Researchers have developed MITE-Net, a novel architecture designed for real-time tiny target perception in 4K video for embodied edge synthetic aperture radar (SAR) missions. This system optimizes for Size, Weight, and Power (SWaP) constraints, addressing the feature loss from downsampling and latency issues of slice-based processing. MITE-Net integrates a learning-free motion-based region proposal network with a small R-CNN-like head, and is evaluated on newly constructed SAR-Tiny datasets. On an NVIDIA Jetson AGX Xavier, it achieved a 100% search success rate at 30.33 FPS and demonstrated superior energy efficiency compared to YOLO models. AI
IMPACT This research offers a specialized solution for edge AI in SAR missions, improving efficiency and accuracy for tiny target detection.
RANK_REASON This is a research paper detailing a novel architecture and datasets for a specific computer vision task. [lever_c_demoted from research: ic=1 ai=1.0]
- MITE-Net
- NVIDIA Jetson AGX Xavier
- R-CNN
- SAR-Tiny Datasets
- SeaDroneSee-Tiny
- Swap
- synthetic aperture radar
- TTM-RPN
- UAVID-Tiny
- YOLO
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
