PROJECT TITLE :
Learning High-level Features for Satellite Image Classification With Limited Labeled Samples
This paper presents a novel technique addressing the classification task of satellite pictures when limited labeled information is offered together with a massive quantity of unlabeled knowledge. Rather than using semi-supervised classifiers, we solve the problem by learning a high-level features, referred to as semisupervised ensemble projection (SSEP). Additional precisely, we propose to represent a picture by projecting it onto an ensemble of weak coaching (WT) sets sampled from a Gaussian approximation of multiple feature spaces. Given a set of pictures with restricted labeled ones, we have a tendency to first extract preliminary features, e.g., color and textures, to create a coffee-level image description. We then propose a brand new semisupervised sampling algorithm to create an ensemble of informative WT sets by exploiting these feature spaces with a Gaussian normal affinity, that ensures both the reliability and diversity of the ensemble. Discriminative functions are subsequently learned from the ensuing WT sets, and every image is represented by concatenating its projected values onto such WT sets for final classification. Moreover, we think about that the potential redundant info existed in SSEP and use sparse coding to scale back it. Experiments on high-resolution remote sensing information demonstrate the efficiency of the proposed methodology.
Did you like this research project?
To get this research project Guidelines, Training and Code... Click Here