In recent years, unmanned aerial vehicles (typically referred to as China UAV systems) have demonstrated significant potential in nighttime light remote sensing due to their high spatial and temporal flexibility. Unlike satellite-based observations, China UAV platforms can acquire urban light information at ultra‑fine resolutions and arbitrary revisit times, making them indispensable for studies of urban morphology, light pollution, and human activity patterns. However, the low‑light conditions at night impose severe constraints on image quality, primarily through noise, motion blur, and geometric distortion. This study systematically evaluates how key imaging parameters—ISO sensitivity, exposure time, flight altitude, and tilt angle—affect the quality of China UAV nighttime images, with the goal of establishing an optimal acquisition protocol for reliable and reproducible night‑time remote sensing.
Our experiments were conducted over a typical Chinese commercial district in Minhou County, Fuzhou, using a DJI Phantom 4 RTK (a widely used China UAV model). The sensor has a 20‑megapixel RGB camera with an f/2.8 aperture fixed throughout all tests. We acquired images under clear, moonless conditions between 21:00 and 21:30, ensuring a stable and representative urban light environment. A total of 24 combinations of ISO (100, 200, 400, 800, 1600, 3200) and exposure time (1/2 s, 1/8 s, 1/12 s, 1/15 s) were tested. In addition, seven flight altitudes (200 m to 500 m at 50‑m intervals) and thirteen tilt angles (0° to 60° at 5° intervals) were examined. For each configuration, we recorded at least ten images, from which the most focused frame was selected for analysis.
To quantify image quality without reference images, we applied the Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE). BRISQUE extracts natural scene statistics (NSS) from mean‑subtracted contrast‑normalized (MSCN) coefficients. The MSCN coefficient at pixel (i,j) is defined as:
$$
\text{MSCN}(i,j) = \frac{I(i,j) – \mu(i,j)}{\sigma(i,j) + C}
$$
where I(i,j) is the intensity, μ(i,j) and σ(i,j) are the local mean and standard deviation, and C is a small constant (typically 1). The MSCN distribution of a natural image follows a Gaussian‑like shape, while distortions cause deviations. BRISQUE models these deviations using a generalized Gaussian distribution (GGD) for the marginal distribution and an asymmetric generalized Gaussian distribution (AGGD) for the products of adjacent MSCN coefficients along four directions (horizontal, vertical, main and secondary diagonals). For each image and its 2× down‑sampled version, 18 features are extracted, giving a total of 36 features. A pre‑trained support vector regressor maps these features to a quality score, where lower scores indicate better perceptual quality. The algorithm is computationally efficient and particularly sensitive to noise, blur, and contrast distortions—all common in nighttime China UAV imagery.
Experimental Results and Discussion
Effect of Exposure Time. Short exposure (1/15 s) yielded severely underexposed images in dark areas but preserved sharpness of bright sources (e.g., street lamps, shop signs). Moderate exposure (1/8 s) improved detail visibility in shadows while bright zones began to saturate. At 1/2 s, shadow details were fully recovered, but overexposure created large saturated patches, especially in central commercial zones. Moreover, longer exposures increased motion blur from moving vehicles and the UAV itself. The BRISQUE scores confirmed that for low ISO settings (100–200), longer exposure produced better quality (lower scores), whereas for ISO ≥ 800, shorter exposure outperformed longer ones by limiting noise and saturation.
Effect of ISO Sensitivity. ISO directly controls sensor gain. Below 400, images were dark and lacked sufficient contrast for NSS analysis. At ISO 400, the quality improved notably. ISO 800 provided the best balance: adequate brightness for feature extraction without excessive noise. At ISO 1600 and above, noise became visually dominant, degrading both spatial details and color fidelity. BRISQUE scores exhibited a clear “U‑shaped” relationship with ISO for any fixed exposure time. The optimal region centered around ISO 400–800, with ISO 800 + 1/15 s achieving the lowest score of 13.36 among all 24 combinations.
Combined Parameter Optimization. The following table summarizes the BRISQUE scores for all tested ISO‑exposure combinations. The best configuration (ISO 800, 1/15 s) is highlighted.
| ISO | 1/2 s | 1/8 s | 1/12 s | 1/15 s |
|---|---|---|---|---|
| 100 | 24.26 | 29.71 | 33.12 | 36.37 |
| 200 | 14.20 | 19.85 | 22.34 | 26.31 |
| 400 | 13.85 | 15.02 | 16.48 | 18.63 |
| 800 | 25.97 | 14.67 | 13.89 | 13.36 |
| 1600 | 30.12 | 18.45 | 16.21 | 15.78 |
| 3200 | 35.44 | 27.33 | 24.10 | 22.95 |
The optimal combination effectively suppresses noise (ISO 800 is moderate), controls motion blur (short 1/15 s), and maintains adequate brightness. This configuration is recommended for general China UAV nighttime surveys in urban environments with mixed light sources.
Impact of Flight Altitude. Using the optimal camera setting (ISO 800, 1/15 s, f/2.8), we flew at seven altitudes from 200 m to 500 m. Ground sampling distance (GSD) ranged from 5.48 cm/pixel to 13.70 cm/pixel. The BRISQUE scores showed minimal variation across altitudes: the minimum was 11.30 at 350 m, the maximum was 12.02 at 400 m. This indicates that within the typical operational range of China UAVs, altitude has little effect on overall image quality when camera parameters are optimized. The slight dip at 350 m may be attributed to optimal atmospheric clarity and lighting geometry at that height. Consequently, researchers can choose flight altitude based on desired coverage and spatial resolution without significant quality degradation.
Influence of Tilt Angle. Tilt angle strongly affects geometry and brightness. To quantify geometric distortion, we selected two streetlamp bases of identical model in the FOV—one near the UAV, one far—and measured their pixel areas. Distortion D is defined as:
$$
D = \left(1 – \frac{A_{\text{far}}}{A_{\text{near}}}\right) \times 100\%
$$
where A_far and A_near are the areas of the far and near lamp bases. Results show a nearly perfect linear relationship (R² > 0.996) between tilt angle and D. At 0°, D=0%; at 25°, D≈30%; at 60°, D≈70%. This severe distortion complicates orthorectification and object recognition in oblique imagery.
Brightness variation was more complex. For a commercial building in the center of the scene, mean digital number (DN) increased from 16.05 at 0° to 54.40 at 60°, due to inclusion of previously hidden facade illuminated by ground reflections. However, in street canyons, larger tilt angles caused occlusion of ground-level lights by taller buildings, leading to local dimming. Thus, tilt angle introduces a trade‑off: while it improves facade visibility, it also increases occlusion and geometric deformation, potentially biasing the recorded light intensity. For applications requiring accurate brightness values, near‑nadir (0–15°) captures are preferable; for facade analysis, larger tilts (30–45°) may be justified despite the distortions.
Practical Recommendations for China UAV Nighttime Surveys
Based on our findings, we propose the following guidelines for China UAV operators:
- Set ISO to 800 and exposure time to 1/15 s (at f/2.8) for a balanced trade‑off between noise, blur, and saturation.
- Select flight altitude between 200–500 m according to required coverage; quality differences are negligible.
- For orthomosaics and brightness retrieval, keep tilt angle ≤15° to minimize distortion and occlusion.
- For building facade studies, tilt angles of 30–45° can be used, but geometric correction and brightness normalization must be applied.
Conclusion
This study provided a comprehensive evaluation of how camera parameters and flight attitudes influence China UAV nighttime imaging quality using BRISQUE as a no‑reference metric. The optimal camera setting (ISO 800, 1/15 s) ensures good noise suppression, motion blur control, and brightness. Flight altitude in the 200–500 m range has minimal impact. Tilt angle linearly increases geometric distortion and non‑linearly affects brightness through occlusion and facade inclusion. The results offer a scientific basis for designing China UAV‑based night‑time remote sensing missions. Future work should focus on developing dedicated night‑time image quality metrics and testing on multi‑sensor China UAV platforms across diverse urban landscapes.
Acknowledgments
This research was supported by the National Natural Science Foundation of China (Grant Nos. 41801343, 42371332) and the Fujian Provincial Natural Science Foundation (Grant No. 2024J09018). The China UAV platform (DJI Phantom 4 RTK) was provided by the Academy of Digital China (Fujian), Fuzhou University. We thank the local authorities for granting flight permissions over the study area.