3. Classifying of Radarsat data landsat TM
After Composite procession of SAR and TM, the composite is used for classification. Unlike traditional remote sensing data, Radarsat data has some moise ant some times the feature of radar data of same objects maybe completely different due to this specialty of radar data, the tone or gray level for one kind of objects at the radar image are not same. Although the image feature for one kind object is nearly same from overall view, actually there exists many individual pixels with different feature. so the conventional classification algorithms based on per pixel classifiers are not suitable for radar data classification.
Neural Network classifiers is unlike the traditional classifiers, it is an algorithm which mimic the computational abilities of the brain. A neuron is the fundamental buiding block of the brain's nervous system, Artificial neurons are simple emulation of biological neurons; they take in information from sensors or other artificial nerons, perform very simple operations on this data, and pass the results to other artificial neurons (PCI Inc., 1997).
The Neural network includes three layers. The B-P (Back-Propagation) algorithm is used for learning process. The first layer is the input layer which is adopted to receive training data and data awaiting classification, the second is a hidden layer which is composed of a sigmoid conversion; function the third layer is an output layer that specified the true output of the network (Li Zhengyuan, et al, 1996).
The 30 learning samples used for the training of network, which was selected from pre-processed RADARSAT data and lands TM data, The input value of learning samples were obtained from the DN value for RADARSAT and lands TM (band 2,3,4). For the training process, the samples which influenced the output accuracy were removed. The network reached a stable state through thousands of training. The classification results were obtained after the data awaiting classification were input.
Resutrs Show the combination of Radarsat data and landsat TM and the classification result map. It is obvious that the classification accuracy has been greatly improved. In the classification image, old building, new construction, forest area, lake, farmland etc are early classified.
4. Result analysis on land use survey and crop recognition
It is shown in the composite of TM and SAR that before compounding, SAR image is a mono-colored image and reflects the information of ground object merely through the difference of gray level, while the composite image is colorful and with rich object information. However, there are clear defects of the composite image. Firstly, the effect of cloud. Visible light and infrared can not penetrate cloud. Therefore, image is very bright when the sensor of TM received the high radiation energy reflected from cloud and dark when from the ground covered by cloud. The bright and dark images cover the true feature of ground objects. For example, there are many areas are bright or dark in the lower-righted part of the image, where is Chuanshazheng district, useful information can not acquired. Along the bank of Huangpu River, there are some areas of heavy gray( piles of coal), similar to pond water in color and shape. Some streets with dense building are not clear in Landsat TM image. The composite image of TM and SAR mainly solves these problems. For example, in the cloud covered areas, object feature can be promoted by SAR. Therefor the piles of coal at river bank is easy discriminated from water body because coal pile is green while water body is blue black in the composite image. These information of linear-shaped objects, such as street, river, irrigation canal, are strengthened in the image. Additionally, at east to the old sea dyke, the paddy land is with a slight high elevation, heavy soil sand, short time for cultivation and different growth stage from the paddy land in the sea dykes. All these differences of soil and topographical conditions are covered by the information of rice with a good growth, while the differences are different in color in the composite image. In paddy land out the old sea dykes is brown while it is drab in the old sea dykes. Obviously, the difference in colors is the result of SAR remote sensor, which penetrates rice and get more information of soil.
The analysis result shows that in the experimental area, 357 polygons are interpretated by the composite image of TM/SAR, 212 polygons less than Landsat TM image before compounding. The decreased polygons are mainly residential areas and orchards. The number of Residential area is decreased 214 polygons from 388 to 174, orchard is reduced 21 polygons. The main reasons are: on the one hand, the experimental area is an developed district in Pudong. Urban area is expanded rapidly. Some regions were residential areas in 1996 while became city in 1997. About 10-15 residential areas distributed at slight east part of Pudong are connected and city areas is added 22.223 km2. Each area changes from residential area to city increases urban area about 1-3.5km2. Some orchards gradually changes into residential areas and land used for public traffic is also reduced gradually. On the other hand, the composite image of TM and SAR is good for land use type interpretation. Of course, due to the effect of cloud and other factors, some land use types, such as paddy land, arid land and wood land are not clear in Landsat TM while very clear in the composite image of TM and SAR. Thus, the polygons of water body increases 13, and 23 for arid land, 8 for wood land. The arid land and wood land in Chuanshazheng, which is covered by cloud, can be clearly recognized. At the bank of Huangpu river, water body and coal piles are discriminated easily. These are the superiority of the TM and SAR composite image. They also proved that in the experimental area, the composite image of TM and SAR are more favorable for land use survey and crop monitoring than Landsat TM.
5. Conclusion
Since 90,s, there is a clear acceleration in the development of space technology in the world. Synthetic aperture radar is the advanced technologies in earth observation field and become the focus for modern remote sensing technologies. In recent years, a series of satellite-board radar system have been studied or already launched successfully.
SAR technology can acquire the information of ground object under all weather conditions. It meets the requirement of dynamic monitoring with remote sensing data of high re-covered rate and short coverage cycle. In the meantime, synthetic aperture radar system records the dielectric and scattering properties of ground object. The former directly relates with moisture rate while the later reflects the rough condition of ground object. For vegetation, the later reflects its structure property. Radar remote sensing can penetrate vegetation and obtain soil information. It also can penetrate surface layer of dry soil and get information from certain depth. The penetration capacity of radar is different from spectral band to band. It is of special favor for land use survey, disaster detection, crop growth monitoring and classification.
However, how to process the huge information quickly, extract the useful information for agriculture investigation, precision farming, and make a full use of remote sensing for agricultural modernization in China, is the most important task at present.
6. References
- Li Zhengyuan, et al, 1996, Analysis of Multi-temporal JERS-1 SAR Data For Forest Mapping Over Zhaoyuan, Longkou Counties of China, Proceeding of the Second Asia Regional Globe SAR Workshop. pp.153.
- PCI Inc., 1997 Using PCI Software Volume I.pp.260.
- Tang Linli and Jiang Ping. 1995. Research on composite procession technology of SAR and TM image. Ground station of satellite, institute of remote sensing applications, CAS.
