Impact of underwater image enhancement on object detection in nadir AUV imagery of goby fish
Document Type
Contribution to Book
Publication Date
6-11-2026
Department
Department of Computer Science; Department of Applied Computing; Department of Electrical and Computer Engineering
Abstract
Accurate underwater object detection is important for aquatic ecosystem monitoring, yet models trained on high-quality benchmark imagery often degrade when applied to real underwater scenes. This study examines how image enhancement affects round goby detection in nadir-view Great Lakes AUV imagery collected by the U.S. Geological Survey. We evaluate four enhancement methods, including ACDC, AutoEnhancer, Semi-UIR, and TUDA, using YOLOv12 under four train-test configurations: original-to-original (O→O), original-to-enhanced (O→E), enhanced-to-enhanced (E→E), and enhanced-to-original (E→O). Quantitative image-quality analysis shows that Enhancement improved perceptual image quality across all methods, with TUDA achieving the highest UIQM (5.31470) and UCIQE (0.57180) and ACDC the highest CCF (28.85804), compared with 1.58066, 0.40227, and 11.11308 for the original imagery. However, these gains did not consistently improve detection. The baseline O→O setting achieved a mean mAP@50 of 0.84307, while matched E→E settings remained comparable (0.83138–0.84125). By contrast, mismatched domains caused substantial degradation, with O→E dropping as low as 0.07592 and E→O as low as 0.18029, demonstrating that train-test domain consistency is more important than perceptual quality improvement alone for round goby detection. Overall, the results suggest that enhancement should not be introduced only at deployment and should instead be validated jointly with the detector under realistic operating conditions. These findings provide practical guidance for freshwater invasive-species monitoring while also underscoring the study’s single-species, detection-only design.
Publication Title
Ocean Sensing and Monitoring XVIII
Recommended Citation
Hromada, A.,
Saleem, A.,
Awad, A.,
Lucas, E.,
Paheding, S.,
&
Havens, T. C.
(2026).
Impact of underwater image enhancement on object detection in nadir AUV imagery of goby fish.
Ocean Sensing and Monitoring XVIII,
14051.
http://doi.org/10.1117/12.3094927
Retrieved from: https://digitalcommons.mtu.edu/michigantech-p2/3110