Raw Imagery Impacting Your AI: Should You Care?

Adrien Dorise; Marjorie Bellizzi; Stéphane May

eess.IVSep 29, 2026

Raw Imagery Impacting Your AI: Should You Care?

Authors: Adrien Dorise, Marjorie Bellizzi, Stéphane May

Organizations: CNES, Toulouse, France · IRT Saint-Exupéry, Toulouse, France

Abstract

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.

Figures & tables

Panel of Figure 1 — Fig. 1 : Examples of GSD, MTF, SNR, and combined degradations at GSDs of 100 and 200 cm.

Fig. 2 : Mean AP 50 gain from GSD 200cm to GSD 100cm (in percentage points)

Panel of Figure 3 — Fig. 3 : Detection performance at GSD = 100 cm. Left panels show the tested configurations and right panels show interpolated AP 50 surfaces for visualization.

Panel of Figure 4 — Fig. 4 : Detection performance at GSD = 200 cm. Left panels show the tested configurations and right panels show interpolated AP 50 surfaces for visualization.

Degradation	Experiment	$\mathrm{MTF}_{\mathrm{Nyq}}$	SNR $(L_{0},L_{1})$ [dB]	NanoDet RangiLyu (2021)			YOLOv5s Ultralytics (2021)			YOLOX-S Ge et al. (2021)
Degradation	Experiment	$\mathrm{MTF}_{\mathrm{Nyq}}$	SNR $(L_{0},L_{1})$ [dB]	AP	AP 50	AP 75	AP	AP 50	AP 75	AP	AP 50	AP 75
MTF only	mtf_025_gsd100	0.250	(100, 250)	0.244	0.305	0.267	0.345	0.412	0.383	0.281	0.342	0.318
MTF only	mtf_015_gsd100	0.150	(100, 250)	0.222	0.276	0.241	0.298	0.370	0.331	0.278	0.338	0.312
MTF only	mtf_010_gsd100	0.100	(100, 250)	0.221	0.271	0.239	0.317	0.383	0.352	0.263	0.336	0.298
MTF only	mtf_007_gsd100	0.070	(100, 250)	0.218	0.265	0.235	0.298	0.364	0.331	0.268	0.333	0.303
MTF only	mtf_003_gsd100	0.030	(100, 250)	0.215	0.264	0.231	0.272	0.336	0.302	0.238	0.297	0.270

TABLE I : Detection performance at GSD = 100 cm.

Degradation	Experiment	$\mathrm{MTF}_{\mathrm{Nyq}}$	SNR $(L_{0},L_{1})$ [dB]	NanoDet RangiLyu (2021)			YOLOv5s Ultralytics (2021)			YOLOX-S Ge et al. (2021)
Degradation	Experiment	$\mathrm{MTF}_{\mathrm{Nyq}}$	SNR $(L_{0},L_{1})$ [dB]	AP	AP 50	AP 75	AP	AP 50	AP 75	AP	AP 50	AP 75
MTF only	mtf_025_gsd200	0.250	(100, 250)	0.176	0.220	0.190	0.248	0.311	0.274	0.210	0.267	0.239
MTF only	mtf_015_gsd200	0.150	(100, 250)	0.166	0.206	0.178	0.246	0.298	0.267	0.211	0.278	0.241
MTF only	mtf_010_gsd200	0.100	(100, 250)	0.198	0.246	0.216	0.256	0.320	0.282	0.236	0.301	0.266
MTF only	mtf_007_gsd200	0.070	(100, 250)	0.179	0.229	0.196	0.242	0.299	0.260	0.229	0.296	0.260
MTF only	mtf_003_gsd200	0.030	(100, 250)	0.184	0.231	0.196	0.245	0.302	0.268	0.209	0.279	0.239

TABLE II : Detection performance at GSD = 200 cm.

Explore similar work

Impact of Atmospheric Turbulence and Pointing Error on Earth Observation
May 21, 2026Celia Sánchez-de-Miguel, Antonio M. Mercado-Martínez, Beatriz Soret +2Earth Observation Ship Detection
TELMA, University of Malaga Malaga, Spain
A Causally Grounded Taxonomy for Image Degradation Robustness Evaluation
May 15, 2026Stefan Becker, Simon Weiss, Wolfgang Hübner +1Visual Degradation Degradation
Fraunhofer IOSB Ettlingen, Germany
GlobalForge: Towards Robust AI-Generated Image Detection
Jul 16, 2026Manni Cui, Ruiqi Liu, Dianyuan Zou +8Ai-Generated Image Detection Degradation
1Huazhong University of Science and Technology · Institute of Automation, Chinese Academy of Sciences · 3Jilin University +1

May 21, 2026, cs.NIJ/K move · Enter open · S save

Impact of Atmospheric Turbulence and Pointing Error on Earth Observation

Celia Sánchez-de-Miguel, Antonio M. Mercado-Martínez, Beatriz Soret, Antonio Jurado-Navas+1

TELMA, University of Malaga Malaga, Spain

Earth Observation Ship Detection Global Navigation Satellite System Impact Analysis

Open paper AskPDF arXiv

Earth Observation (EO) imagery is often degraded by atmospheric turbulence and pointing jitter; yet, these effects are rarely considered in datasets used to train AI-based detection models. Based on prior work, this paper presents an enhanced image simulator that enables the incorporation of vertical-path atmospheric turbulence and satellite pointing jitter, arising from platform and sensor vibrations, to generate physically realistic distorted images. As a case study, vessel detection is evaluated using YOLOv8 and RetinaNet on images generated by the proposed simulator under different levels of turbulence and pointing errors. Results show that YOLOv8 recall decreases from 91% under ideal conditions to 60% in the presence of weak turbulence, and falls below 40% under strong turbulence or jitter. In contrast, RetinaNet demonstrates greater robustness, maintaining approximately 75% recall across degraded conditions. These results highlight the importance of incorporating realistic physical degradations into EO training datasets to ensure reliable performance of AI-based models in operational environments, as demonstrated in maritime surveillance applications.

Abstract

Explore similar work

Impact of Atmospheric Turbulence and Pointing Error on Earth Observation

A Causally Grounded Taxonomy for Image Degradation Robustness Evaluation

GlobalForge: Towards Robust AI-Generated Image Detection