Abstract: Firstly,we employ a technique based on the knowledge of fractal dimension （FD） and multi-fractal dimension （MFD）.The FD and MFD features are extracted from the original image and the differential gray images.Then a tree-structured wavelet transform is employed to extract the entropy for each frequency channel of the wavelet transform output as the features for texture classification and the features are normalized according to their own degree of dispersion.The features of both fractal and wavelet are then integrated to fulfill the recognition task.Recognition experiments have been made on the sonar images in different signa-l to-noise ratios and the test results show that a recognition ratio greater than90％ can be achieved by the approach.