Land Cover Classification From Remote Sensing Images Using A Hierarchical Decision Tree Integrated OBIA Method

Zhongwei Oasis is typical in arid region, which heavily influenced by human activities with significant imprint. With the advances of regional modernization, Zhongwei Oasis suffers a problem of harmonizing ecological and environmental protection with economic growth. In recent years, land cover distribution pattern has changed dramatically with industrialization and urbanization, rapid development, agricultural upgrading, and accelerated pace. It results in environmental pollution and a series of related ecological problems. Remote sensing technology can quickly monitor large areas of land changes. Therefore, it provides the scientific background for decision making, environmental protection and ecological restoration. In this paper, based on three periods of Landsat 8 OLI and TIRS image data, the integrated method of Object Based Image Analysis (OBIA) and hierarchical decision tree, was used to classify land cover type. Main conclusions are as follows:(1) OBIA is an effective technology to classify land cover type. This method divides the image into objects by multi-scale resolution scales, which considers spectral, spatial geometry, texture and plant phenological characteristics. Besides, the advantage of OBIA technology can also be seen from the effective elimination of "salt and pepper phenomenon" in image classification.(2) OBIA can effectively present levels of land cover system. Super-object layer on a larger segmentation scale can produce a higher level of land cover classification, while sub-object layer con a smaller segmentation scale can generate a lower level of land cover classification.(3) A decision tree in data mining techniques can be used to build model in classifying land cover objects. Classification And Regression Tree (CART) and J48 decision tree excavates land cover information respectively, and CART decision tree is more simple and easier to understand than J48 decision tree. In hierarchical classification models, the accuracy of combination of CART+CART combination of models is the highest, increased by 3.52% compared with the nearest neighbor method, and Kappa coefficient increased by 0.06%.(4) The results from image classification show that farmland of, Zhongwei oasis on 2014mainly distributed in Zhongwei plains, where rice and corn are the main crops in this region, a new type of solar greenhouse in the north of the Yellow River, woodland in the southern edge of the Tengger Desert, garden in Nanshan platform, and desert shrub on a hillside around oasis.(5) The integration of hierarchical decision tree and OBIA method was proved to be an effective approach for land cover classification. This method can automatically obtain object-based rule sets and show up in the form of a decision tree, which can help analysis and understanding.