Classification of remote sensing data using partially trained neural network

The feasibility of a partially trained artificial neural network technique for classification of remote sensing sea ice coverage is presented. The neural network technique used is feed-forward back-propagation, and the sensing object is the ice coverage over Arctic region. This ice coverage information is obtained from the special sensor microwave imager (SSMI) microwave radiative measurements. Seven channels brightness temperature are used to identify six different surface classes. Different stages of partially trained feed-forward back-propagation artificial neural networks have been applied for the classification of ice coverage in order to investigate the performance of partial trained network at different training stages and to reduce the lengthy training time required by most BP ANN architectures.<>