Visual Attention and Background Subtraction With Adaptive Weight for Hyperspectral Anomaly Detection

Anomaly detection (AD) in hyperspectral target is of particular interest because no prior knowledge ground object spectra required. However, it difficult to utilize the salient features image (HSI) and mitigate effects noise AD, which greatly limits performance. Here, we report a strategy implement AD by visual attention model background subtraction with adaptive weight. Through band selection method, most discriminating bands are selected as input images for subsequent processing. Then, introduced, first time, into extracting feature map images. Furthermore, process that can reduce developed via curvature filter. Using this operation, initial anomaly area obtained. Finally, incorporating spectral information, an weight applied further suppress background. In experiment, proposed method compared seven other state-of-the-art methods on synthetic real-world HSI. Most importantly, results demonstrate effective performs better than alternative methods. We believe open new avenue processing AD.