The mapping of wetland habitats requires image data of high spatial resolution in order to establish the precise contours and space occupies a specific habitat. However, the spectral deficiency of high resolution images accentuates the problems of proximity and spectral mixing between image objects, which makes them very sensitive classification operations in such environments. The present work offers a solution based on an unsupervised approach to habitat classification of the wetland lagoon of Oualidia and its surroundings. To do this, a picture RBV (1m) covering the study area was segmented from the software GRASS, followed by extraction optimal segments as polygons from QGIS software. The partitioning algorithm K-means was used for classification of selected polygons in the respective classes, and this using three (3) discrimination criteria (color, shape, and size). The objective is to propose a solution in the discrimination of different types of wetland habitats from a poor image spectral resolution, but harboring a very high spatial resolution. As such, the algorithm permits to classify the different habitats with an accuracy of 0.88 according to the index of Kappa.

Spatial accuracy is important information in the mapping of wetland habitats, hence the recourse to the use of data with very high spatial resolution such as the IKONOS satellite images. However, the mediocrity of these spectral images; the presence of mixed pixels or spectral confusion between different objects in the image, make the process of discrimination of wetlands habitats difficult. This difficulty is amplified because these areas are home to diversified habitats, and in most times have spectral similarities between them. Taking into consideration the problems mentioned above, this work proposes a hybrid classification approach to better discriminate the habitats from an IKONOS data covering the Wetland of Merja Zerga (Moroccan Wetland of International Importance). This approach combines a supervised classification (per pixel) and an object-oriented classification (unsupervised) using DBSCAN algorithm. This classification approach allows discriminating all wetland habitats of the Merja Zerga and produces a map with an accuracy of 0.86 according to the index Kappa. The results suggest that this classification approach may also provide better results by applying it on images having similar characteristics to the IKONOS image.