KARST ROCKY DESERTIFICATION INFORMATION EXTRACTION BASED ON THE DECISION TREE

Rocky desertification is a common geo-ecological disasters in China are mainly distributed in southwest karst region, and a wide range of further deterioration. Based on the theory of decision tree Guangxi rocky information extraction, selection of experimental data of Guangxi Zhuang Autonomous Region in 2005 TM image. First of remote sensing images after geometric correction image registration and other pretreatment. Secondly based on binary model of pixel, the Guangxi Zhuang Autonomous Region NDVI values and vegetation cover and slope analysis combining the results of Guangxi Zhuang Autonomous Region, the use of decision tree classification of remote sensing images, and finally get different levels of Guangxi Zhuang Autonomous Region rocky area and spatial distribution. The experimental results showed that: 2005 Guangxi rocky area of about 22,000 km2, accounting for 9% of the total land area in Guangxi, accounting for 24.30% of the karst area the overall classification accuracy of 89.03%, Kappa coefficient was 0.8417. From the classification results and the accuracy evaluation shows that the use of the information extracted rocky achieve better results.  *Corresponding author. Tel:+86-15878346093 E-mail :yuetao@glut.edu.cn Sponsored by The project was funded by Guangxi Key Laboratory of Spatial Information and Geomatics (17-259-16-13) and 2019 Guangxi Young and Middle-aged Teachers' Research Fundamental Ability Enhancement Project (2019KY1363) The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-3/W10, 2020 International Conference on Geomatics in the Big Data Era (ICGBD), 15–17 November 2019, Guilin, Guangxi, China This contribution has been peer-reviewed. https://doi.org/10.5194/isprs-archives-XLII-3-W10-367-2020 | © Authors 2020. CC BY 4.0 License. 367


INTRODUCTION
Karst rocky is recognized in recent years to a new geo-ecological disasters, the impact on the economy, environment and society is growing, widespread national attention (Li-Yangbing 2004). Advantages of using remote sensing technology that can accurately, quickly and economically to get the status of karst rocky desertification, distribution, etc. for monitoring desertification, evolution and subsequent analysis of rocky provide effective governance scientific basis Common information extraction method is mainly rocky vegetation index or improving the use of vegetation index. Relatively simple method in the classification, the classification results may not be ideal. As a key of decision tree classification method based on knowledge, the algorithm complexity low and high efficiency, widely used in the field of remote sensing, which is characteristic variables to extract information and select the node threshold (Mahesh Pal 2003). In this paper, the above methods, the use of decision tree classification method for remote sensing images in Guangxi 2005 TM data rocky classified according to NDVI, vegetation cover and slope analysis.

TREE
Decision tree is an inductive learning training samples to generate decision rules, and decision rules for the use of new data through mathematical method of classification. A decision tree is a tree structure consists of a root node, consisting of a series of internal nodes and leaf nodes of three parts, a parent of two or more child nodes and nodes constitute each node is connected via a branch between nodes.
Decision tree is an intuitive knowledge representation method, which uses information theory mutual information (information gain) to find property field with the maximum amount of information in the database, create a node tree, and then created based on different values of the field branch of the tree; concentrated repetition to build the tree at each branch and sub-branch of the lower node processes (HansenM, 1996, DudaRO, 2003. In addition to the tree in the form of a tree (Figure 1), but also can be expressed as a set of IF-THEN rules generated form.
Each tree in the path from the root to the leaf corresponds to a rule, the rule is to determine the condition of all the property values of the nodes on this path, the conclusion is the rule class properties on this path leaf nodes.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-3/W10, 2020 International Conference on Geomatics in the Big Data Era (ICGBD), 15-17 November 2019, Guilin, Guangxi, China Decision tree for remote sensing image classification can be divided into three stages: (1) Data preparation phase, the remote sensing image data preprocessing, feature determines the type of statistical training in the region around the object type information (spectral and non-spectral information).
(2) The construction of rule phase, the larger the amount of information the property containing the rules as to form an internal node tree until the same category included within each leaf node so far.
(3) Decision tree trimming stage, check the classification results can meet the needs of, if not, you need to adjust the tree (pruning and add nodes) until the establishment of a proper tree so far.

Study area
Guangxi is a typical area of karst landforms, the

Data Introduction
In this paper, the use of remote sensing data classification images of Landsat-5 TM acquired imaging sensors and data processing through ASTER GDEM come digital elevation (DEM) product (data from the International Computer Network Information Center, Chinese Academy of Sciences data mirroring website (http: // www.gscloud.cn)). The basic parameters of the experimental data shown in Table 1.
The obtained of formula (2) and (3) (Table 2). In Envi4.8 in the above-mentioned indicators as a condition of rocky decision tree classification, decision tree classification rule base ( Figure 5), run the initial classification decision tree to get results.     Table 4.

CONCLUSION
(1) In this paper as a decision tree to determine the The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-3/W10, 2020 International Conference on Geomatics in the Big Data Era (ICGBD), 15-17 November 2019, Guilin, Guangxi, China