Mapping of Freshwater Lake Wetlands Using Object-Relations and Rule-based Inference

Inland freshwater lake wetlands play an important role in regional ecological balance.Hongze Lake is the fourth biggest freshwater lake in China.In the past three decades,there has been significant loss of freshwater wetlands within the lake and at the mouths of neighboring rivers,due to disturbance,primarily from human activities.The main purpose of this paper was to explore a practical technology for differentiating wetlands effectively from upland types in close proximity to them.In the paper,an integrated method,which combined per-pixel and per-field classification,was used for mapping wetlands of Hongze Lake and their neighboring upland types.Firstly,Landsat ETM+ imagery was segmented and classified by using spectral and textural features.Secondly,ETM+ spectral bands,textural features derived from ETM+ Pan imagery,relative relations between neighboring classes,shape features,and elevation were used in a decision tree classification.Thirdly,per-pixel classification results from the decision tree classifier were improved by using classification results from object-oriented classification as a context.The results show that the technology has not only overcome the salt-and-pepper effect commonly observed in the past studies,but also has improved the accuracy of identification by nearly 5%.

Inland freshwater lake wetlands play an important role in regional ecological balance. Hongze Lake is the fourth biggest freshwater lake in China. In the past three decades, there has been significant loss of freshwater wet- lands within the lake and at the mouths of neighboring rivers, due to disturbance, primarily from human activities. The main purpose of this paper was to explore a practical technology for differentiating wetlands effectively from upland types in close proximity to them. In the paper, an integrated method, which combined per-pixel and per-field classifi- cation, was used for mapping wetlands of Hongze Lake and their neighboring upland types. Firstly, Landsat ETM＋ imagery was segmented and classified by using spectral and textural features. Secondly, ETM＋ spectral bands, textural features derived from ETM＋ Pan imagery, relative relations between neighboring classes, shape fea^xes, and elevation were used in a decision tree classification. Thirdly, per-pixel classification results from the decision tree classifier were improved by using classification results from object-oriented classification as a context. The results show that the technology has not only overcome the salt-and-pepper effect commonly observed in the past studies, but also has im- proved the accuracy of identification by nearly 5%.