RSOD: Reliability-Guided Sonar Image Object Detection with Extremely Limited Labels

TL;DR

RSOD tackles sonar object detection under extreme label scarcity by a reliability-guided teacher–student framework that evaluates pseudo-labels across multiple views, balances learning with object-mixed pseudo-labels, and applies reliability-weighted constraints. It introduces a multi-view pseudo-label reliability score, an object-mixed pseudo-label strategy, and a corner-point regression loss to suppress noisy labels and improve localization, achieving strong results on UATD and the newly collected FSOD dataset with only a small fraction of labeled data. The approach yields notable improvements in small-object detection and surpasses several SSOD baselines, highlighting the practical impact for underwater detection where annotation is costly. A new FSOD dataset is provided to foster further research in sonar imagery.

Abstract

Object detection in sonar images is a key technology in underwater detection systems. Compared to natural images, sonar images contain fewer texture details and are more susceptible to noise, making it difficult for non-experts to distinguish subtle differences between classes. This leads to their inability to provide precise annotation data for sonar images. Therefore, designing effective object detection methods for sonar images with extremely limited labels is particularly important. To address this, we propose a teacher-student framework called RSOD, which aims to fully learn the characteristics of sonar images and develop a pseudo-label strategy suitable for these images to mitigate the impact of limited labels. First, RSOD calculates a reliability score by assessing the consistency of the teacher's predictions across different views. To leverage this score, we introduce an object mixed pseudo-label method to tackle the shortage of labeled data in sonar images. Finally, we optimize the performance of the student by implementing a reliability-guided adaptive constraint. By taking full advantage of unlabeled data, the student can perform well even in situations with extremely limited labels. Notably, on the UATD dataset, our method, using only 5% of labeled data, achieves results that can compete against those of our baseline algorithm trained on 100% labeled data. We also collected a new dataset to provide more valuable data for research in the field of sonar.

Figures (6)

• Figure 1: Compared to previous approaches for training sonar image object detection detectors, our method reduces the requirement for labeled data.
• Figure 2: (a) The collection process of the FSOD dataset. (b) The number in each category of the FSOD dataset.
• Figure 3: The overall architecture of RSOD. (a) Illustrating the basic structure of the RSOD, (b) the process of assessing the reliability of pseudo-labels, (c) the generation process of object mixed pseudo-labels, and (d) the loss optimization process by reliability-guided adaptive constraint.
• Figure 4: Qualitative comparison of state-of-the-art object detection methods on the UATD dataset with 5% labeled data.
• Figure 5: mAP on the UATD dataset with 5% labeled data.