Pixel-associated autoencoder for hyperspectral anomaly detection

Pei Xiang, Shahzad Ali, Jiajia Zhang, Soon Ki Jung, Huixin Zhou

Abstract: Highlights•A novel anomaly detection framework by using correlation between pixels was built.•Two representative dictionaries were constructed by superpixel distance estimation.•The similarity metric replaced the original HSI as the input for the AE network.•A network combining two hidden-layer feature similarity constraint was proposed.