📄 中文摘要
量子机器学习（QML）正日益受到关注，被视为解决未来计算需求挑战的潜在方案。地球观测（EO）领域已进入大数据时代，而利用复杂的深度学习模型有效分析海量EO数据的计算需求已成为瓶颈。为了应对这一挑战，本研究旨在利用量子计算技术进行EO数据分类，并探索其潜在优势。通过构建混合量子网络，将经典神经网络与量子电路相结合，以期在处理EO数据分类任务时，能够充分利用量子计算在并行处理和高维特征空间映射方面的独特能力。这种混合架构允许模型同时学习多个EO分类任务，从而提高模型的泛化能力和数据利用效率。具体而言，该方法通过共享量子层或经典层中的参数，实现不同分类任务之间的知识迁移，减少了对大量标注数据的依赖。
📄 English Summary
Multitask Learning for Earth Observation Data Classification with Hybrid Quantum Network
Quantum machine learning (QML) is garnering increasing attention as a potential solution to address future computational demands. Earth observation (EO) has transitioned into the Big Data era, where the computational requirements for effectively analyzing vast EO datasets with complex deep learning models have become a significant bottleneck. Motivated by this, the present work aims to leverage quantum computing for EO data classification and explore its inherent advantages. A hybrid quantum network is constructed, integrating classical neural networks with quantum circuits, to fully exploit quantum computing's unique capabilities in parallel processing and high-dimensional feature space mapping for EO data classification tasks. This hybrid architecture enables the model to concurrently learn multiple EO classification tasks, thereby enhancing generalization capabilities and data utilization efficiency. Specifically, the approach facilitates knowledge transfer between different classification tasks by sharing parameters within either quantum or classical layers, thereby reducing reliance on extensive labeled data. The quantum layers are designed to capture intricate non-linear relationships within the data, while the classical layers handle traditional feature extraction and classification logic. Experimental results demonstrate that the hybrid quantum network effectively improves classification accuracy for EO image classification tasks, particularly under conditions of limited data or complex features. Furthermore, the multitask learning paradigm enhances the model's robustness and adaptability, enabling it to better cope with the diversity and complexity of EO data, thus offering a novel computational paradigm and solution for large-scale data analysis in the Earth observation domain.