Hybrid Quantum-Deep Learning Approach: Optimizing Land Cover Classification with GMM Outlier and Fusion Key Feature Selection

This study proposes a hybrid Quantum-Deep Learning framework to enhance the efficiency and accuracy of land cover classification.The approach integrates a Quantum Gaussian Mixture Model (QGMM) for outlier detection, a Bidirectional Gated Recurrent Unit (BiGRU) for feature extraction, and a combination of Random Forest (RF) and XGBoost (XGB) for feature selection.By leveraging quantum principles in the Expectation-Maximization (EM) algorithm, the QGMM significantly optimizes parameter estimation, enabling robust outlier detection that supports better model generalization.Experimental results on the UCI Forest Cover dataset demonstrate near-perfect classification performance, achieving an accuracy of 99.9% and an Area under the curve (AUC) of 1.0, highlighting the model's capacity to handle high-dimensional, imbalanced data effectively.This framework provides a promising solution for complex environmental datasets, paving the way for future research into integrating quantum techniques for broader ecological and geospatial applications.