Abstract:

Aimed to improve the classification accuracy of SPOT5 imagery, image segmentation, object feature extraction and selection, multi-classifiers and combination had been studied in this paper. A systemic approach to do classification of imagery was present. The proposed method is a five-step object-oriented classification routine that involves the integration of: imagery segmentation with larger scale, rule-based classification, classification-based segmenting, hierarchical classification under sub-region control and topper layer synthesizing. Five classifiers were employed to the classification, included minimum distance classifier, Mahalanobis distance classifier, Bayes rule classifier, Fuzzy classifier and support vector machine. The result indicated that in the study area with the fragmentized distribution of forest, species and type diversity, complex structure, using the Bayes classifier, the total accuracies of the third level, the second level and the first level were 79.38%, 81.82% and 86.98% respectively, where the third level contained twenty-two categories based on age-group of trees, the second level contained fifteen categories based on species, and the first level included nine categories based on species groups. With hierarchical classification, the result of upward synthesizing from lower levels to upper level was better than that classified from top levels to lower levels. Used as ancillary data, Landsat 7 ETM+ data were helpful to improve the classification accuracy of SPOT5 imagery. However, they could only be used to extract object features and couldn’t be involved in segmentation, as they would reduce the homogeneity and increase the heterogeneities of objects, and then affect the classification accuracy.

Key words: object-oriented, SPOT5 imagery, forest cover classification, multi-classifier, object features, selection