Deep anomaly detection in hyperspectral images based on membership maps and object area filtering

作者：

Highlights：

• A framework is developed to detect anomalies in hyperspectral images accurately.

• First, principal component analysis (PCA) is applied to reduce dimensionality.

• A convolutional network is then proposed to learn the pixel pair similarities.

• A novel method entitled Object Area Filtering is used to remove undesired objects.

• The results are compared to other anomaly detection methods in time and accuracy.

摘要

•A framework is developed to detect anomalies in hyperspectral images accurately.•First, principal component analysis (PCA) is applied to reduce dimensionality.•A convolutional network is then proposed to learn the pixel pair similarities.•A novel method entitled Object Area Filtering is used to remove undesired objects.•The results are compared to other anomaly detection methods in time and accuracy.

论文关键词：Anomaly detection,Convolutional neural network,Hyperspectral image,Object area filtering

论文评审过程：Received 25 February 2021, Revised 5 November 2021, Accepted 5 November 2021, Available online 2 December 2021, Version of Record 7 December 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.116200