A Multiscale GAN- SOD Yolov7 Framework for Robust Small Object Detection in Complex Underwater Environment

DOI

10.2991/978-94-6239-678-4_7How to use a DOI?

Keywords

DAS; Data augmentation; YOLOv7; Multiscale GAN; Underwater Object Detection

Abstract

The detection of underwater objects is very important in such areas as the marine search, underwater surveillance and environmental monitoring. But it is a difficult task because underwater domain is complex. Where traditional object detection methods can fall short in these cases, due to problems such as differing illumination conditions, low contrast, turbidity and light scattering, obfuscation/occlusion or biological camouflage of marine life with respect to their background. These difficulties contribute to the degradation of performance of traditional algorithms. Against this background, in this paper we propose an advanced underwater object detection framework that combines a Multiscale GAN with the SOD-YOLOv7 detector. The proposed model is comprised of several main components: Adapto Denoise Block, CNN-based Dual Attention Selective (DAS) Network, synthetic data augmentation and Multi-scale Unpaired GAN. The Adapto Denoise Block efficiently diminishes noise and makes the critical aspects in the underwater photos stand out. The DAS network enhances feature extraction via spatial and channel attention mechanism. In this case, we use the multi scale GAN to generate super-resolved images which are more closer to those generated under real-world settings. We evaluate the proposed method on the publicly available UTDAC2020 dataset which includes a total 6164 images that have different resolutions, including (3840 × 2160), (1920 × 1080), (720 × 405), (704 × 576) and (586 × 480). The results show significant improvement over the classical methods, with a mAP = 97.40%, precision = 97.18% and recall = 94.85%. The model works at fast 115 FPS detection speed and about 105.4 GFLOPs computational complexity. In summary, the SOD-YOLOv7 model put forward in this paper offers a competitive way to detect small object in complex underwater.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Atul Patil
AU  - Vinay Kumar Singh
PY  - 2026
DA  - 2026/05/28
TI  - A Multiscale GAN- SOD Yolov7 Framework for Robust Small Object Detection in Complex Underwater Environment
BT  - Proceedings of the 2nd International Conference on Recent Advancement and Modernization in Sustainable Intelligent Technologies & Applications (RAMSITA-2026)
PB  - Atlantis Press
SP  - 73
EP  - 89
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-678-4_7
DO  - 10.2991/978-94-6239-678-4_7
ID  - Patil2026
ER  -