AI Research & Development

Real-time AI-powered streaming analysis for offshore platform inspections via fiber optic transmission. Our advanced transfer learning models detect surface installation defects including corrosion, structural cracks, coating degradation, and mechanical damage with 90%+ accuracy. The system processes 15-30 frames per second from platform-mounted cameras, delivering instant alerts for critical issues through temporal tracking that reduces false positives by 60%. Operators receive real-time defect overlays, confidence scores, and maintenance recommendations during live inspections, dramatically reducing vessel time and enabling immediate operational decisions. Applications include topside equipment monitoring, flare tower inspections, helideck structural assessment, and safety barrier verification across Norwegian Continental Shelf installations.

Pioneering subsea inspection through real-time ROV/AUV video streaming with AI-powered defect detection. Our transfer learning framework analyzes live underwater video transmitted via fiber optic cables, identifying pipeline defects (corrosion, cracks, weld anomalies, biofouling), subsea equipment degradation, and structural integrity issues in real-time. The system integrates YOLOv8 models fine-tuned on subsea datasets with multi-sensor fusion (visual, LiDAR, sonar) for comprehensive underwater asset monitoring. Frame-by-frame inference delivers <500ms latency, enabling operators to make immediate decisions during inspection operations. Advanced temporal tracking validates defects across multiple frames, dramatically improving accuracy and reducing post-inspection analysis time from 24-48 hours to real-time. Our offline training pipeline supports continuous model improvement using customer-specific data, adapting detection capabilities to unique field conditions and asset types across deepwater installations.

Advanced computer vision and passive sonar analysis for comprehensive aquaculture monitoring. Our multi-sensory platform integrates underwater 8K camera arrays with hydroacoustic networks (10Hz-100kHz) for real-time feeding behavior recognition (92% accuracy), stress detection through acoustic signatures, and species identification. Advanced sonar systems distinguish farmed from wild salmon using swim bladder resonance patterns and behavioral fingerprinting with 95% accuracy. The system processes 10TB daily per facility through edge computing (NVIDIA Jetson Orin) and central AI clusters, providing behavioral analysis, escape risk prediction, and automated visual documentation. Applications include fish health monitoring through thermal imaging, feeding optimization via acoustic analysis, and escape prevention using behavioral conditioning. Our image and sonar-based AquaGuard platform achieved 99.5% escape prevention rates in pilot deployments across Norwegian salmon farms.