Satellite Image Scene Categorization Based on Topic Models

We present a scene classification method for satellite images based on the topic model-probabilistic latent semantic analysis(pLSA) and feature-combination.Firstly,three kinds of features(SIFT,geometric blur and colorhistogram) are extracted from images.Then,we apply the the pLSA model on these image features to obtain the probabilities of latent topics which will be combined subsequently.Finally,we implement SVM classification based on these probability features.The experimental results on the 12-category dataset show that our proposed method performs better in scene classification.