Hyperspectral Anomaly Detection Using Quantum Potential Clustering

Quantum machine learning has attracted significant attention in recent years due to its capacity to reflect a particle's aggregation in the quantum domain. In this work, quantum machine learning is used to perform anomaly detection in hyperspectral images (HSIs) by considering pixels as particles in the quantum domain, and using the quantum potential to cluster the anomalies and background. Specifically, a new anomaly detection method is proposed based on quantum potential by first reformulating and applying quantum potential concepts to traditional HSI anomaly detection. Then, by using the weight matrix (also called distance matrix), the original wave function is modified to improve the anomaly detection accuracy. Finally, by considering the similarities between anomaly detection and human vision in the task of highlighting targets, an unsupervised salient strategy is also defined to achieve the final detection results. Through this method, we build the internal relationship between quantum potential energy clustering and hyperspectral anomaly detection (HAD). Our experimental results on four real HSIs reveal that the proposed method compares favorably with regards to other anomaly detection methods. Specially, the proposed method called quantum potential anomaly detection (QPAD) obtains a 0.2% increase on the average of the background suppression ability indicator.