Improving satellite image processing via hybridization of fusion, feature extraction & neural nets

Abstract

Satellite image classification is useful for many applications including but not limited to, crop classification, military equipment identification, movement tracking and forest cover detection. These applications involve image segmentation, feature extraction and application of a classifier to perform the final categorization task. This texts presents a hybrid approach which uses multispectral image fusion using brovey and principal component analysis methods, with the purpose of boosting the eminence of the image segmentation method, this when combined with hybrid feature extraction and classification process, tends to produce highly accurate classification results. We compare the classification accuracy of a standard support vector machine (SVM) with cascaded neural networks and observe that the neural network performs 20% better than SVM when applied to crop identification application