Classification of high spatial resolution images using semantic allocation level-probabilistic topic model

Scene classification is treated to be a philosophical and the powerful method for HSR (High spatial resolution) images. Many academicians and researchers paid their attention towards the agglomeration of numerous features. Initially, a survey on PTM (Probabilistic Topic Model) was done in a finest manner and they have concluded that, a single feature (i.e., a spectral feature) was not best suited for HSR images. The next investigation was performed on CAT-PTM and their basic theory behind this method was that, the words of visual dictionary are highly correlated. Due to these inadequacies, the above methods are not valid for HSR images. Thus, the paper proposes SAL-PTM (Semantic Allocation Level - Probabilistic Topic Model) method through which three features (i.e., Texture, Scale Invariant Feature Transform and Spectral) are extracted for better performance. The semantic description of low level descriptors is generated by means of K-means clustering. Finally, the features obtained from the latent semantic allocations are isolated by means of PTM and their performance was evaluated and compared by using LDA (Latent Dirichlet Allocation) and PLSA (Probabilistic Topic Model). A U.S geological survey dataset and UC Merced Dataset was tested on SAL-PTM (Semantic Allocation Level-Probabilistic Topic Model). In response to that, a precise outcome was obtained suggesting that our proposed SAL-PTM method was confined to prove its effectiveness.