Unsupervised Classification of Remotely Sensed High resolution Images using RP-CNN

In recent years, various deep learning frameworks have been developed for the classification of remotely sensed images. However, the network models proposed in these frameworks exhibit high complexity and do not yield high classification accuracy when applied to unlabeled scenarios. This paper introduces a Multi spectral image (MSI) classification approach that combines the random patches network with self-supervised branch (RPSS) to extract informative deep features. The proposed method involves convolving image bands with random patches to obtain multi-level deep features. Subsequently, we use panchromatic image (PAN) to extract spatial features. The MS spectral features, the derived RPSS features and spatial features then merged to classify the MSI using a support vector machine (SVM) classifier. The experimental results on real remotely sensed images have been presented.