The Analysis Of The Application Of The BP Neural Network In The Land Use Classification

Remote sensing is a burgeoning branch of science and technology. It is a comprehensive technology, and integrates mathematic, geoscience, computer science and technology, etc.. One of the main content of the technology is image classification, and it gains moe and more extensive and deeply applications. Of all the applications, to get the information about the land use and land cover through remote sensing technology has become one of the research hot points.In the beginning, in order to gain the land use and land cover information with remote sensing, visual interpretation is the only method. In the recently twenty years, devoted to some scholars, the classification technology by computer has been become more and more widely used. The computer classification technology is the application of the pattern identificaiton technology in the field of remote sensing. The core of this technology is to ascertain the identification interface and identification rules between different classes of land use and land cover. The technology can be used by many times, and can realize accurately location and fast processing. It is good at the time effect of the remote sensing information. However, because the remotely sensed data have such features as many classes, blend pixels, and high dimensions, it is rather difficult to realize high-accuracy classification. With the development of software in computer science, the classification system of remote sensing image can deal with not only single sensor data type, but also multisensor data types, which include many data formats. The procedure of image technology hs become quantative and intelligent.In this thesis, based on the electromagnetic reflection features of the ground objects, through digital image processing, we do the land use and land cover classification according to man-machine interactive processing. This is the thought of our classification.The study of spectrum features is the physics basis for the Remote Sensing research and it's also the premise for the land use and land cover classification. It is shown that the objects of the same type have the similar spectrum curves, and every object has its own feature spectrum property. As to the study area, the water has low reflection in TM1, 2, and 3, and from TM4 its reflection is nearly equal to zero. At some place, the water, affected by the saline or salt, has a relatively high reflection in all channels. This will be helpful to the study of the saline of the soil. The reflection of botany has high in TM4, 5 and 7, and in TM1, 2, and 3 the reflection is low, which helps us to easily identify it. Because the image is gained in January, the paddy field and the water is hard to identify from each other. But the reflection of paddy field has a high in TM4, and has the regular geometric outline. The reflection of dry land has high in TM1-4, and low in TM5 and 7. In the false color composite image of TM7, 4 and 2, the dry farm is lurid, and is regular in the geometric outline. Because the grass is withered at this time, its spectrum features is approximate to the sand's, but the reflection of the former is low the latter's. In the image, the grassland spreads widely and continuously, and sometimes it is mixed with the forest, the dry land and the saline, which makes it difficult to be identified. In the composite image of TM4, 3 and 2 the grass is cyan. All the spectrum features of these land cover and land use classes provide the basis for the land use and land cover classification of the region.The conventional classification methods are hard classification based on the spectral statistic features of the pixels. But these methods are not easy in dealing with such questions: they are different objects but have the same spectral features, and are same objects but have the different spectral features, and are mixed pixels. For these reasons, the results of the classification alwalys happen some uncertainty and fuzziness. The accuracy of these methods are too far to satisfy the requirement of constructing the land...