Mathematical modeling and feature extraction architecture of a quantum-inspired attention network for UAV image classification

Abstract

In this study, a novel Quantum Self-Attention Neural Network for Multi-Model Aerial Image Classification (QSANN-MAIC) is proposed. The objective of this paper is to develop a high-performance learning framework capable of accurately classifying UAV-acquired aerial images. Initially, the proposed QSANN-MAIC model pre-processes the input images through several enhancement steps, including noise reduction, sharpening, contrast enhancement, and color correction, to eliminate unwanted distortions and improve image clarity for further analysis. We have used a multi-model feature extraction framework to obtain rich and complementary feature representations, integrating three architectures: a compact Vision Transformer, an enhanced ConvNeXt model, and a fine-tuned VGG16 network. Subsequently, a quantum self-attention neural network is utilized to perform the final classification by effectively capturing long-range dependencies among the extracted features. To validate the effectiveness of the proposed QSANN-MAIC model, extensive simulations are conducted and evaluated using multiple performance metrics. Comparative analysis demonstrates that the QSANN-MAIC approach achieves improved performance across several evaluation measures.