Quantum-Inspired Multi-Scale Object Detection in UAV Imagery: Advancing Ultra-Small Object Accuracy and Efficiency for Real-Time Applications

Abstract

Ultra-small object detection in UAV imagery presents significant challenges due to scale variation, environmental complexity, and computational constraints. This study introduces a quantum-inspired multi-scale object detection model designed to address these issues effectively. By integrating quantum-inspired sub-pixel convolution, adversarial training, and self-supervised learning, the model enhances detection accuracy, robustness, and computational efficiency. These advancements are particularly critical for UAV applications such as surveillance, precision agriculture, disaster response, and environmental monitoring. The proposed model was evaluated on the VisDrone2019 dataset and benchmarked against state-of-the-art methods, including YOLOv4, YOLO11, RT-DETR, and EfficientDet. It achieved 65.3% precision, 52.4% recall, and a mean Average Precision (mAP) of 34.5%, outperforming conventional models in detecting ultra-small objects. Efficiency optimizations, including structured pruning and quantization, reduced computational load to 30 GFLOPS with an inference time of 8.1 milliseconds, ensuring suitability for real-time UAV applications on resource-constrained platforms. This research offers a practical and robust solution for UAV-based object detection tasks, combining state-of-the-art accuracy with operational efficiency. It also establishes a foundation for future advancements, including scalability to diverse datasets, integration with edge computing platforms, and the exploration of quantum computing techniques. These contributions pave the way for enhanced capabilities in computer vision and autonomous aerial systems.

Keywords

• Ultra-small object detection
• UAV imagery
• quantum-inspired feature pyramids
• adversarial training
• self-supervised learning
• real-time applications