An Embedded AI Perception Framework for Autonomous Long-Range Loitering Munition Strike Systems Operating in Cluttered and GNSS-Denied Environments

Autonomous long-range loitering munition systems require robust onboard perception architectures capable of detecting, classifying, and discriminating tactical targets under cluttered battlefield environments while operating within strict computational and power constraints. This paper presents the design and implementation framework of an embedded artificial-intelligence-enabled vision pipeline for real-time onboard target detection in enduranceclass autonomous strike UAV platforms. The proposed architecture integrates lightweight deep learning objectdetection networks, multi-sensor preprocessing, confidence-driven discrimination logic, and GPU-accelerated edge-inference modules to enable reliable perception in GNSS-degraded and communication-limited operational environments. Simulation-based evaluation demonstrates that optimised convolutional neural-network detectors achieve inference latency below 120 ms with detection precision exceeding 90% under cluttered terrain conditions. The framework establishes a scalable baseline for indigenous development of autonomous loitering munition strike perception systems supporting ISR-strike convergence architectures.