Anomaly Detection in Noisy Multi and Hyper Spectral Images of Urban Environments

The recent surge in remotely sensed imagery with multi to hyper spectral cubes has made it very difficult to detect features because of the sheer volume of data. In a sense it is locating the needle in the haystack which in urban areas is extremely difficult unless we have specific knowledge of the anomaly spectrum. Adding noise and atmospheric masking makes it even more complex a problem. In this paper we will present an approach by which we base our study on a novel spectral-segmentation algorithm for multi-or hyper spectral images and consider how to detect multi-pixel environmental anomalous objects in the urban space. In particular, we have developed several filters to compensate for noise which may be present in the initial cube. We also assume no a priori knowledge on the objects other than the fact that they are different from the background and are regularly shaped. We show that for speckle noise, a modification of our morphology technique allows us to detect features without correcting for atmospheric influences nor producing an enhanced false alarm result. We will show several results from urban environs in Israel and the USA using a variety of sensors.