Title: Underwater Object Detection Using Yolov12 Model
Authors: Daniely Mapepele, Chanda Mfilinge, Angel Msonge, Sara Mulungu, Paskalina Michaely, Samwel Mwaijonga, Rayson Komba, Rahabu Harrison, Pius Lumbanga
Volume: 10
Issue: 4
Pages: 65-71
Publication Date: 2026/04/28
Abstract:
Underwater object detection plays a crucial role in marine biodiversity monitoring, infrastructure inspection, autonomous underwater vehicles, and search-and-rescue missions, but it is severely hindered by visibility degradation caused by light absorption, scattering, color distortion, low contrast, and turbidity. These factors significantly impair the performance of advanced object detectors like YOLOv12, which is primarily trained on clear terrestrial images. This study proposes an enhanced underwater object detection system by integrating a lightweight adaptive visibility enhancement module into the YOLOv12 pipeline. The module dynamically assesses image degradation and adapts enhancement parameters, incorporating a Visibility Estimation Module , Visibility-Conditioned Area Attention , and adaptive gating in the feature pyramid neck to improve feature extraction and robustness while preserving real-time performance. The main objective is to design, implement, and evaluate this visibility-aware framework using PyTorch and publicly available underwater datasets, aiming for superior mean Average Precision, precision, and recall compared to standard YOLOv12 and static enhancement approaches. The proposed system is expected to deliver consistent improvements in detecting small, occluded, or distorted objects, advancing underwater computer vision and supporting practical applications in marine conservation and environmental monitoring.