Techniques Of Automatic Classifying And Identifying The TM Remote Sensing Images Of Typical Mixed Forest With Needle And Broad Leaves

Forest resource information is one of the nationally important foundations and is necessary for various decisions of forestry construction. How to economically, and efficiently obtain the forest resource information is an important work of realizing digital forestry. Remote sensing technique, resulting for its characteristics such as macro, comprehensive, repeatable, express delivery and economical etc., become an ideal tool that studies the forest resources status and its dynamic change. However, for a long time, the high cost of obtaining the data and the poor technique of classifying and identifying forest remote sensing images have been a technical bottleneck of forestry.In the paper, according to the demand of developing the forest, the TM remote sensing images of typical needle-broad mixed forest are research objects. For improving the precision of the automatic classifying and identifying, after traditional techniques are tested and analyzed in precision, new SCIS and BNN are explored and applied, which have importantly theoretical and practical meaning for providing new techniques in forest remote sensing.The main contents and conclusions of this paper are induced as follows:(1) The Specialist Classify and Identify System (SCIS) aided by Geography Information System (GIS) can be well applied to classify the TM images of mixed forest. With the aid of GIS, the data of Digital Elevation Model (DEM), slope direction, soil type and other geography information, highly related with forest vegetation distribution, are combined with pretreated image spectral information to form the knowledge system. Furthermore, by eliciting the information from the knowledge system and forming the rule and decision tree, SCIS is established. The test result shows that the whole index of classify precision is 81.67% ,and the whole Kappa index is 0.7556 ,which is improved 14.22% more than that of traditional technique. And the quality of TM image is better and has attained to the purpose of differentiating forest types.(2) The modified BP Nerve Network Classify and Identify System has a very wide application foreground in TM images. By a series of measures such as standardizing of input vectors, transforming of PCI, modifying arithmetic of exercising, amplifying the network insize and so on, the classified precision has been improved greatly. By using the net, its whole classify precision is 76.00% and the whole Kappa index is 0.6800 which is raised 19.14% more than before. So the result shows that this system can be used to TM image classify and it has a potential to improve its precision with network pattern's augmentation.(3) The proper methods can reduce the effects from terrain shade and cloudy. Through revising radiation, revising topography, combining specialist system with GIS and constructing nerve network system, the results of decreasing effects, to resolve some problems that cloudy zones can not be classified in images and weaken terrain shade, are obvious.