Class imbalance is a serious problem that plagues the semantic segmentation task in urban remote sensing images. Since large object classes dominate the segmentation task, small object classes are usually suppressed, so the solutions based on optimizing the overall accuracy are often unsatisfactory. In the light of the class imbalance of the semantic segmentation in urban remote sensing images, we developed the concept of the Down-sampling Block (DownBlock) for obtaining context information and the Up-sampling Block (UpBlock) for restoring the original resolution. We proposed an end-to-end deep convolutional neural network (DenseU-Net) architecture for pixel-wise urban remote sensing image segmentation. The main idea of the DenseU-Net is to connect convolutional neural network features through cascade operations and use its symmetrical structure to fuse the detail features in shallow layers and the abstract semantic features in deep layers. A focal loss function weighted by the median frequency balancing (MFB_Focal _loss ) is proposed; the accuracy of the small object classes and the overall accuracy are improved effectively with our approach. Our experiments were based on the 2016 ISPRS Vaihingen 2D semantic labeling dataset and demonstrated the following outcomes. In the case where boundary pixels were considered (GT), MFB_Focal _loss achieved a good overall segmentation performance using the same U-Net model, and the F1-score of the small object class “car” was improved by 9.28% compared with the cross-entropy loss function. Using the same MFB_Focal _loss loss function, the overall accuracy of the DenseU-Net was better than that of U-Net, where the F1-score of the “car” class was 6.71% higher. Finally, without any post-processing, the DenseU-Net+MFB_Focal _loss achieved the overall accuracy of 85.63%, and the F1-score of the “car” class was 83.23%, which is superior to HSN+OI+WBP both numerically and visually.

Social media analytics has drawn new quantitative insights of human activity patterns. Many applications of social media analytics, from pandemic prediction to earthquake response, require an in-depth understanding of how these patterns change when human encounter unfamiliar conditions. In this paper, we select two earthquakes in China as the social context in Sina-Weibo (or Weibo for short), the largest Chinese microblog site. After proposing a formalized Weibo information flow model to represent the information spread on Weibo, we study the information spread from three main perspectives: individual characteristics, the types of social relationships between interactive participants, and the topology of real interaction networks. The quantitative analyses draw the following conclusions. First, the shadow of Dunbar's number is evident in the ``declared friends/followers'' distributions, and the number of each participant's friends/followers who also participated in the earthquake information dissemination show the typical power-law distribution, indicating a rich-gets-richer phenomenon. Second, an individual's number of followers is the most critical factor in user influence. Strangers are very important forces for disseminating real-time news after an earthquake. Third, two types of real interaction networks share the scale-free and small-world property, but with a looser organizational structure. In addition, correlations between different influence groups indicate that when compared with other online social media, the discussion on Weibo is mainly dominated and influenced by verified users.

In image retrieval tasks, the single‐layer convolutional feature has insufficient image semantic representation ability. A new image description algorithm ML‐RCroW based on multilayer multiregion cross‐weighted aggregational deep convolutional features is proposed. First, the ML‐RCroW algorithm inputs an image into the VGG16 (a deep convolutional neural network developed by researchers at Visual Geometry Group and Google DeepMind) network model in which the fully connected layer is discarded. The visual feature information in the convolutional neural network (CNN) is extracted, and the target response weight map is generated by combining with the spatial weighting algorithm of the target fuzzy marker. Then, visual features in the CNN are divided into multiple regions, and the pixels of each region are weighted by regional spatial weight, regional channel weight, and regional weight. The image global vector is generated by aggregating and encoding every region in the weighted feature map. Finally, features of each layer of the VGG16 network model are extracted and then aggregated and dimensionally reduced to obtain the final feature vector of the image. The experiments are carried out on the Oxford5k and Paris6k datasets provided by Oxford VGG. The experimental results show that the average accuracy of image retrieval based on the image feature description algorithm ML‐RCroW is better than that achieved by the other commonly used algorithms such as SPoC, R‐MAC, and CroW.

With the rapid growth of the Internet, the scale of graphs has increased dramatically, which poses special challenges in representing both web graphs and social network graphs. In the adjacency matrix of web and social network graphs, only a very small proportion of the elements are “1“s. Furthermore, the authors find that using the aggregation of scattered 1s to form a high density of adjacency matrices is beneficial to the compression of storage space. Based on these findings, the authors propose the DGC-K2-tree compression approach based on K2-tree, which can greatly increase the density of 1s among the existing algorithms and adequately compress the blank area in the adjacency matrix. Then, the authors design a query algorithm for this mechanism to support the operation on the graph. The experimental results show that compared with the state-of-the-art algorithms, including the K2-tree based on a diagonal clustering mechanism(K2-BDC), the K2-tree, Re-Pair and LZ78, the authors’ approach achieves better compression ratio and shorter time consumption. In terms of storage efficiency, the author’s approach reduces the space by an average of 34.07the author’s approach reduces the time by an average of 80.63LZ78.

The two-terminal reliability calculation for wireless sensor networks (WSNs) is a #P-hard problem. The reliability calculation of WSNs on the multicast model provides an even worse combinatorial explosion of node states with respect to the calculation of WSNs on the unicast model; many real WSNs require the multicast model to deliver information. This research first provides a formal definition for the WSN on the multicast model. Next, a symbolic OBDD_Multicast algorithm is proposed to evaluate the reliability of WSNs on the multicast model. Furthermore, our research on OBDD_Multicast construction avoids the problem of invalid expansion, which reduces the number of subnetworks by identifying the redundant paths of two adjacent nodes ands-tunconnected paths. Experiments show that the OBDD_Multicast both reduces the complexity of the WSN reliability analysis and has a lower running time than Xing’s OBDD- (ordered binary decision diagram-) based algorithm.