USA · University of Luxembourg, Kirchberg, Luxembourg
Onboard AI is gaining interest for space applications such as vessel, wildfire, and cloud detection, where real-time processing can improve mission reactivity and reduce downlink needs. However, onboard models may operate on raw or minimally processed imagery rather than on restored ground products. This study evaluates how image degradation affects object detection by varying Signal-to-Noise Ratio (SNR), Modulation Transfer Function (MTF) at Nyquist, and Ground Sampling Distance (GSD). Controlled degradations are applied to Very High Resolution Maxar imagery, and three lightweight detectors, YOLOv5s, YOLOX-S, and NanoDet, are evaluated on the resulting operating points. The results show that the impact of image quality depends on the degradation mechanism, and that increasing degradation does not necessarily lead to a proportional decrease in vessel detection performance. GSD produces the most consistent performance shift, while MTF and SNR effects depend more on the model and resolution. Severe combinations of blur and noise produce the largest losses. These results provide task-level information that can support sensor, processing, and AI trade-offs for future onboard systems.
Adrien Dorise, Marjorie Bellizzi, Stéphane May
CNES, Toulouse, France · IRT Saint-Exupéry, Toulouse, France
Image restoration aims to recover high quality images from inputs degraded by various factors, such as adverse weather, blur, or low light. While recent studies have shown remarkable progress across individual or unified restoration tasks, they still suffer from limited generalization and inefficiency when handling unknown or composite degradations. To address these limitations, we propose RAR, a Restore, Assess and Repeat process, that integrates Image Quality Assessment (IQA) and Image Restoration (IR) into a unified framework to iteratively and efficiently achieve high quality image restoration. Specifically, we introduce a restoration process that operates entirely in the latent domain to jointly perform degradation identification, image restoration, and quality verification. The resulting model is fully trainable end to end and allows for an all-in-one assess and restore approach that dynamically adapts the restoration process. Also, the tight integration of IQA and IR into a unified model minimizes the latency and information loss that typically arises from keeping the two modules disjoint, (e.g. during image and/or text decoding). Extensive experiments show that our approach consistent improvements under single, unknown and composite degradations, thereby establishing a new state-of-the-art.
I-Hsiang Chen, Isma Hadji, Enrique Sanchez +5
Samsung AI Center Cambridge · National Taiwan University · Technical University of Iasi +3