New fusion method applied on IKONOS satellite images of urban and rural areas

Nowadays different countries have used remote sensing technology to get physical parameters of the objects without direct physical contact with them. Using remote sensing sensors, huge amount of data is available from the Earth surface which is used in different sustainable developing programs. Due to the various available operating sensors and valuable unique information acquired by each type of sensors, then there is a need to suite from different available images acquired by different sensors simultaneously which is called image fusion technique. In image fusion, different images from different sensors are used as inputs to the fusion algorithms and finally an output image is produced which has more useful information about the interesting object comparing to each of the input images. Of course better fusion algorithms result in better fusion output images. In one case, a higher resolution gray scale image is fused with a lower resolution color image to produce a high resolution color image which inherits the spatial resolution from high resolution gray scale image and spectral information from low resolution color image simultaneously in the output fused image. In this research, we have used panchromatic and multispectral bands of IKONOS satellite image of Malard region in Iran. At first small window from the panchromatic and multispectral bands are introduced to the neural networks (NNs) and after learning phase we have predicted the panchromatic band for other parts from the multispectral data. Then the predicted panchromatic band has been used in HSI fusion method and the results are compared with the original panchromatic band. Various artificial neural networks are designed and tested to find the suitable one. It is found that in the test area using the mentioned parameters in the paper and various implemented neural networks, a multilayer feed-forward with back propagation could be the best choice. Although the results are promising themselves but it needs- to carry out more tests on other type of images using more complex neural networks to achieve better results.