Anomaly Detection in Hyperspectral Image Based on SVDD Combined with Features Compression

Anomaly detection in hyperspectral image has been a research hot topic in recent years, and it has rich applications in many fields, such as disaster warning and military reconnaissance. Traditional anomaly detection methods in hyperspectral image usually need to assume that the data fits a certain distribution or use low-order statistical features, resulting in poor detection accuracy. This paper proposes an anomaly detection method in hyperspectral image based on SVDD combined with nonlinear feature mapping and feature compression. The selected bands of hyperspectral image are used to construct an autoencoder. The parameters of the autoencoder are adjusted by minimizing the reconstruction error. The output of the encoder is regarded as the compressed feature. Then the compressed feature is used to train the SVDD. The proposed method uses fewer features to construct an anomaly detection model. The experimental results on real datasets show that the proposed method has achieved outstanding results compared with other state-of-the-art methods.