Underwater Sonar Image Object Detection Data Labeling Project
Image
SegmentationThis project involved 2D bounding box annotation for underwater sonar images to support small object detection model training. The task included identifying and accurately labeling multiple target categories (e.g., cylinders, spheres, and human-shaped objects) across complex, noisy underwater acoustic backgrounds. Over 9,000 images were annotated with strict quality standards: - Followed clear annotation guidelines to ensure consistent box placement and tight alignment with target boundaries - Conducted double-checking and spot validation to reduce labeling errors - Handled challenges like low contrast, speckle noise, and small target sizes common in sonar imagery The labeled dataset was used to train a deep learning-based underwater target detection model, improving its performance on weak and small objects.