This paper deals with the optimal detection of man-made targets in UWB SAR images obtained using a FOPEN radar. Conventional detection algorithms perform poorly in this case, mainly because of the spiky aspect of the foliage clutter. We propose to use the K-compound distribution to model this kind of impulsive clutter. A method to overcome the problem of the closed forms for the detection statistics and to obtain reference ROC curves in this case is also proposed in the paper. The detection results are compared to those obtained for a Rayleigh distributed clutter.