A rich line of works focus on designing elegant loss functions under the deep metric learning (DML) paradigm to learn a discriminative embedding space for remote sensing image retrieval (RSIR). Essentially, such embedding space could efficiently distinguish deep feature descriptors. So far, most existing losses used in RSIR are based on triplets, which have disadvantages of local optimization, slow convergence and insufficient use of similarity structure in a mini-batch. In this paper, we present a novel DML method named as global optimal structured loss to deal with the limitation of triplet loss. To be specific, we use a softmax function rather than a hinge function in our novel loss to realize global optimization. In addition, we present a novel optimal structured loss, which globally learn an efficient deep embedding space with mined informative sample pairs to force the positive pairs within a limitation and push the negative ones far away from a given boundary. We have conducted extensive experiments on four public remote sensing datasets and the results show that the proposed global optimal structured loss with pairs mining scheme achieves the state-of-the-art performance compared with the baselines.

As satellite observation technology improves, the number of remote sensing images significantly and rapidly increases. Therefore, a growing number of studies are focusing on remote sensing image retrieval. However, having a large number of remote sensing images considerably slows the retrieval time and takes up a great deal of memory space. The hash method is being increasingly used for rapid image retrieval because of its remarkably fast performance. At the same time, selecting samples that contain more information and greater stability to train the network has gradually become the key to improving retrieval performance. Given the above considerations, we propose a deep hash remote sensing image retrieval method, called the hard probability sampling hash retrieval method (HPSH), which combines hash code learning with hard probability sampling in a deep network. Specifically, we used a probability sampling method to select training samples, and we designed one novel hash loss function to better train the network parameters and reduce the hashing accuracy loss due to quantization. Our experimental results demonstrate that HPSH could yield an excellent representation compared with other state-of-the-art hash approaches. For the university of California, merced (UCMD) dataset, HPSH+S resulted in a mean average precision (mAP) of up to 90.9% on 16 hash bits, 92.2% on 24 hash bits, and 92.8% on 32 hash bits. For the aerial image dataset (AID), HPSH+S achieved a mAP of up to 89.8% on 16 hash bits, 93.6% on 24 hash bits, and 95.5% on 32 hash bits. For the UCMD dataset, with the use of data augmentation, our proposed approach achieved a mAP of up to 99.6% on 32 hash bits and 99.7% on 64 hash bits.

Image retrieval based on a convolutional neural network (CNN) has attracted great attention among researchers because of the high performance. The pooling method has become a research hotpot in the task of image retrieval in recent years. In this paper, we propose the feature distribution entropy (FDE) to measure the difference of regional distribution information in the feature maps from CNNs. We propose a novel pooling method, which fuses our proposed FDE with region maximum activations of convolutions (R-MAC) features to improve the performance of image retrieval, as it takes the advantage of regional distribution information in the feature maps. Compared with the descriptors computed by R-MAC pooling, our proposed method considers not only the most significant feature values of each region in feature map, but also the distribution difference in different regions. We utilize the histogram of feature values to calculate regional distribution entropy and concatenate the regional distribution entropy into FDE, which is further normalized and fused with R-MAC feature vectors by weighted summation to generate the final feature descriptors. We have conducted experiments on public datasets and the results demonstrate that our proposed method could produce better retrieval performances than existing state-of-the-art algorithms. Further, higher performance could be achieved by performing these post-processing on the improved feature descriptors.