The orthogonal subspace projection (OSP) is a supervised classification method

which needs the information of the objects to be classified. To broaden its application fields

we propose the local orthogonal subspace projector (LOSP) and apply it in the anomaly detection of the hyperspectral image. The anomaly detection algorithms find the isolated man-made objects in the nature background

and the kind of the substances in the small local region is usually uniform. Based on the principle

the local projector is constructed through using the detected pixel as the interested object and the mean of its nearby pixels as the object to be suppressed. The experiments show that LOSP can detect the sub-pixel target whose content is greater than 30%

and the target that occupies more pixels can be detected by enlarging the size of the local region. It has been also been tested that LOSP is not affected by the Hughes phenomenon and it costs less than 10% of the RX algorithm when the image has 80 channels. LOSP is effective in both the precision and efficiency

and is applicable to the real-time detection of the hyperspectral image.