Edge Device Integration to Visualize Blue Whale Tracking Using Space-Borne Remote Sensing Data

Abstract

The global whale population is shrouded in uncertainty, primarily due to the substantial costs and resource demands associated with traditional detection methods such as sighting surveys, acoustic monitoring, and high-resolution imagery analysis. This study presents a groundbreaking approach that employs transformer-based models, specifically RTDetr with customized backbone and Segformer-based encoder–decoder architecture with skip connections, for the autonomous detection, classification, and tracking of blue whales in the Indian Ocean using space-borne satellite imagery. By integrating datasets from SASPlanet, UK Polar data, and Worldview-2 imagery around Sri Lanka, and validating with Cartosat-2E Satellite data (1.16 m) from NSIL Bangalore, ISRO. The proposed research developed a robust system capable of processing high-resolution satellite images for cost-effective whale detection. This system is accessible through Telegram and WhatsApp bots, facilitating real-time detection and tracking via deployment on a Jetson Nano board. Our model achieved impressive performance metrics, including an F1 score of 90%, mean average precision (mAP) of 83%, precision of 90%, and recall of 98%. These results demonstrate the efficacy of our approach in automating whale detection, offering a scalable and efficient tool for advancing marine conservation efforts.