The remote sensing constitutes a vast field of study whose repercussions are many and varied on environmental management. The phenomenon of dust clouds is a major climatic event in Africa. But the observation means of this phenomenon are still too much limited. The development of an approach consisting in the detection of dust clouds from satellite images can be a solution. In this work, we present a new approach for dust clouds detection in the infrared images coming from the METEOSAT satellite. It is then proved necessary of finding automatic or semi-automatic analysis methods to assist their detection and interpretation. Thus we are interested in image fusion methods for detection structures in the images. In this paper, we present some statistical methods which enable to extract texture features from the images. Then, we describe the method used for selection the best attributes for the images segmentation into three classes: "water clouds", "ocean" and "continent". We then use a method which enable us to segment the class "continent" of the image for dust clouds detection. Finally, we compare our results with another one which shows the zone of presence or absence of dust clouds. This comparison shows that we are in concord because visually, we have a good analogy of shape on the dust clouds zone as well as on the part without dust clouds.