Multicore Feature Learning Approach for Maximizing Retrieval Information in Remote Sensing Images

Remote sensing (RS) images incorporate valuable information about the different objects captured from space or satellites. In the last two decades, a large amount of research has been carried out to retrieve the appropriate and related information from RS images. In this work, we propose a novel multicore feature learning (McFL) technique leveraging deep learning methods for retrieving content-based RS images. First, two convolutional neural network (CNN) architectures, i.e., VGG16 and ResNet50, pretrained with identical short-span images, are employed to extract the features. Second, the proposed McFL technique is used to improve the semantic representation of RS images for feature learners. The McFL technique helps the network to learn the most critical and optimized semantic features. Finally, with increased learning capacity, the similarity measure is computed between the query and database images. The Experimental results are conducted on two benchmark datasets to show the improved efficiency and accuracy of the proposed method for retrieving RS images.