Shahriar Kabir

• Modular Pilot Application: Enabled compatibility with diverse drones (DJI M300 & MAVLink). • Real-time Video Streaming: Sub-200ms latency for smooth video feeds (FFmpeg & GPU acceleration). • Localized Geofence Database: Optimized on-device geofence retrieval (SQLite). • Scalable Geofence Storage (AWS S3): Efficient database management. • Dockerized Gazebo Simulator: Streamlined drone mission simulation. • Custom Python Image Viewer: Facilitated wind turbine blade damage inspections.

• Implemented self-supervised contrastive learning in PyTorch to pre-train a model on the Global Mining Sites dataset, reducing reliance on expensive annotated data. • The pre-trained model improved mining site segmentation with minimal labeled data. • Collaborated with researchers across TUM AI and DLR to advance this state-of-the-art approach and develop effective solutions.

Master Thesis Student Jan 2022 - Jun 2022 • Created a Sentinel-2 image-based dataset of 150 annotated Chilean mining sites, with regions classified into eight categories (e.g., mine sites, open-pits, leaching heaps). • Implemented a U-Net model with a ResNet-34 backbone in TensorFlow/Keras for semantic segmentation. • The model achieved 89% accuracy and a mean Intersection over Union (mIoU) of 70.5%, demonstrating its effectiveness in classifying various mining site features.