Rough Set, ELM Classifier and Deep Architecture for Remote Sensing Images

The progress in Remote Sensing (RS) technology provides high- definition images of land cover. The precise classification of RS images lies in leveraging its rich feature spectrum. In this context, Convolutional Neural Network (CNN) is a prominent tool for feature extraction, producing a high number of extracted features. With more number of input features the network architecture and its training becomes complex and time consuming, respectively. Our study addressed these issues by using rough set concepts to select the most informative features for classification with extreme learning machine (ELM). Thether, the ELM network connection is partially established through a random rule matrix, effectively reducing network complexity and computational requirements without compromising model performance. The designed model is evaluated on two data sets i.e., UC Merced and RSSCN7. The model’s superior classification performance is compared with Support Vector Machine (SVM) and similar methods on the ground of overall accuracy, precision, F1 score etc. measures.