IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Hybrid Quantum-Classical Convolutional Neural Network With Attention-Based Feature Fusion for Remote Sensing Image Classification
Guoqing Zhou, R A N Xu, Jianrong Xiao
Abstract
Convolutional neural networks (CNNs) effectively capture spatial patterns but are inherently limited by linear, local weighting when modeling complex, high-order correlations in remote sensing imagery. To address this, we propose the Quantum-Classical Synergy Network (QCSN), a hybrid architecture combining a lightweight CNN branch with a Parameterized Quantum Circuit designed to capture high-order feature interactions via quantum entanglement. A channel attention mechanism is introduced to dynamically recalibrate and fuse these heterogeneous representations based on input characteristics. Under strict parameter parity, comprehensive evaluations on EuroSAT, RSI-CB128, NaSC-TG2, SIRI-WHU, and PatternNet indicate that QCSN achieves superior performance on the majority of benchmarks, while remaining competitive on the others. Furthermore, feature space analysis reveals near-zero cosine similarity between the quantum and classical branches, quantitatively establishing their representational orthogonality. Empirical analyses verify that the attention mechanism operates dynamically per sample, effectively mitigating the spatial information loss associated with aggressive input downscaling required for near-term quantum simulation. Finally, the architecture exhibits graceful degradation under various simulated quantum noises, affirming its operational viability in the Noisy Intermediate-Scale Quantum era.