Anomaly detection has been one of the most important applications for hyperspectral imagery. A new lossy compression method for hyperspectral imagery oriented anomaly detection is proposed. In order to keep the performance of anomaly detection, the anomalous vectors detected by the improved RX algorithm are preprocessed. Furthermore, virtual dimensionality algorithm is introduced to estimate the Intrinsic Dimensionality (ID) of original data while Karhunen-Loeve transform is used to provide spectral decorrelation. In addition, based on the virtual dimensionality, a new method for the number of principle component determination is presented. The bit rate of each principle component is distributed by optimal rate allocation strategy for spatial compression by SPIHT algorithm. Experimental results show that the proposed algorithm provides better compression performance, as well as efficient preservation for anomalous pixels.