191 
Water body 
Paddy field 
Arid land 
Building area 
Road 
Vegetation 
Subsi deneeland 
Figure 3(b). The output image classified by Supervised 
Classification 
4.4 Accuracy Assessment 
Accuracy assessment is one of the indispensable jobs in the 
process of the remote sensing data classification.Through 
accuracy assessment, the classifying person can ascertain the 
validity of the classification signature, improve the 
classification signature, enhance classification accuracy; the 
user can gain information in the classified image correctly and 
effectively according to the classification accuracy. [l21 The 
method based on the confusion matrix is universally 
recommended classification accuracy assessment method. In 
this study, we chose 1,024 detecting samples randomly 
referring to TM image and 1:50000 topographic maps, then 
through visual interpretation to structure the confusion matrix, 
carried out classification accuracy assessment based on the 
correlation index calculated. 
5. CONCLUSIONS 
This study used the thought of CART analysis whose tree 
shaped was simple,clear and intuitionistic. It caused the multi 
characteristic and multi-model land types of the trial area to be 
clearer, so it easily realized the automatic recognition in the 
compute. 
This study used CART analysis to classify and extract the 
mining area land resources, had yielded some result, specially 
had extracted the subsidence lands.But,because of the time and 
insufficiently experienced myself, the CART decision tree was 
not too perfect. After ground truth investigation, we found one 
subsidence land was omited and two arid lands had classified 
into subsidence land by mistake, but this thought was feasible, 
in later work can improve the CART decision tree shape, cause 
it to be more perfect. 
REFERENCES : 
[1] Dazhi Guo, Yehua Sheng, Mingxing Hu etc., 1998. Mining 
Area Environment Disaster Dynamic Monitor and Analysis 
Assessment. The press of Chinese Mining University, Xuzhou. 
[2] Yinhui Zhang, Gengxing Zhao, 2002a. The summary of 
remote sensing data classification methods about land use/ land 
cover. Chinese agricultural resources and districts, 23(3), pp. 
21-25. 
[3] Kaichang Di, 2001. Spatial Data Mining and Knowledge 
Discovery. The press of Wuhan University, Wuhan. 
[4] Yurong Gao, 2006b. Study on land use information 
extraction based on decision tree method. Dissertation 
Submitted to Zhejiang University For Degree of Master, 
Zhejiang. 
[5] Breiman L, Friedman J H, Olshen R A, etc., 1984. 
Classification and Regression Tees. Monterey, California, 
U.S.A.: Wadsworth International Group, pp. 1-358. 
[6] Yohannes Y, Hoddinott J, 1999. Classification and 
Regression Tree: An Introduction. Washington, D.C., U.S.A.: 
International Food Policy Research Institute. 
[7] Ping Zhao, 2003b. Knowledge-based Landuse/cover 
Classification in the Typical Testareas of the Lower Reaches of 
Yangtze River. Dissertation Submitted to Nanjing University 
For Degree of Doctor, Nanjing. 
[8] Haralic R M, Shanmugam K, 1973a. Dinstein I Texture 
Features for Image Classification. IEEE Tmnsactions on 
Systems, Man and Cybernetics, (6), pp. 610-621. 
[9] Deshen Xia, Desheng Fu, 1997. The Modern Image 
ProcessingTtechnology and Application. The press of 
Southeast University, Nanjing. 
[10] Treitz P, Howarth P, 2000a. Integrating Spectral Spatial 
and Terrain Variables for Forest Ecosystem Classification. 
Photogrammetric Engineering & Remote Sensing, 66(3), pp. 
305-317. 
[11] Franklin S E, Pebble D R, 1989a. Spectral Texture for 
Improved Class Discrimination in Complex Terrain. 
International Journal of Remote Sensing, 54, pp. 1727- 1734. 
[12] Jianping Wu, Xingwei Yang, 1995a. Accuracy analysis of 
classification of remote sensing data. Remote Sensing 
Technology and Application, 10(1), pp. 17-24.