An Approach of Classification Based on Pixel Level and Decision Level Fusion of Multi-source Images in Remote Sensing

With the availability of multi-sensor,multi-temporal,multi-resolution and multi-spectral image data from operational Earth observation satellites,the image fusion has become a valuable tool in remote sensing image evaluation.It is a relatively new and rapidly developing research field in remote sensing.In this paper,a pixel-level fusion algorithm of multi-source images in remote sensing based on high frequency modulation is studied.According to the characters of imaging system and principle of Heisenberg,a Gaussian filter is designed and used in the algorithm,which is proved to be effective.A back-propagation feed forward artificial neural network using momentum and adjusting learning rate by self-adaptation is studied.The speed and reliability of BP neural network are improved.A pixel-level fusion procedure and a decision-level fusion procedure for classification of multi-source remotely sensed images are proposed.A multi-source image set including Landsat TM3,4,5 and SAR has been used in classification.Compared with their classification accuracy obtained by the two procedures,the results show that the two procedures applied in classification of multi-source remotely sensed images are effective.