Study On Classification Of Remote Sensing Image Based On Wavelet Transform And BP Artificial Neural Network

With the application of remote sensing image classification is more andmore widely, computer automatic classification method is more and moreimportant. The method of traditional remote sensing image classification isusing the statistical characteristic value of remote sensing image or thestatistical relationship between the sample data of training area for patternrecognition. The surface is complex, remote sensing image exist "differentbody with same spectrum" or "same body with different spectrum"phenomenon. In the past decade, with the great development of the researchon Artificial Neural Networks(ANN),ANN has become an important methodfor the application of RS image classification, neural network provides thepossibility to solve the deficiency of the traditional remote sensing imageclassification. BP neural network classification is an effective method of Landcover classification,which，when compared with other traditional methods，has better abilities of self-learning and self-configuring, to improve theclassification accuracy.This study first reviewed and analyzed the latest progress of research onRS image classification home and abroad, take the TM remote sensing data inAugust2009and IRS-P5image in DaLiNuoEr nature reserve as data source,use the ENVI and Matlab as platform, Based on the original band, principalcomponent analysis and wavelet transform，we classify the remote sensingimage using BP neural network. Through choosing the training samplebuilding methods and training algorithm, confirming the number of besthidden layer neurons, to system study on the processes of BP neural networkclassification and determine the best parameters of suitable for this study.This paper makes a comparative precision study on traditional classification(supervised classification and non-supervised classification) results ofstatistical pattern recognition.Research shows that using BP neural network classification based on Wavelet transform the total accuracy is86.74%，Kappa coefficient0.8407.The accuracy of Artificial neural network classification is better than thetraditional supervised classification and non-supervised classification, theaccuracy of BP neural network classification based on Wavelet transform ishigher than BP neural network classification based on the original band andprincipal component analysis. Under the similar classification accuracyconditions, the principal component transformation or wavelet transform canmake more rapid convergence and faster simulation speed.It proves that it is feasible that remote sensing image classification inDaLiNuoEr nature reserve using BP neural network classification based onWavelet transform.