Adaptive Visual Enhancement and Semi-Supervised Learning for Robust Underwater Object Detection

Abstract

Underwater object detection remains a challenging task due to the degradation of image quality caused by light absorption and scattering. To address this, we propose an innovative method combining adaptive visual enhancement and semi-supervised learning for robust underwater object detection. Our approach incorporates a visual enhancement and joint training-based image preprocessing method (VEJT-IP), which utilizes existing visual enhancement algorithms guided by object detection loss to preprocess underwater images. Additionally, we introduce a semi-supervised object detection framework utilizing high-quality pseudo-labels (HQPL-SSOD), which includes a localization quality-based pseudo-label filter to ensure the use of only high-quality pseudo-labels during training. Extensive experiments on the URPC2022 dataset demonstrate that our method significantly improves detection accuracy compared to baseline models, achieving a good balance between accuracy and annotation cost. The proposed approach improves underwater object detection performance under the evaluated URPC2022 conditions and provides a basis for further validation in broader underwater environments.