Search

In the field of object detection in large-scale remote sensing images, achieving a good tradeoff between model accuracy and speed has been a long-standing challenge. The majority of inference time is spent on background regions without objects, making real-time detection difficult in practical applications. Common approaches involve partitioning large-scale remote sensing images into smaller patches, followed by using additional classification networks or detectors on the final layer of the backbone's feature map to identify and filter out patches devoid of objects, ultimately enhancing detection efficiency. This article proposes a novel model, called OPD-Swin-Transformer, for ship detection in large-scale remote sensing images. This model integrates a simple and lightweight object presence detector (OPD) at each stage of the Swin-transformer and uses a step-by-step, progressively challenging strategy to filter out background image patches, achieving an overall improvement in detection speed. The model optimizes the entire network end-to-end using a multitask loss function, leading to simultaneous improvements in detection accuracy. By employing an optimal threshold generation strategy based on the weighted Youden index, the model effectively maintains a higher recall rate for ships while filtering out background images, achieving an optimal balance between speed and accuracy. Our OPD-Swin-Transformer is integrated into two mainstream detectors and evaluated on two popular benchmarks for ship detection. The experiments demonstrate that, when compared to other state-of-the-art methods, this approach increases inference speed by more than 40% while also improving detection accuracy.

We propose and experimentally demonstrate a tunable frequency-doubling optoelectronic oscillator (FD-OEO) based on a single-bandpass dispersion-induced microwave photonic filter (MPF) consisting of a Mach–Zehnder modulator (MZM), a linearly chirped fiber Bragg grating and polarization-multiplexed dual-loop. Thanks to the polarization dependence of the MZM, a special double sideband modulation is implemented where the optical carrier (OC) and subcarriers are orthogonally polarized. By simply tuning the PC in the OEO loop, the phase difference between the orthogonal polarization carrier and two sidebands can be controlled, and thus the center frequency of the fundamental OEO can be tuned. Furthermore, a PC and a polarizer are placed outside the OEO to achieve optical carrier suppression (OCS) modulation, which ensures that a frequency-tunable microwave signal at the second-harmonic frequency is generated. In the experiment, a fundamental frequency signal with tunable frequency from 3.6 to 6.85 GHz and FD-OEO with a tunable frequency range from 7.2 to 13.7 GHz are generated.

Event argument extraction, which aims to identify arguments of specific events and label their roles, is a challenging subtask of event extraction. Previous approaches solve this problem in a two-stage manner that first extracts named entities as argument candidates and then determines their roles. However, many nested entities may be missed or wrongly predicted during the argument candidate extraction procedure, which substantially affects the performance of the downstream classifier. In this paper, we propose a novel one-step question answering based framework, which performs argument candidate extraction and argument role classification simultaneously to mitigate the error propagation problem in conventional two-stage methods. Since the conventional question answering based framework cannot be applied directly to this task, we design a Question Answering based Sequence Labeling (QA-SL) model to tackle inexistent argument roles and multiple argument token spans. Moreover, considering the overwhelming number of parameters in question answering based neural network models and the relatively small size of event extraction corpus, we fine-tune the pre-trained model from BERT to mitigate the data scarcity problem. Extensive experiments demonstrate the benefits of the proposed method, leading to a competitive performance compared with state-of-the-art methods. To the best of our knowledge, this is the first work to cast event argument extraction as a question answering task.

Knowledge Base Question Answering (KBQA) is a promising approach for users to access substantial knowledge and has become a research focus in recent years. Our paper focuses on relation detection, a subtask of KBQA and proposes an adversarial training improved multi-path multi-scale relation detector (AdvT-MMRD) to improve the performance of a common KBQA system. To solve the problem of matching the casual form of a question with the logical form of a predicate, we use question pattern-relation matching, in which an attention-based bidirectional recurrent neural network with gated recurrent units (Bi-GRUs) is used to match semantic similarity and a convolutional neural network (CNN) is used to learn literal similarity between question and relation. We also explore two ways to measure the relevance of entity type-relation pairs through several level representations. Additionally, an adversarial training strategy is conducted to enhance our model. The experimental results demonstrate that our approach not only achieves a state-of-the-art accuracy of 93.8% on relation detection task, but contributes our KBQA system to reaching an outstanding accuracy of 79.0% on the SimpleQuestions benchmark.

Network embedding (NE) aims to represent network information appropriately by learning low-dimensional and dense vectors for the nodes and edges of information network. Actually, the real world is almost full of heterogeneous information networks, which stimulates the emergence of heterogeneous information networks (HINs) embedding models. However, parts of existing HIN embedding models like meta-path-based methods only capture limited and aggregated information of relations, whereas some models based on metric or distance learning are usually of high computational complexity and slow training speed. In this paper, we present a novel heterogeneous information network embedding model, which applies dynamic projection metrics and translation mechanisms to the complicated heterogeneous information networks including multiple nodes and different relations. In order to overcome the imbalance of the distribution of relations in HIN and optimize the training process, we introduce an adaptive loss function for model optimization. Further more, we propose a hybrid model with baseline method as the initialization of the model. Experiments have been implemented on some real-world HIN datasets. And empirical results show that our model significantly outperforms the state-of-the-art representation learning models.

Aspect-based sentiment analysis, which aims to predict the sentiment polarities for the given aspects or targets, is a broad-spectrum and challenging research area. Recently, pre-trained models, such as BERT, have been used in aspect-based sentiment analysis. This fine-grained task needs auxiliary information to distinguish each aspect. But the input form of BERT is only a words sequence which can not provide extra contextual information. To address this problem, we introduce a new method named GBCN which uses a gating mechanism with context-aware aspect embeddings to enhance and control the BERT representation for aspect-based sentiment analysis. Firstly, the input texts are fed into BERT and context-aware embedding layer to generate BERT representation and refined context-aware embeddings separately. These refined embeddings contain the most correlated information selected in the context. Then, we employ a gating mechanism to control the propagation of sentiment features from BERT output with context-aware embeddings. The experiments of our model obtain new state-of-the-art results on the SentiHood and SemEval-2014 datasets, achieving a test F1 of 88.0 and 92.9 respectively.

Aspect-level sentiment analysis is a fine-grained task in sentiment analysis that aims at detecting sentiment polarity towards a specific target in a sentence. Previous studies focus on using global attention mechanism that attends to all words in the context to model the interaction between target and sentence. However, global attention suffers from assigning high-attention score to irrelevant sentiment words in the cases where sentence contains noisy words or multiple targets. To address this problem, we propose a novel syntax-directed hybrid attention network (SHAN). In SHAN, a global attention is employed to capture coarse information about the target, and a syntax-directed local attention is used to take a look at words syntactically close to the target. An information gate is then utilized to synthesize the information from local and global attention results and adaptively generate a less-noisy and more sentiment-oriented representation. The experimental results on SemEval 2014 Datasets demonstrate the effectiveness of the proposed method.

In the electromagnetic space, a single channel radar receiver will often intercept several periodic pulse trains radiating from the surrounding emitters simultaneously. The aim of radar pulse deinterleaving is to sort out the pulses coming from different emitters. Most traditional pulse repetition interval (PRI) deinterleaving methods are easy to sort out the pulses with small PRI fluctuations but difficult in dealing with relatively bigger fluctuation or staggered PRIs. In addition, this searching procedure can easily cause pulse omission phenomenon and even generate false pulses. In this paper, we propose an improved histogram method for PRI deinterleaving based on pulse correlation to overcome the above-mentioned shortcomings. After calculating the multi-level time difference histogram, we introduce the mean filter and interquartile range algorithm to optimize the estimated PRI values. Our method extracts the pulse pairs based on pulse correlation directly instead of searching for the pulses and then determines whether the PRI is staggered or not. The experiments on simulation data show that our method can achieve better performance on both the pulse trains of jittered PRIs and the staggered PRIs.

Graph layout investigates the structure of the graph in order to better obtain the information implied in the graph. To solve the shortcomings of dimension reduction layouts on local adjustment and the insufficiency of energy models to maintain the overall structure of the graphs, this paper proposes a new graph layout framework called “tNEM” that layouts graphs by combining t-distributed neighbor retrieval visualizer (t-NeRV) and energy models. In the process of layout, our algorithm considers global and local structures at the same time. The layout results are more conform to aesthetic standards, meanwhile, maintain the structural information of the graph. We evaluate our algorithm on a wide variety of datasets and compare it with many other methods. We produce better visualization results than tsNET and tsNET* methods by reducing the tendency to crowd points together, and can better capture the global structure of the graph.

Named entity recognition (NER) is one of the basic techniques in natural language processing tasks. Chinese NER is complicated and difficult which remains a major challenge. One of the main reasons is that the boundaries of entities in Chinese are blurred and closely related to word segmentation results. Previous studies for this task have broadly divided into two categories, word-based, and character-based methods. However, the former class suffers from the word segmentation errors, and the latter cannot make full use of the information on multiple granularities. To address these problems, we investigate a new dynamic meta-embeddings method and apply it to Chinese NER task, which utilizes attention mechanism to combine features of both character and word granularity in embedding layer. The meta-embeddings created by our method are dynamic, data-specific, and task-specific, since the meta-embeddings for same characters in different sentence sequences are distinct. The experiments on MSRA and LiteratureNER datasets validate the effectiveness of our model, and this method achieves the state-of-the-art results on LiteratureNER.

Chinese Named Entity Recognition (Chinese NER) is an important task in Chinese natural language processing field. It is difficult to identify the boundary of entities because Chinese texts lack natural delimiters to separate words. For this task, two major methods can be distinguished by the model inputs, i.e., word-based model and character-based model. However, the word-based model relies on the result of the Chinese Word Segmentation (CWS), and the character-based model cannot utilize enough word-level information. In this paper, we propose a multi-granularity information fusion model (MIFM) for the Chinese NER task. We introduce a novel multi-granularity embedding layer that utilizes the attention mechanism and an information gate to fuse the character and word level features. The results of this embedding method are dynamic and data-specific because they are calculated based on different contexts. Moreover, we apply the reverse stacked LSTM layer to gain deep semantic information for a sequence. Experiments on two benchmark datasets, MSRA and ResumeNER, show that our approach can effectively improve the performance of Chinese NER.

Graph visualization with proper layout is widely applied to understand the relationship between entities in a complex system and the topological structure information is mainly used. Real-world graphs often have the community structures property which is ignored in many existing graph layout methods. Thus, we propose a multivariate network layout method using the force-directed method with attribute constraints. This method can effectively take into account the hierarchical structure, connection strength, and quantitative comparison between communities. First, the layout of community centers is generated by a force-directed algorithm in which node count of the community is taken as constraints to enable area balance of community; Second, community force based on node attribute is added in the force-directed algorithm to maintain the community clarity. A visualization system is also developed to allow users to interactively generate community structure-aware layout results. qualitative and quantitative evaluation of the results verifies the usability and effectiveness of the proposed method by comparing it with other methods.

Multiple-object tracking (MOT) has received an increasing attention due to the rapid development of autonomous driving. However, the MOT problem is still challenging mainly due to the occlusion and scale variation. Motivated by the fact that the discriminative correlation filters-based (DCFB) tracking algorithms can tackle these problems and significantly improve the accuracy of single object tracking, how to exploit the DCFB tracking algorithms for MOT is worthy studying. Moreover, the corrupted training samples due to the occlusion make DCFB tracking methods to update the appearance model of target uncorrected and result in tracking drift. In this paper, we exploit Markov decision process to integrate the DCFB tracking method into our MOT framework and address the update problem of the appearance model in DCFB tracking method. Moreover, in order to overcome the challenges of occlusion and scale variation, to prevent target drift during tracking, we use two DCFB trackers with different update frequencies and a novel update strategy to predict the location of targets. The part-based method is used to extract robust features to tackling the challenges of occlusion and scale change. In order to verify the efficiency of our algorithm, experiments are performed based on KITTI tracking benchmark. The results demonstrate that our method achieves state-of-the-art performance and outperforms the state-of-the-art algorithms in road scenarios.

Unlike single geospatial objects extraction from high-resolution remote sensing images, the task of road extraction faces more challenges, including its narrowness, sparsity, diversity, multiscale characteristics, and class imbalance. Focusing on these challenges, this paper proposes an end-to-end framework called the multiple feature pyramid network (MFPN). In MFPN, we design an effective feature pyramid and a tailored pyramid pooling module, taking advantage of multilevel semantic features of high-resolution remote sensing images. In the optimization stage, a weighted balance loss function is presented to solve the class imbalance problem caused by the sparseness of roads. The proposed novel loss function is more sensitive to the misclassified and the sparse real labeled pixels and helps to focus on the spare set of hard pixels in the training stage. Compared with the cross-entropy loss function, the weighted balance loss can reduce training time dramatically for the same precision. Experiments on two challenging datasets of high-resolution remote sensing images which illustrate the performance of the proposed algorithm have achieved significant improvements, especially for narrow rural roads.

Community detection plays a vital role in network analysis, simplification, and compression, which reveals the network structure by dividing a network into several internally dense modules. Among plenty of methods, those based on statistical inference are widely used because they are theoretically sound and consistent. However, in many of them, the number of communities needs to be provided in advance or computed in a time-consuming way and parameters are usually initialized randomly, resulting in unstable accuracy and low convergence rate. In this paper, we present a community detection method based on modified PageRank and stochastic block model, which is able to compute the number of communities by finding community centers and initialize community assignments according to the centers and distance. Experiments on both synthetic and real-world networks prove that our method can intuitively give the number of communities, steadily get results of high NMI and modularity and efficiently speed up the convergence of optimizing likelihood probability.

Relation detection plays a crucial role in knowledge base question answering, and it is challenging because of the high variance of relation expression in real-world questions. Traditional relation detection models based on deep learning follow an encoding-comparing paradigm, where the question and the candidate relation are represented as vectors to compare their semantic similarity. Max- or average-pooling operation, which is used to compress the sequence of words into fixed-dimensional vectors, becomes the bottleneck of information flow. In this paper, we propose an attention-based word-level interaction model (ABWIM) to alleviate the information loss issue caused by aggregating the sequence into a fixed-dimensional vector before the comparison. First, attention mechanism is adopted to learn the soft alignments between words from the question and the relation. Then, fine-grained comparisons are performed on the aligned words. Finally, the comparison results are merged with a simple recurrent layer to estimate the semantic similarity. Besides, a dynamic sample selection strategy is proposed to accelerate the training procedure without decreasing the performance. Experimental results of relation detection on both SimpleQuestions and WebQuestions datasets show that ABWIM achieves the state-of-the-art accuracy, demonstrating its effectiveness.

With the development of information technology in modern military confrontation, specific emitter identification has become a hot and difficult topic in the field of electronic warfare, especially in the field of electronic reconnaissance. Specific emitter identification requires a historical reconnaissance signal as the matching template. In order to avoid being intercepted by enemy electronic reconnaissance equipment, modern radar often has multiple sets of working parameters, such as pulse width and signal bandwidth, which change when performing different tasks and training. At this time, the collected fingerprint features cannot fully match the fingerprint template in the radar database, making the traditional specific emitter identification algorithm ineffective. Therefore, when the working parameters of enemy radar change, that is, when there is no such variable working parameter signal template in our radar database, it is a bottleneck problem in the current electronic reconnaissance field to realize the specific emitter identification. In order to solve this problem, this paper proposes a network model based on metric learning. By learning deep fingerprint features and learning a deep nonlinear metric between different sample signals, the same individual sample signals under different working parameters can be associated. Even if there are no samples under a certain kind of working parameter signal, it can still be associated with the original individual through this network model, so as to achieve the purpose of specific emitter identification. As opposed to the situation in which the traditional specific emitter identification algorithm cannot be associated with the original individual when the signal samples of changing working parameters are not collected, the algorithm proposed in this paper can better solve the problem of changing working parameters and zero samples.

High-quality digital road maps are essential prerequisites of location-based services and smart city applications. The massive and accessible GPS trajectory data generated by mobile GPS devices provide a new means through which to generate maps. However, due to the low sampling rate and multi-level disparity problems, automatically generating road maps is challenging and the generated maps cannot yet meet commercial requirements. In this paper, we present a GPS trajectory data-based road tracking algorithm, including an active contour-based road centerline refinement algorithm as the necessary post-processing. First, the low-frequency trajectory data were transferred into a density estimation map representing the roads through a kernel density estimator, for a seeding algorithm to automatically generate the initial points of the road-tracking algorithm. Then, we present a template-matching-based road-direction extraction algorithm for the road trackers to conduct simple correction, based on local density information. Last, we present an active contour-based road centerline refinement algorithm, considering both the geometric information of roads and density information. The generated road map was quantitatively evaluated using maps offered by the OpenStreetMap. Compared to other methods, our approach could produce a higher quality map with fewer zig-zag roads, and therefore more accurately represents reality.

As most networks come with some content in each node, attributed network embedding has aroused much research interest. Most existing attributed network embedding methods aim at learning a fixed representation for each node encoding its local proximity. However, those methods usually neglect the global information between nodes distant from each other and distribution of the latent codes. We propose Structural Adversarial Variational Graph Auto-Encoder (SAVGAE), a novel framework which encodes the network structure and node content into low-dimensional embeddings. On one hand, our model captures the local proximity and proximities at any distance of a network by exploiting a high-order proximity indicator named Rooted Pagerank. On the other hand, our method learns the data distribution of each node representation while circumvents the side effect its sampling process causes on learning a robust embedding through adversarial training. On benchmark datasets, we demonstrate that our method performs competitively compared with state-of-the-art models.

The growing use of deep neural networks in critical applications is making interpretability urgently to be solved. Local interpretation methods are the most prevalent and accepted approach for understanding and interpreting deep neural networks. How to effectively evaluate the local interpretation methods is challenging. To address this question, a unified evaluation framework is proposed, which assesses local interpretation methods from three dimensions: accuracy, persuasibility and class discriminativeness. Specifically, in order to assess correctness, we designed an interactive user feature annotation tool to provide ground truth for local interpretation methods. To verify the usefulness of the interpretation method, we iteratively display part of the interpretation results, and then ask users whether they agree with the category information. At the same time, we designed and built a set of evaluation data sets with a rich hierarchical structure. Surprisingly, one finding is that the existing visual interpretation methods cannot satisfy all evaluation dimensions at the same time, and each has its own shortcomings.

In recent years, convolutional neural networks (CNNs) have shown great success in the scene classification of computer vision images. Although these CNNs can achieve excellent classification accuracy, the discriminative ability of feature representations extracted from CNNs is still limited in distinguishing more complex remote sensing images. Therefore, we propose a unified feature fusion framework based on attention mechanism in this paper, which is called Deep Discriminative Representation Learning with Attention Map (DDRL-AM). Firstly, by applying Gradient-weighted Class Activation Mapping (Grad-CAM) algorithm, attention maps associated with the predicted results are generated in order to make CNNs focus on the most salient parts of the image. Secondly, a spatial feature transformer (SFT) is designed to extract discriminative features from attention maps. Then an innovative two-channel CNN architecture is proposed by the fusion of features extracted from attention maps and the RGB (red green blue) stream. A new objective function that considers both center and cross-entropy loss are optimized to decrease the influence of inter-class dispersion and within-class variance. In order to show its effectiveness in classifying remote sensing images, the proposed DDRL-AM method is evaluated on four public benchmark datasets. The experimental results demonstrate the competitive scene classification performance of the DDRL-AM approach. Moreover, the visualization of features extracted by the proposed DDRL-AM method can prove that the discriminative ability of features has been increased.

The requirements of location-based services have generated an increasing need for up-to-date digital road maps. However, traditional methods are expensive and time-consuming, requiring many skilled operators. The feasibility of using massive GPS trajectory data provides a cheap and quick means for generating and updating road maps. The detection of road intersections, being the critical component of a road map, is a key problem in map generation. Unfortunately, low sampling rates and high disparities are ubiquitous among floating car data (FCD), making road intersection detection from such GPS trajectories very challenging. In this paper, we extend a point clustering-based road intersection detection framework to include a post-classification course, which utilizes the geometric features of road intersections. First, we propose a novel turn-point position compensation algorithm, in order to improve the concentration of selected turn-points under low sampling rates. The initial detection results given by the clustering algorithm are recall-focused. Then, we rule out false detections in an extended classification course based on an image thinning algorithm. The detection results of the proposed method are quantitatively evaluated by matching with intersections from OpenStreetMap using a variety of distance thresholds. Compared with other methods, our approach can achieve a much higher recall rate and better overall performance, thereby better supporting map generation and other similar applications.

In the above article [1], the authors’ units need to be changed because of the requirements of the school.

In recent years, entity-based ranking models have led to exciting breakthroughs in the research of information retrieval. Compared with traditional retrieval models, entity-based representation enables a better understanding of queries and documents. However, the existing entity-based models neglect the importance of entities in a document. This paper attempts to explore the effects of the importance of entities in a document. Specifically, the dataset analysis is conducted which verifies the correlation between the importance of entities in a document and document ranking. Then, this paper enhances two entity-based models—toy model and Explicit Semantic Ranking model (ESR)—by considering the importance of entities. In contrast to the existing models, the enhanced models assign the weights of entities according to their importance. Experimental results show that the enhanced toy model and ESR can outperform the two baselines by as much as 4.57% and 2.74% on NDCG@20 respectively, and further experiments reveal that the strength of the enhanced models is more evident on long queries and the queries where ESR fails, confirming the effectiveness of taking the importance of entities into account.

Link prediction is a task predicting whether there is a link between two nodes in a network. Traditional link prediction methods that assume handcrafted features (such as common neighbors) as the link’s formation mechanism are not universal. Other popular methods tend to learn the link’s representation, but they cannot represent the link fully. In this paper, we propose Edge-Nodes Representation Neural Machine (ENRNM), a novel method which can learn abundant topological features from the network as the link’s representation to promote the formation of the link. The ENRNM learns the link’s formation mechanism by combining the representation of edge and the representations of nodes on the two sides of the edge as link’s full representation. To predict the link’s existence, we train a fully connected neural network which can learn meaningful and abundant patterns. We prove that the features of edge and two nodes have the same importance in link’s formation. Comprehensive experiments are conducted on eight networks, experiment results demonstrate that the method ENRNM not only exceeds plenty of state-of-the-art link prediction methods but also performs very well on diverse networks with different structures and characteristics.

The spatial resolution and clarity of remote sensing images are crucial for many applications such as target detection and image classification. In the last several decades, tremendous image restoration tasks have shown great success in ordinary images. However, since remote sensing images are more complex and more blurry than ordinary images, most of the existing methods are not good enough for remote sensing image restoration. To address such problem, we propose a novel method named deep memory connected network (DMCN) based on the convolutional neural network to reconstruct high-quality images. We build local and global memory connections to combine image detail with global information. To further reduce parameters and ease time consumption, we propose Downsampling Units, shrinking the spatial size of feature maps. We verify its capability on two representative applications, Gaussian image denoising and single image super-resolution (SR). DMCN is tested on three remote sensing datasets with various spatial resolution. Experimental results indicate that our method yields promising improvements and better visual performance over the current state-of-the-art. The PSNR and SSIM improvements over the second best method are up to 0.3 dB.

Ship detection plays an important role in automatic remote sensing image interpretation. The scale difference, large aspect ratio of ship, complex remote sensing image background and ship dense parking scene make the detection task difficult. To handle the challenging problems above, we propose a ship rotation detection model based on a Feature Fusion Pyramid Network and deep reinforcement learning (FFPN-RL) in this paper. The detection network can efficiently generate the inclined rectangular box for ship. First, we propose the Feature Fusion Pyramid Network (FFPN) that strengthens the reuse of different scales features, and FFPN can extract the low level location and high level semantic information that has an important impact on multi-scale ship detection and precise location of dense parking ships. Second, in order to get accurate ship angle information, we apply deep reinforcement learning to the inclined ship detection task for the first time. In addition, we put forward prior policy guidance and a long-term training method to train an angle prediction agent constructed through a dueling structure Q network, which is able to iteratively and accurately obtain the ship angle. In addition, we design soft rotation non-maximum suppression to reduce the missed ship detection while suppressing the redundant detection boxes. We carry out detailed experiments on the remote sensing ship image dataset, and the experiments validate that our FFPN-RL ship detection model has efficient detection performance.

Current popular abstractive summarization is based on an attentional encoder-decoder framework. Based on the architecture, the decoder generates a summary according to the full text that often results in the decoder being interfered by some irrelevant information, thereby causing the generated summaries to suffer from low saliency. Besides, we have observed the process of people writing summaries and find that they write a summary based on the necessary information rather than the full text. Thus, in order to enhance the saliency of the abstractive summarization, we propose an attentive information extraction model. It consists of a multi-layer perceptron (MLP) gated unit that pays more attention to the important information of the source text and a similarity module to encourage high similarity between the reference summary and the important information. Before the summary decoder, the MLP and the similarity module work together to extract the important information for the decoder, thus obtaining the skeleton of the source text. This effectively reduces the interference of irrelevant information to the decoder, therefore improving the saliency of the summary. Our proposed model was tested on CNN/Daily Mail and DUC-2004 datasets, and achieved a 42.01 ROUGE-1 f-score and 33.94 ROUGE-1, recall respectively. The result outperforms the state-of-the-art abstractive model on the same dataset. In addition, by subjective human evaluation, the saliency of the generated summaries was further enhanced.

Aircraft type recognition plays an important role in remote sensing image interpretation. Traditional methods suffer from bad generalization performance, while deep learning methods require large amounts of data with type labels, which are quite expensive and time-consuming to obtain. To overcome the aforementioned problems, in this paper, we propose an aircraft type recognition framework based on conditional generative adversarial networks (GANs). First, we design a new method to precisely detect aircrafts’ keypoints, which are used to generate aircraft masks and locate the positions of the aircrafts. Second, a conditional GAN with a region of interest (ROI)-weighted loss function is trained on unlabeled aircraft images and their corresponding masks. Third, an ROI feature extraction method is carefully designed to extract multi-scale features from the GAN in the regions of aircrafts. After that, a linear support vector machine (SVM) classifier is adopted to classify each sample using their features. Benefiting from the GAN, we can learn features which are strong enough to represent aircrafts based on a large unlabeled dataset. Additionally, the ROI-weighted loss function and the ROI feature extraction method make the features more related to the aircrafts rather than the background, which improves the quality of features and increases the recognition accuracy significantly. Thorough experiments were conducted on a challenging dataset, and the results prove the effectiveness of the proposed aircraft type recognition framework.

Automatic question answering (QA), which can greatly facilitate the access to information, is an important task in artificial intelligence. Recent years have witnessed the development of QA methods based on deep learning. However, a great amount of data is needed to train deep neural networks, and it is laborious to annotate training data for factoid QA of new domains or languages. In this paper, a distantly supervised method is proposed to automatically generate QA pairs. Additional efforts are paid to let the generated questions reflect the query interests and expression styles of users by exploring the community QA. Specifically, the generated questions are selected according to the estimated probabilities they are asked. Diverse paraphrases of questions are mined from community QA data, considering that the model trained on monotonous synthetic questions is very sensitive to variants of question expressions. Experimental results show that the model solely trained on generated data via the distant supervision and mined paraphrases could answer real-world questions with the accuracy of 49.34%. When limited annotated training data is available, significant improvements could be achieved by incorporating the generated data. An improvement of 1.35 absolute points is still observed on WebQA, a dataset with large-scale annotated training samples.

Recently, inspired by the power of deep learning, convolution neural networks can produce fantastic images at the pixel level. However, a significant limiting factor for previous approaches is that they focus on some simple datasets such as faces and bedrooms. In this paper, we propose a multiscale deep neural network to transform sketches into Chinese paintings. To synthesize more realistic imagery, we train the generative network by using both L1 loss and adversarial loss. Additionally, users can control the process of the synthesis since the generative network is feed-forward. This network can also be treated as neural style transfer by adding an edge detector. Furthermore, additional experiments on image colorization and image super-resolution demonstrate the universality of our proposed approach.

With the increasing use of mobile GPS (global positioning system) devices, a large volume of trajectory data on users can be produced. In most existing work, trajectories are usually divided into a set of stops and moves. In trajectories, stops represent the most important and meaningful part of the trajectory; there are many data mining methods to extract these locations. DBSCAN (density-based spatial clustering of applications with noise) is a classical density-based algorithm used to find the high-density areas in space, and different derivative methods of this algorithm have been proposed to find the stops in trajectories. However, most of these methods required a manually-set threshold, such as the speed threshold, for each feature variable. In our research, we first defined our new concept of move ability. Second, by introducing the theory of data fields and by taking our new concept of move ability into consideration, we constructed a new, comprehensive, hybrid feature–based, density measurement method which considers temporal and spatial properties. Finally, an improved DBSCAN algorithm was proposed using our new density measurement method. In the Experimental Section, the effectiveness and efficiency of our method is validated against real datasets. When comparing our algorithm with the classical density-based clustering algorithms, our experimental results show the efficiency of the proposed method.