Remote Sensing Classification Of Forest Vegetation Based On Artificial Neural Network

Remote sensing image classification is always the important content of remote sensing research and forest resources inventory and monitoring. Application level and practical value are influenced by classification precision.The problems of multi-class image recognition and satisfaction precisions are key problems of remote sensing images which have very important significance.This paper base on domestic and abroad forest vegetation remote sensing classification study progress. The BP neural network methods study on forest vegetation remote sensing classification which SPOT5 remote sensing image of 10m resolution and geographic data Saihanba Mechanical Tree Farm in 2003. This paper makes a comparative precision study on traditional classification results of statistical pattern recognition. The results show that:(1)Because of the characteristics of Saihanba Mechanical Forest Farm, forest resources monitoring makes use of the method of SPOT5 Remote Sensing data pre-processing. The technical processes in the actual application as follows: the optimal band combination (2-4-1), Image Enhancement (Histogram Equalization, Principal Component Analysis Transformation, Tasseled Cap Transform, Geometric Rectification), orthophoto correction.(2)In this study, SPOT5 multi-spectral remote sensing images are main data source. Forest resource current situation distribution maps of classification study area are considered as geographic data. Overall accuracy of classification results types and total quality precision are attained 83.6%, Kappa coefficient is 0.7943 while BP neural network methods use in forest vegetation remote sensing classification. The result shows that classification quality is very well.(3)This research adopts class accuracy method and amount accuracy method making precision comparative analysis for maximum likelihood , simple and complex unsupervised classification,it is shown that classification types total precision of neural network methods are more greater 4%,47.6% and 35.6% than traditional three classification methods. BP neural network methods plus geographic data use in forest vegetation remote sensing classification having better effects and it is an effective image classification method. (4)While network node of hidden layer is 17, learning rate is 0.01, training times reach 12000 times,the network reach determining as prior classification precision of 95%. It can meet the demand of remote sensing classification of forest vegetation using BP neural network.