AI-Powered Object Detection in Satellite Imagery for Military Reconnaissance

Abstract

Timely and accurate interpretation of satellite imagery plays a vital role in modern military reconnaissance. This paper proposes VisionAI, a robust, AI-powered object detection system built on the YOLOv8 architecture, optimized for detecting military assets such as tanks, aircraft, trucks, and naval vessels. The system was fine-tuned on a custom remote sensing dataset and deployed using Google Cloud’s T4 GPU infrastructure for real-time inference. The model achieved a mean Average Precision (mAP@0.5) of 0.79 for aircraft and maintained high precision and recall across key object categories. VisionAI demonstrates strong resilience to environmental distortions including cloud occlusion, low lighting, and motion blur. This work builds upon previous efforts in lightweight detection frameworks using MobileNetV2 for pest surveillance in locust management campaigns [1], as well as scalable AI pipelines for real-time monitoring in resource-constrained settings [2]. Furthermore, it aligns with recent advances in satellite surveillance and small object detection using cross-scale and pyramid fusion methods [3], [4]. Challenges related to detecting camouflaged or low-resolution naval targets persist, underscoring the need for hybrid approaches combining multispectral data and transformer-based architectures. Ethical considerations around adversarial manipulation and dual-use of AI in military contexts are also discussed. This research offers a cost-effective, adaptable, and ethically aware solution for defense-oriented remote sensing operations in developing regions.