Hybrid Quantum-Classical Convolutional Neural Network Model for Image Classification.

Area of Science:

• Computer Science
• Quantum Computing
• Remote Sensing

Background:

• Earth observation (EO) data analysis faces computational bottlenecks due to big data challenges.
• Sophisticated machine learning models require significant computational power for remote sensing image classification.
• Quantum computing offers potential solutions to overcome these computational limitations.

Purpose of the Study:

• To introduce a hybrid quantum-classical convolutional neural network (QC-CNN) for efficient feature extraction in EO data classification.
• To leverage quantum properties for accelerating the analysis of large-scale remote sensing datasets.
• To reduce quantum bit resource requirements using amplitude encoding.

Main Methods:

• Development of a hybrid quantum-classical convolutional neural network (QC-CNN).
• Implementation of amplitude encoding to minimize qubit usage.
• Complexity analysis to compare computational speed with classical CNNs.
• Evaluation using diverse EO benchmarks (Overhead-MNIST, So2Sat LCZ42, PatternNet, RSI-CB256, NaSC-TG2) on the TensorFlow Quantum platform.

Main Results:

• The proposed QC-CNN model demonstrates accelerated convolutional operations compared to classical counterparts.
• The model achieved superior performance and higher generalizability across multiple EO datasets.
• Amplitude encoding effectively reduced the necessary quantum bit resources.

Conclusions:

• The hybrid QC-CNN model is a valid and effective approach for Earth observation data classification.
• Quantum computing integration offers a promising solution for handling big data challenges in remote sensing.
• The QC-CNN approach enhances classification accuracy and generalizability in remote sensing applications.