Classification Of Remote Sensing Image Based On Object-oriented And Decision Tree Algorithm

Classification of Remote Sensing image is the basis for thematic mapping and the study ofremote sensing applications. It has an important position in processing remote sensing image.The classification accuracy directly affects the quality of the image products’ accuracy. Theclassification with high accuracy will help the qualitative and quantitative analysis of the imageand help the thematic information extraction. In all aspects of processing remote sensing image,the classification method is the key technologies, which affect the quality of classification.Therefore, with the object-oriented technology combined with the decision tree algorithm, thepaper studied the method of classification, designed to improve the classification accuracy. Makeit more widely used in land resources survey, land dynamic monitoring, basic geographicinformation updates, and so on.This paper studies the contents as follows:(1) A method of Object-oriented Remote Sensing Image Classification Based on C4.5algorithm is advanced in this paper. Firstly, the image is segmented into image objects using themulti-scale segmentation method. Then the features of objects, such as spectral、texture、shapeand layer characteristics, are measured. The rules are automatically obtained using C4.5algorithm. At last, classification of the remote sensing is carried out using the rules. Also,Classified the high-resolution image with this method, and the result shows the superiority of thismethod.(2) The extraction of classification rules is achieved automation and visualization. Throughthe study of the C4.5algorithm, use it to obtain the rulers automatically by training samples.Also, the rules are visible. It breaks that the rulers obtained by experiences in traditional decisiontree classification. Those who do not need to have more knowledge of remote sensing andgeosciences can do the work of classification.(3) The tree editor is established. So, the function that classification rulers obtained by C4.5algorithm combined with manual intervention has been changed into reality. The design andimplementation of the Tree Editor provides the possibility of human intervention in theclassification. In classification, if the human intervention and the mathematical theory arecombined, especially in more complex terrain areas, Will help to improve the accuracy ofclassification.