Hybrid Quantum Deep Learning With Superpixel Encoding for Earth Observation Data Classification.

Area of Science:

• Earth Observation
• Quantum Computing
• Artificial Intelligence

Background:

• Earth observation (EO) data is rapidly growing, creating Big Data challenges.
• Analyzing large EO datasets with deep learning models is computationally intensive.
• Quantum computing offers potential but faces data encoding efficiency issues.

Purpose of the Study:

• To develop a hybrid quantum deep learning model for efficient EO data classification.
• To address the bottleneck of encoding large EO data into quantum states.
• To validate the model's effectiveness on benchmark EO datasets.

Main Methods:

• Introduced a hybrid quantum deep learning model.
• Implemented an efficient superpixel encoding technique for EO data.
• Evaluated the model on Overhead-MNIST, So2Sat LCZ42, and SAT-6 datasets.
• Analyzed the impact of interaction gates and measurements on performance.

Main Results:

• The hybrid model demonstrated effective encoding and analysis of EO data.
• Superpixel encoding significantly reduced quantum resource requirements.
• Accurate classification performance was achieved on multiple EO benchmarks.
• Model optimization insights were gained through gate and measurement studies.

Conclusions:

• The proposed hybrid quantum deep learning model is effective for EO data classification.
• Superpixel encoding is a viable strategy for efficient quantum data representation.
• This approach offers a promising solution for Big Data challenges in Earth observation.