The automatic detection of temporal changes in sonar images is of high interest for monitoring critical areas as ports or channels used by submarines for instance. This problem is addressed here as incoherent change detection between two sonar passes. To be able to compare two surveys, sonar images have first to be registered to each other. A mean shift between the master image and the slave image has been estimated here at a coarse-to-fine image resolution by minimizing the Kullback-Leibler divergence between the first and the second sets of amplitude distributions related to homogeneous areas of the master image. Then two different methods have been proposed to detect changes in terms of contacts in the common area of the registered images. The first method is one of the well-known techniques used for synthetic aperture radar imagery: the log ratio of images. The second one is made of two stages: firstly a goodness-of-fit test is applied to every image divided into small overlapped snippets in order to detect statistical deviations due to possible contacts; secondly a pixelwise difference is simply performed between these two images of detected snippets put in a non-zero pixel value. For both methods morphological operations and local correlations have been performed in order to discard false alarms. Advantages and drawbacks of the two proposed methods are discussed with results on a set of data collected by DRDC-Atlantic with a high-frequency sidescan sonar a month apart in the winter of 2008 in a port area.