Deep Active Learning for Multi-Label Classification of Remote Sensing Images

In this letter, we introduce deep active learning (AL) for multi-label classification (MLC) problems in remote sensing (RS). particular, investigate the effectiveness of several AL query functions MLC RS images. Unlike existing (which are defined single-label or semantic segmentation problems), each function paper is based on evaluation two criteria: i) uncertainty; and ii) diversity. The uncertainty criterion associated to confidence neural networks (DNNs) correctly assigning multi-labels image. To assess criterion, three strategies: loss ordering; measuring temporal discrepancy predictions; iii) magnitude approximated gradient embeddings. diversity selection a set images that as diverse possible other prevents redundancy among them. exploit clustering strategy. We combine above-mentioned strategies with strategy, resulting different functions. All considered introduced first time framework RS. Experimental results obtained benchmark archives show these result highly informative samples at iteration process.