| Information Extraction of Forest is the core issue in the forestry resource monitoring, and object-oriented classification of tree species as the key part of information extraction of forest vegetation, is the focus of the present study. Based on object-oriented species classification, this paper is to explore the establishment of the object-oriented species classification and develop species classifications the ALOS image for data source in Fujian Jiangle State-owned forest farm. By constructing the object-oriented classification method based on rough set rules extracted, a set of classification rules was established and the classification of the image in the experimental zone was completed. This study initially formed methods and techniques of forest vegetation information extraction method based on object-oriented technology.The main results of this study were as follows:(1)The pre-processing of the experimental data of the study area was to be done, including geometric correction, the best band selection, image fusion and cutting. The best band combination was selected by evaluating different band combinations.At the same time, the study analyses a variety of fusion method and select the most suitable one for the ALOS. It turned out that the means that we selected was the best one for our research.(2)The segmentation method of object-oriented were discussed in detail. The emphasis was to analyze segmentation algorithm based on edge information and its application to image segmentation of the study area, building the image information of different scales hierarchical structure to complete the Segmentation of the image in the study area.(3)As the object was formed by segmentation, the attribute of the object characteristics was reduced based on rough set algorithm.13parameters of the most representative of the species classification was selected from the34characteristics of the object to build a suitable feature space. The rules set of the classification was established to classify the species of the study area.(4)The classification accuracy assessment of object-oriented classification results was carried out. The compare between object-oriented classification results based on rule extraction of rough set、 object-oriented classification based on the nearest neighbor algorithm and supervised classification results showed that when extracting forest vegetation based on object-oriented classification with Rough set rules, the extracted species and the actual distribution of the species have high shape and attribute consistency. The classification accuracy was higher and it can effectively avoid the "salt and pepper phenomenon" which consists in the conventional classification method. The classification results was easier to interpret and understand. |
