Remote Sensing Image Scene Classification in Hybrid Classical-Quantum Transferring CNN with Small Samples.

Area of Science:

• Computer Science
• Quantum Computing
• Remote Sensing

Background:

• Deep learning, particularly Convolutional Neural Networks (CNNs), excels in Remote Sensing Image Scene Classification (RSISC).
• Training CNNs typically requires extensive annotated data, which is often scarce in RSISC.
• Pre-trained CNNs on natural image datasets are a common workaround but may falter with remote sensing data due to differing imaging mechanisms.

Purpose of the Study:

• To propose and evaluate an improved hybrid classical-quantum transfer learning CNN for RSISC.
• To address the challenge of limited annotated data in remote sensing image analysis.
• To enhance classification accuracy while reducing model complexity and data requirements.

Main Methods:

• Developed a hybrid CNN model integrating a classical ResNet for feature extraction and a quantum tensor circuit for refinement.
• Employed transfer learning by tuning parameters on near-term quantum processors.
• Tested the hybrid model on an open-source remote sensing image dataset.

Main Results:

• The hybrid classical-quantum CNN demonstrated superior performance compared to existing pre-trained CNN-based RSISC methods, especially with small training datasets.
• Achieved improved classification accuracy.
• Significantly reduced the number of model parameters and the total training data required.

Conclusions:

• The proposed hybrid classical-quantum CNN is an effective approach for RSISC, particularly when dealing with limited annotated data.
• This method offers a promising direction for advancing remote sensing image analysis through the integration of quantum computing.
• The hybrid model provides a more efficient and accurate solution for RSISC than traditional deep learning techniques.