Exploring the Intrinsic Probability Distribution for Hyperspectral Anomaly Detection

In recent years, neural network-based anomaly detection methods have attracted considerable attention in the hyperspectral remote sensing domain due to their powerful reconstruction ability compared with traditional methods. However, actual probability distribution statistics hidden latent space are not discovered by exploiting error because of anomalies is explicitly modeled. To address issue, we propose a novel representation detector (PDRD) that explores intrinsic both background and for this paper. First, represent data multivariate Gaussian distributions from probabilistic perspective. Then, combine local obtained leverage spatial information. Finally, difference between test pixel average expectation pixels Chebyshev neighborhood measured computing modified Wasserstein distance acquire map. We conduct experiments on three real sets evaluate performance our proposed method. The experimental results demonstrate accuracy efficiency method state-of-the-art