Spectra-Difference based anomaly-detection for infrared hyperspectral dim-moving-point-target detection

• Published in: Infrared physics & technology 2023-01, Vol.128, p.104489, Article 104489
• Main Authors: Wu, Tianxiao, Wen, Maoxing, Wang, Yueming, Yao, Yi, Zhang, Dong, Chen, Fansheng, Wang, Jianyu
• Format: Article
• Language: English
• Subjects: Anomaly detection; Dim-moving-point-target; Hidden covariance matrix; Spectra-difference
• ISSN: 1350-4495; 1879-0275
• DOI: 10.1016/j.infrared.2022.104489

•Dim-moving-point-target is detected under complex background using hyperspectral sequence images.•Spectra-difference is used to suppress complex background and highlight target.•Covariance matrices constructed with Gaussian mixture model are applied to further dig detection performance. Traditional anomaly detection methods are mainly based on single frame detection, and the model is unstable due to the influence of complex optical background. In order to improve the performance of dim-moving-point-target detection in infrared hyperspectral sequence images under complex background clutter, this paper proposes effective anomaly detection algorithms based on spectra-difference. Firstly, two frames of hyperspectral images without targets are used for spectra-difference, the optical background is effectively eliminated. Secondly, in order to describe the residual image, the Gaussian mixture model is used to classify the residual image. Thirdly, the maximum class sample, equal-probability sample and all-classes sample are constructed by using the classified data, and the covariance matrices of three samples are calculated to further tap the description ability of the background. Finally, spectra-difference is performed between the two hyperspectral images to be detected, and the corresponding covariance matrix is used for anomaly detection. Theoretical and experimental results show that the algorithm has high performance, and the three detectors have different application scenarios. Dim-target detection is not affected by multi-targets and bright targets. The target velocity can be calculated by using the target coordinates and the imaging time interval.