Where are the Whales: A Human-in-the-loop Detection Method for Identifying Whales in High-resolution Satellite Imagery

The paper operates within the established paradigm of using computational analysis of very high-resolution (VHR) satellite imagery for marine mammal monitoring. This paradigm seeks to overcome the limitations of traditional, resource-intensive survey methods. The specific puzzle being addressed is the bottleneck created by the lack of large, annotated datasets required to train fully automated deep learning models for whale detection.

This work is classified as Normal Science because it does not challenge the fundamental principles of its paradigm but instead focuses on puzzle-solving to improve its efficiency and scope. The authors develop a semi-automated, human-in-the-loop system that uses statistical anomaly detection to pre-filter imagery, directing expert attention to the most promising areas. This is a methodological refinement designed to solve the well-defined problem of scalable data annotation. By creating a tool to more efficiently generate training data, the research strengthens and extends the existing machine-learning-based monitoring paradigm rather than proposing a revolutionary alternative.