Deep Unsupervised Weighted Hashing for Remote Sensing Image Retrieval

Deep unsupervised hashing methods are gaining attention in the field of remote sensing (RS) image retrieval due to rapid growth volume unlabeled RS data. Most previous research used only natural image-based pre-trained models generate label matrices; however, this method cannot capture semantic information images well and limits accuracy retrieval. To solve problem, authors propose a deep weighted (DUWH) model that uses similarity matrix updating strategy based on structure achieve mutual optimization hash network. The devise novel combinatorial loss function improve performance can be obtain higher quality codes by assigning different weights sample pairs with difficulties. Experiments were conducted two datasets verify excellent proposed method.