<p>Hyperspectral anomaly detection (HAD) is a critical remote sensing task challenged by background variability and the absence of prior information about anomalies. Many existing methods model the background from the observed scene, which can be incomplete, contaminated, or require scene-specific parameterization. This paper introduces a novel approach, synthetic background Learning for anomaly detection (SBL-AD), centered on advanced background learning via high-quality synthetic spectral samples. The core idea is to construct a comprehensive synthetic background library (SBLib) by augmenting signatures from standard spectral libraries. Two distinct strategies are employed for augmentation: an algorithmic method that introduces random noise, angular variations, scaling, and bias to pristine spectra, and a generative adversarial network (GAN) trained to produce realistic spectral variations. This synthetic library creates an extensive training set of common background materials. This dataset is then used to train a support vector machine (SVM) to learn the characteristics of various background classes. Anomalies in hyperspectral scenes are subsequently identified as pixels exhibiting a low probability of belonging to any of the learned background classes. This proactive approach aims to overcome the limitations of scene-dependent estimation, improving accuracy and robustness by minimizing reliance on test scene statistics for more blind detection. Preliminary evaluations on benchmark datasets indicate the promising capability of the SBL-AD method.</p>