« IEEE fyrirlestur: Advances in Spectral-Spatial Classification of Hyperspectral Imagery | Main | IEEE — VFI »
IEEE fyrirlestur: Neural networks for classification and change detection
Skrifað af mou | October 21, 2009
Þriðjudaginn 27 október klukkan 16:40 í stofu 158 í VR II, Háskóla Íslands mun Dr. Prashanth Reddy Marpu frá Háskólanum í Pavia á Ítalíu halda fyrirlestur sem nefnist Neural networks for classification and change detection. Fyrirlesturinn verður fluttur á ensku.
Útdráttur/abstract
Feed forward neural networks are well established in several fields of science for regression and classfication applications. However, in most cases, a poor choice of the training algorithm has been the biggest hindrance for utilizing the complete potential of neural networks. Traditionally, backpropogation
algorithm is used to train the neural networks. It is notoriously slow and has the problem that it can converge at the local minima of the cost function. This talk will focus on applications of neural networks in remote sensing image analysis with an emphasis on the results of two advanced training algorithms namely, scaled conjugate gradients and Kalman filter which do not converge at the local minima. Also, neural networks can approximate any functional continuous mapping from one finite dimensional space to another. So, neural network regression can be used for normalization of multi-temporal images even if the pixel intensities have non-linear variations. A new method
for automatic change detection using iterative neural network regression will be presented.
Topics: Óflokkað | No Comments »