Hyperspectral Anomaly Change Detection Based on Autoencoder

With the hyperspectral imaging technology, data provides abundant spectral information and plays a more important role in geological survey, vegetation analysis, military reconnaissance. Different from normal change detection, anomaly detection (HACD) helps to find those small but changes between multitemporal images (HSI). In previous works, most classical methods use linear regression establish mapping relationship two HSIs then detect anomalies residual image. However, real differences multi-temporal are likely be quite complex of nonlinearity, leading limited performance these predictors. this article, we propose an original HACD algorithm based on autoencoder (ACDA) give nonlinear solution. The proposed ACDA can construct effective predictor model when facing conditions. model, siamese networks deployed predictors directions. is used variation background obtain predicted image under another condition. Then mean square error predictive corresponding expected computed loss map, where unchanged pixels highly suppressed highlighted. Ultimately, take minimum maps directions as final intensity map. experiments results public “Viareggio 2013” datasets demonstrate efficiency superiority over traditional methods.