Détection des cuvettes oasiennes du Centre-Est du Niger par classification d’une image texturale basée sur la variance - Archive ouverte HAL Accéder directement au contenu
Article Dans Une Revue Photo-Interprétation. European Journal of Applied Remote Sensing (PIEJARS) Année : 2017

Detection of oasis basins in East-Central Niger by textural image classification based on the variance

Détection des cuvettes oasiennes du Centre-Est du Niger par classification d’une image texturale basée sur la variance

Résumé

This study assesses the possibility of detection of oasis basins in East-Central Niger through textural analysis of satellite images based on the variance. Texture analysis serves to find descriptive attributes that quantify images granularity characteristics, directionality or finesse. These textural characteristics are usually related to image objects properties. An image texture analysis provides information about the nature of bjects, constituting the segments of homogeneous regions, and improves image quality. The study area is located in the Sahel-Saharan region of Niger. Note that agriculture and animal husbandry are the main economic activities in this area. Irrigated and rainfed crops guarantee almost all cereal production. Irrigated crops grown in the oasis basins provide a minimum level of output less sensitive to fluctuations in rainfall than rainfed crops. Oasis basins are closed depressions within which flow is endorheic. These basins are the only places where irrigated agriculture is practiced during the dry season. They therefore constitute the main source of income for farmers in the region.Indeed, no study has previously made an exhaustive inventory of these closed depressions. The only censuses carried were often limited to exploited oasis basins. In this context this study has been initiated to preserve the oasis basins and assess their socio-economic roles. The developed method detects oasis basins based on the image variance of the grey levels (digital counts) in a 5x5 pixels moving window of the infrared band of a SPOT5-THX image (2.5 m spatial resolution) acquired in November 2013. This procedure is performed in three steps. The first stage analyses the cumulative frequency curve of variance values determined by a threshold value. The high window variance values, representing the contours of the basins, were determined from a threshold adapted to the classification results. This threshold value allowed with the «erase» tool of ArcGIS to make a mask of the oasis basins and their surroundings, and another mask representing the other items: dry valleys, dune edifices and hills. Both masks led to a segmentation of the composite image into two parts. To improve the accuracy of the results, a supervised classification, using the Mahanalobis distance algorithm, was performed on the first part of the image, oasis basins and some surrounding bare soil dunes. The second part of the image, representing dune edifices, dry valleys and hills, was used in step 2. In this step, a second supervised classification, using the same algorithm, was applied on the composite image without cuvettes oasis, hence allowing to distinguish between the various constituent objects. In the third step, the results from step 1 and 2 were combined. Classification evaluation is based on a confusion matrix. The evaluation indicators calculated are: Kappa coefficient, overall accuracy, rate of reliability of a class and correct classification rate. Following thresholding, binarization of the textural channel derived from the infrared band of SPOT 5 THX image first delineates the oasis basins; then this result has been improved through supervised classification of the oasis basins boundaries. The combined results of the two images show four classified geomorphic units: oasis basins (11512 ha or 7.4% of the study area), dry valleys (32386ha or 20.7% of the study area), dune edifices (112499 ha or 71.8% of the study area) and hills (218 ha or 0.1% of the study area). Classification evaluation has given satisfactory conclusions with a Kappa coefficient of 0.96 and a rate of correct oasis basins classification of 82.7%. Our results confirm several studies showing that the performance of image classification by textural analysis is greater than using simple pixel by pixel analysis. But some confusions remain between classes. These confusions can be explained by the definition of classes identified in the study, which are geomorphological units. Yet, these geomorphological units may have similar land cover. For example, a temporary water can be located at a top of a hill, in a dry valley or in an oasis basin. However, the performance of the oasis basin detection method can be improved by combining several textural attributes to create a textural composite image. This study demonstrates the possible identification and measurement of the surface of oasis basins in a part of the Gouré Department; this could be done even with free remote sensing software allowing computation of a textural channel. The inventory of these oasian basins resources is a necessary first step in their monitoring and the evaluation and management of their agricultural potential.
Cette étude évalue les capacités de détection des cuvettes oasiennes du centre-est du Niger par analyse texturale d’images-satellites, basée sur la variance. Ces cuvettes constituent les seuls endroits où l’agriculture irriguée est pratiquée durant la saison sèche. La méthode mise au point est une détection des cuvettes oasiennes à partir de la bande infrarouge d’une image SPOT5-THX (2,5 m de résolution spatiale) en analysant la courbe de la fréquence cumulée des valeurs de variances déterminées dans une fenêtre de 5*5 pixels, et en s’appuyant sur une valeur seuil. L’évaluation des résultats de la classification d’image a donné des conclusions satisfaisantes avec un coefficient Kappa égal à 0,96 et un taux de bonne classification des cuvettes de 82,7%. Cette étude permet de localiser et de mesurer la superficie des cuvettes dans la zone d’étude. L’inventaire de ces cuvettes permettra de les caractériser et de les suivre afin d’évaluer leurs potentialités agricoles.
Fichier non déposé

Dates et versions

hal-01630700 , version 1 (08-11-2017)

Identifiants

  • HAL Id : hal-01630700 , version 1

Citer

Mahamadou Karimou Barké, Bernard Tychon, Issa Ousseini, André Ozer. Détection des cuvettes oasiennes du Centre-Est du Niger par classification d’une image texturale basée sur la variance. Photo-Interprétation. European Journal of Applied Remote Sensing (PIEJARS), 2017, 51 (3), pp.24-34. ⟨hal-01630700⟩
182 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More