Neighbor-Based Label Distribution Learning to Model Label Ambiguity for Aerial Scene Classification

Many aerial images with similar appearances have different but correlated scene labels, which causes the label ambiguity. Label distribution learning (LDL) can express ambiguity by giving each sample a distribution. Thus, contributes to of its ground-truth as well improve data utilization. LDL has gained success in many fields, such age estimation, be easily modeled on basis prior knowledge about local similarity and global correlations. However, never been applied classification, because there is no correlations thus it hard model In this paper, we uncover neighbors that cause jointly capturing propose neighbor-based (N-LDL) for classification. We define subspace problem, formulates neighboring relations coefficient matrix regularized sparse constraint The provides few nearest neighbors, captures similarity. are predefined according confusion matrices validation sets. During learning, encouraged agree correlations, ensures uncovered labels. Finally, propagation among forms distributions, leads smoothing terms distributions used train convolutional neural networks (CNNs). Experiments image dataset (AID) NWPU_RESISC45 (NR) datasets demonstrate using clearly improves classification performance assisting feature mitigating over-fitting problems, our method achieves state-of-the-art performance.