Research And System Implementation Of A Multi-Classifier Combined Decision Tree Hierarchical Classification Method

Classification of remote sensing image is an important aspect of image processing,which is the basis of the follow-up extraction of thematic information, detection ofdynamic change, production of thematic maps, databases of remote sensing image, etc.With the development of modern remote sensing technology, remote sensing informationhas become richer. Therefore, the accuracy and real-time requirement of classificationhave become more important. However, the improvement in classification accuracy oftraditional classification methods is not so good to meet the needs of practicalapplications.On the basis of existing classification algorithms, the paper analyzed the advantagesand disadvantages of the traditional single classifier, and the influence factors ofclassification accuracy. Study the classification method in the overall process level. Themain idea of the method is: First, using the decision tree algorithm produces the initialdecision tree, and then connecting the initial decision tree with other single classifiers toform a compound decision tree, then using the mask method, to reduce the interferencebetween class and class in classification procedure. Finally，merging the results ofone-class element mask overlay method as final general classification result. Features ofthis paper are as follows:Firstly, the two main factors found to affect the classification accuracy are theselection of samples and interference between the classes. The accuracy of samplesselection is limited by operators and tools. The latter is mainly affected by the influence ofthe algorithm. Thus the main problems could be figured out.Secondly, two problems mentioned above are solved. Classifier combinationmethod based on decision tree was used to compensate for the dependence ofsamples.One-class element mask overlay method was used to reduce themisclassification or reclassification.Finally, a fast, automatic and stable classification system based on the architecture ofserver/client was proposed by means of engineering, modular, process-oriented,hierarchical design model. This system can achieve fast and accurate classification of remote sensing image processing. And the experiments prove the efficiency of the parallelclassification.