The Application Of Remote Sensing Technology In Zhalong Wetland Resources Survey

Zhalong wetland lies in Heilongjiang province of Northwest China. Its environment is facing degeneration because of continuous drought in recent years. Especially, Zhalong wetland suffered fire from the end of 2001 to the beginning of 2002. Therefore, artificial recharge projects were developed to hasten its recovery. To evaluate the effect of the fire and the recharge on Zhalong wetland scientifically, remote sensing is used to investigate the resource of Zhalong wetland. A four-layer neural network, which adopts improved learning algorithm, is presented to classify remotely sensed images. The changes brought by the recharge have been analyzed. The thought that images with similar spectra can adopt the same samples to do the classification is presented in this paper. And based on this thought, an automatic classification system is performed.The ETM+ images taken on March 14, 2002, May 17, 2002 and September 22, 2002 are used to investigate the resources of Zhalong wetland. To recognize different types of land use, the remotely sensed images come through pretreatment, classification, mapping and area measurement. After the pretreatment, the four-layer neural network, which adopts improved learning algorithm, is used to classify images of Zhalong wetland. Results show that the four-layer neural network appears to be feasible to classify the images of Zhalong wetland. Compared with three-layer neural network and maximum likelihood classifier, it has the highest classification accuracy. The improved algorithm, that is, the adaptive back-propagation learning algorithm based on the error robust function, avoids the occurrence of big error, accelerates decreasing rate of error and shortens the learning time. Statistics about the wetland area based on classification results suggest that the recharge reduces the effect of fire on marsh and promotes the recovery of the wetland. Additionally, the automatic classification system is realized under .NET platform. The thought that images with similar spectra can adopt the same samples to do the classification is presented, and the complicated flow design for establishing the relation between image and sample is accomplished. The system makes full use of collected samples and realizes the automatic classification of images.