Search

Detecting rotated ships is difficult in optical remote sensing images due to the challenges of complex scenes. Existing advanced rotated ship detectors are typically anchor-based algorithms that require plenty of predefined anchors. However, the use of anchors brings three critical problems: 1) a large number of anchors bring a huge amount of calculation; 2) the attributes (e.g., size and aspect ratios) of anchors are designed via ad hoc heuristics; and 3) only a tiny fraction of anchors that overlap with ground-truth bounding boxes of ships tightly can be considered as positive samples, which causes an extreme imbalance between positive and negative samples. As a result, the detection accuracy will be influenced seriously when the design of anchors is not suitable. To address the above problems, this article proposes a novel anchor-free rotated ship detection framework, called SKNet, which detects rotated ships as keypoints in optical remote sensing images. In SKNet, a ship target is modeled as its center keypoint and morphological sizes, including the width, height, and rotation angle. Accordingly, we design two customized modules: orthogonal pooling and soft-rotate-nonmaximum suppression (NMS), where the former is to improve the prediction accuracy of the center keypoint and the morphological size, and the latter is to effectively remove redundant rotated ship detection results. Extensive experiments are conducted to demonstrate the effectiveness of SKNet on three optical remote sensing image data sets: HRSC2016, DOTA-ship, and HPDM-OSOD, which is collected by ourselves and published in this article. Empirical studies show that SKNet achieves state-of-the-art detection performance while being time-efficient. Overall, SKNet achieves the best speed–accuracy tradeoff.

In dynamic scene deblurring, recent neural network–based methods have been very successful. But with the improvement of deep deblurring performance, network structure and learning become more complicated. Compared with large‐scale network parameters and complex network structures, an iterative multi‐scale residual network to achieve a more effective parameter sharing scheme is proposed. In each iterative unit, fast multi‐scale residual blocks to replace superimposed convolutional layers or classic residual blocks are used. On the basis of preventing model overfitting, the receptive field of the network is increased. At the same time, the gated recurrent unit is introduced to connect modules of different stages. The model does not rely on the estimation of the blur kernel and directly generates sharp images in an end‐to‐end manner. The experimental structure on the benchmark dataset and real‐world images showed that this method has better quality than the existing methods in terms of large‐scale blur and subjective perception effects, both in quantitative and qualitative terms.

With the emergence and development of blockchain technology, due to its innovation in the underlying technology, many potential innovations have been created in the financial aspects of different industries, and even disruptive changes have occurred. Especially mature applications in supply chain finance are more extensive, and at the same time, the technology also effectively promotes the development of finance-related technology. We proposed a blockchain-based framework and used an accessory technology, namely smart contracts. The credit mechanism can be reformed to promote the flow of credit value and make it highly coupled with financial scenarios to obtain the feasibility of supply chain process design.

With the continuous in-depth implementation of the ”Internet +” development strategy, supply chain finance relying on Internet technology is gradually becoming the main way for small and medium-sized enterprises (SMEs) to finance. Supply chain finance is closely integrated with the physical industry and finance, which has greatly promoted the continuous development of the main bodies of the supply chain, especially the small and medium-sized enterprises. Therefore, by integrating the components and characteristics of the blockchain into the various application links of supply chain finance, it can effectively solve the dilemma of SME financing in supply chain finance.

This paper proposes a robust obstacle detection and recognition method for driver assistance systems. Unlike existing methods, our method aims to detect and recognize obstacles on the road rather than all the obstacles in the view. The proposed method involves two stages aiming at an increased quality of the results. The first stage is to locate the positions of obstacles on the road. In order to accurately locate the on-road obstacles, we propose an obstacle detection method based on the U-V disparity map generated from a stereo vision system. The proposed U-V disparity algorithm makes use of the V-disparity map that provides a good representation of the geometric content of the road region to extract the road features, and then detects the on-road obstacles using our proposed realistic U-disparity map that eliminates the foreshortening effects caused by the perspective projection of pinhole imaging. The proposed realistic U-disparity map greatly improves the detection accuracy of the distant obstacles compared with the conventional U-disparity map. Second, the detection results of our proposed U-V disparity algorithm are put into a context-aware Faster-RCNN that combines the interior and contextual features to improve the recognition accuracy of small and occluded obstacles. Specifically, we propose a context-aware module and apply it into the architecture of Faster-RCNN. The experimental results on two public datasets show that our proposed method achieves state-of-the-art performance under various driving conditions.

Drug–drug interactions (DDIs) are crucial for drug research and pharmacovigilance. These interactions may cause adverse drug effects that threaten public health and patient safety. Therefore, the DDIs extraction from biomedical literature has been widely studied and emphasized in modern biomedical research. The previous rules-based and machine learning approaches rely on tedious feature engineering, which is labourious, time-consuming and unsatisfactory. With the development of deep learning technologies, this problem is alleviated by learning feature representations automatically. Here, we review the recent deep learning methods that have been applied to the extraction of DDIs from biomedical literature. We describe each method briefly and compare its performance in the DDI corpus systematically. Next, we summarize the advantages and disadvantages of these deep learning models for this task. Furthermore, we discuss some challenges and future perspectives of DDI extraction via deep learning methods. This review aims to serve as a useful guide for interested researchers to further advance bioinformatics algorithms for DDIs extraction from the literature.

This paper considers a problem that visual servo control for an aerial manipulator removes foreign objects of power transmission lines. A position-based visual servoing (PBVS) combing a foreign objects locating method based on the point cloud with a hierarchical task-priority control method is employed to drive the aerial manipulator to remove the foreign object. Firstly, the RGB-D camera mounted on the drone obtains the point cloud of the environment, and the foreign object will be localized by the detection and localization algorithm. Then, a new visual servo error is proposed to decouple linear speed and angular speed, allowing the aerial manipulator to grasp accurately in the dangerous environment. In addition, the redundant characteristics of the aerial manipulator will be fully used by the hierarchical task priority control scheme. Finally, experimental results of a drone equipped with a 4-DOF delta manipulator removing foreign objects of power transmission line are provided to demonstrate the effectiveness of the control method.

Fill-in-the-blank question generation has become enormously popular and attracted lots of attention recently. However, most of the existing question generation datasets are developed for machine reading comprehension, which are not specifically designed for examination. To fill in the gap, in this paper, we propose a Chinese sentence fill-in-the-blank question generation dataset for examination (named CSFQGD), which will be released to the public 11Resources are available at https://github.com/tianlin668/CSFQGD. The dataset is composed of 20.5K questions from many real examinations in Chinese that cover a wide spectrum of learning subjects. Based on the proposed dataset, we test several well-known methods for fill-in-the-blank question generation and compare their performance. Our baseline study on this dataset shows that CSFQGD is a challenging test bed for further research.

Grasping objects in a large-scale area has been investigated extensively for mobile robots. But for unmanned aerial manipulator(UAM), this is still a challenging task. In order to solve the problem of accurate grasping in a large-scale area, we propose a novel detection and control framework for UAM, which consists of two stages: preliminary localization and precise localization. At the preliminary localization stage, the RGB-D sensor mounted on the end-effector of the manipulator scans to obtain the point cloud surrounding the UAM. By bringing the known target shape and processed point cloud parameters into our proposed loss function, we select the highest priority area as the best potential region proposal, which can help the UAM to screen out the target for precise localization from the obtained point cloud. At precise localization stage, after UAM reaches the best potential region, the RGB-D sensor mounted on the drone uses the deep object pose estimation(DOPE) to estimate the 6D pose of the target. Through independently compensating for the disturbance of the UAV and manipulator, the UAM can accurately grasp the target with the estimated 6D pose. To evaluate the performance of the UAM, we conducted experiments in four different scenes. Experimental results demonstrate that the UAM can grasp the target with an average success rate of 83.4% for the large-scale scene. The above results prove the feasibility and robustness of the framework. The code is available at https://github.com/skywoodsz/CatchIt

Learning effective feature interactions behind user behavior is challenging in credit scoring. Existing machine learning methods seem to have a strong bias towards low-order or high-order interactions, or require expertise feature engineering. In this paper, we present a novel neural network approach AttentionFM, which incorporates Factorization Machines and Attention mechanism for credit scoring. The proposed model focuses more on critical features and emphasizes both low- and high-order feature interactions, with no need of manually feature engineering on raw data representation. Experimental results demonstrate that our proposed model significantly outperforms the baselines based on two public datasets.

Neural Networks have achieved great success in many computer vision tasks, especially in image recognition. However, as neural networks grow deeper and deeper, to some extend, we've found them becoming difficult to train, and requiring samples in large scale dramatically, even with the help of Dropout and Dropconnect methods, which do improve the accuracy a bit but burdens the training process as a sacrifice. To overcome this, we propose a novel neuron connection model to generate dynamic graphs of computation. As synapses have two kinds: excitatory and inhibitory ones, our model also has two kinds of connections for neurons. In addition, we propose a training algorithm that deals with non-differentiable because the equations of the connections and activation function of neurons in our model are not really differentiable. To evaluate the effectiveness the proposed method, we apply it to the image recognition task, and the results show that our proposed model achieves state-of-the-art performance on three public datasets: MNIST, CIFAR-10, and CIFAR-100.

Suicide is one of the leading causes of death worldwide. At the same time, the widespread use of social media has led to an increase in people posting their suicide notes online. Therefore, designing a learning model that can aid the detection of suicide notes online is of great importance. However, current methods cannot capture both local and global semantic features. In this paper, we propose a transformer-based model named TransformerRNN, which can effectively extract contextual and long-term dependency information by using a transformer encoder and a Bi-directional Long Short-Term Memory (BiLSTM) structure. We evaluate our model with baseline approaches on a dataset collected from online sources (including 659 suicide notes, 431 last statements, and 2000 neutral posts). Our proposed TransformerRNN achieves 95.0%, 94.9% and 94.9% performance in P, R and F1-score metrics respectively and therefore outperforms comparable machine learning and state-of-the-art deep learning models. The proposed model is effective for classifying suicide notes, which in turn, may help to develop suicide prevention technologies for social media., Highlights • Early suicide note identification may help to develop suicide prevention technologies for social media. • TranformerRNN was proposed, using the transformer encoder and BiLSTM. • TranformerRNN can extract contextual information and latent features to identify suicide notes. • Our TranformerRNN method outperforms comparable machine learning and state-of-the-art deep learning models.

Convolutional neural networks have been shown successful in extracting features from images and texts. However, it is difficult to apply convolutional neural networks directly on ubiquitous graph data since the graph data lies in an irregular structure. A significant number of researchers engrossed themselves in studying graph convolutional networks transformed from Euclidean domain. Previous graph convolutional networks overviews mainly focus on reviewing recent methods in a comprehensive ways. In this survey, we review the convolutional networks from the perspective of receptive fields. Roughly, the convolutional networks fall into three main categories: spectral based methods, sampling based methods and attention based methods. We analysis the differences of these methods and propose three potential directions for future research of graph convolutional networks.

Traffic congestion seriously affect citizens’ life quality. Many researchers have paid much attention to the task of short-term traffic congestion forecasting. However, the performance of the traditional traffic congestion forecasting approaches is not satisfactory. Moreover, most neural network models cannot capture the features at different moments effectively. In this paper, we propose an Attention-based long short-term memory (LSTM) recurrent neural network. We evaluate the prediction architecture on a real-time traffic data from Gray-Chicago-Milwaukee (GCM) Transportation Corridor in Chicagoland. The experimental results demonstrate that our method outperforms the baselines for the task of congestion prediction.

Road passage time is an important measure of urban traffic. Accurate estimation of road passage time contributes to the route programming and the urban traffic planning. Currently, the estimation of road passage time for a particular road is usually based on its historical data which is simple to express the general law of road traffic. However, with the increase of the number of roads in the urban area, the connection between roads becomes more complex. The existing methods fail to make use of the connection between different roads and the road passage time, merely based on its own historical data. In this paper, we propose a road passage time estimating model, called “CA-RPT”, which utilizes the contextual information between road connections as well as the date and time period. We evaluate our method based on a real geolocation information data set collected by mobile APP anonymously. The results demonstrate that our method is more accurate than the state-of-the-art methods.

Question Generation is a rising research field of artificial intelligence in education and staff training. It is an effective strategy to conduct knowledge evaluation and performance appraisal. How to generate a variety of good quality questions has been an essential issue. In this paper, we propose an automatic method that generates Chinese multiple-type question by combing diverse knowledge bases and similar language features. Our research is mainly focused on Tradition Chinese Medicine (TCM) domain because it contains many Chinese-based resources and complex relationships. But the method is flexible and can be easily applied to other domains. Experimental results demonstrate that the proposed method obtain a good result.

Current state-of-the-art detectors typically classify candidate proposals using their interior features. However, the valuable contexts are not fully exploited by existing detectors yet, which limits the detection performance. In this paper, we present a context learning network (CLN), which aims to capture pairwise relation between objects and global contexts of each object. The proposed CLN consists of two subnetworks: a multi-layer perceptron (MLP) with three layers and a convolutional neural network (ConvNet) with two layers. The MLP is first designed to capture the pairwise relation context. Pairwise relationship context is then gathered and concatenated to further learn the global contexts by the ConvNet. Finally, we obtain the desired context feature maps with rich contextual information that are useful for accurate object detection. The proposed CLN is lightweight and it is easy to embed in any existing networks for object detection. In this paper, we present a context-aware Faster-RCNN with the proposed CLN and conduct extensive experiments to evaluate its performance. Experimental results demonstrate that the context-aware Faster-RCNN achieves state-of-the-art performance with the 82.1%, 80.7% and 38.4%mAPs on VOC 2007, VOC 2012 and COCO datasets, respectively.

A small variation in mobile hardware and software can potentially cause a significant heterogeneity or variation in the sensor data each device collects. For example, the microphone and accelerometer sensors on different devices can respond very differently to the same audio or motion phenomena. Other factors, like the instantaneous computational load on a smartphone, can cause key behavior like sensor sampling rates to fluctuate, further polluting the data. When sensing devices are deployed in unconstrained and real-world conditions, examples of sharply lower classification accuracy are observed due to what is collectively known as the sensing system heterogeneity. In this work, we take an unconventional approach and argue against solving individual forms of heterogeneity, e.g., improving OS behavior, or the quality/uniformity of components. Instead, we propose and build classifiers that themselves are more tolerant of these variations by leveraging deep learning and a data-augmented training process. Neither augmentation nor deep learning has previously been attempted to cope with sensor heterogeneity. We systematically investigate how these two machine learning methodologies can be adapted to solve such problems, and identify when and where they are able to be successful. We find that our proposed approach is able to reduce classifier errors on an average by 9% and 17% for a range of inertial- and audio-based mobile classification tasks.

As domain terminology plays a crucial role in the study of every domain, automatic domain terminology extraction method is in real demand. In this paper, we propose a novel parsing-based method which generates the domain compound terms by utilizing the dependent relations between the words. Dependency parsing is used to identify the dependent relations. In addition, a multi-factor evaluator is proposed to evaluate the significance of each candidate term which not only considers frequency but also includes the influence of other factors affecting domain terminology. Experimental results demonstrate that the proposed domain terminology extraction method outperforms the traditional POS-base method in both precision and recall.