Search

Convolutional neural networks (CNNs) have been achieving remarkable results in medical image segmentation. However, for accurate segmentation of subcortical brain structure in MR images, it is still a challenge due to the ambiguous boundaries, the complex structures, and the various shapes, which limits their clinical application. In this paper, we focus on utilizing multi-scale image contexts and attention mechanisms to improve networks’ ability to learn discriminative feature representation for accurate segmentation and present a novel FCNN architecture called multis-scale guided squeeze-and-excitation network (MSGSE-Net). In particular, we first propose the multi-scale guided squeeze-and-excitation (MSGSE) attention module which can progressively and selectively aggregate discriminative features. In contrast to existing attention modules, the MSGSE module performs an adaptive recalibration that features at different locations of the feature map are recalibrated under the guidance of multi-scale contexts. Then multi-scale spatial attention supervision is adopted to enhance the intra-class homogeneity and inter-class distinction of the attention weights. Moreover, we propose a novel entropy-weighted Dice loss (EDL) to force the network to focus on the ambiguous voxels around the boundaries of subcortical structures. We evaluate the proposed method on two challenging benchmark datasets (the IBSR dataset and the MALC dataset). The experimental results show that our model consistently yields better segmentation performance than several state-of-the-art methods and improves the segmentation Dice score by 1.6% at most compared with baseline method U-Net. Our code is available at https://github.com/neulxlx/MSGSE-Net .

Efforts to understand the complex, multidimensional nature of environmental vulnerability can generate new knowledge by deploying a convergence research framework within a community-engaged approac...

Nowadays, news recommendation has become a popular channel for users to access news of their interests. How to represent rich textual contents of news and precisely match users' interests and candidate news lies in the core of news recommendation. However, existing recommendation methods merely learn textual representations from in-domain news data, which limits their generalization ability to new news that are common in cold-start scenarios. Meanwhile, many of these methods represent each user by aggregating the historically browsed news into a single vector and then compute the matching score with the candidate news vector, which may lose the low-level matching signals. In this paper, we explore the use of the successful BERT pre-training technique in NLP for news recommendation and propose a BERT-based user-news matching model, called UNBERT. In contrast to existing research, our UNBERT model not only leverages the pre-trained model with rich language knowledge to enhance textual representation, but also captures multi-grained user-news matching signals at both word-level and news-level. Extensive experiments on the Microsoft News Dataset (MIND) demonstrate that our approach constantly outperforms the state-of-the-art methods.

Personalized news recommendation is a critical technology to help users find interested news, and how to precisely match users' interests and candidate news lies in the core of news recommendation. Existing studies generally learn user's interest vector by aggregating his/her browsed news and then match it with the candidate news vector, which may lose the textual semantic matching signals for recommendation. In this paper, we propose an Attentive Multi-field Matching (AMM) framework for news recommendation which captures the semantic matching representations between each browsed news and candidate news, and then aggregates them as final user-news matching signal. In addition, our method incorporates multi-field information and designs a within-field and cross-field matching mechanism, which leverages complementary information from different fields (e.g., titles, abstracts and bodies) and obtain the multi-field matching representations. To achieve a comprehensive semantic understanding, we employ the most popular language model BERT to learn the matching representation of each browsed-candidate news pair, and incorporate the attention mechanism in aggregating procedure to characterize the importance of each matching representation for the final user-news matching signal. Experiments on the real world datasets validate the effectiveness of AMM.

As a representative product of the sharing economy era and a powerful supplement to public transportation shared cars have the characteristics of convenience, efficiency, environmental protection, and green travel, and to a certain extent alleviate the contradiction between supply and demand, and solve the problem of long-term idle vehicles and overloaded operation of roads problems. But the uneven distribution of shared cars, the coexistence of no cars, and empty seats will happen. To solve the above problems, this article first analyzes data outliers, data missing values, and data standardization processing on the attached data, and then builds a BP neural network demand prediction model to obtain the distribution of shared car usage in the city.

Automatic subcortical brain segmentation in a magnetic resonance image (MRI) is crucial to disease diagnosis and various clinical applications. In this paper, we proposed an efficient segmentation method based on atlas registration (AR) and linearized kernel sparse representative classifier (SRC) to segment thalamus, putamen, caudate, pallidum, hippocampus, and amygdala. Many multi-atlas-based segmentation methods (MAS) have been proposed with success. However, the optimization problems in MAS for each voxel cause heavy computational burden. Besides, brain structures in the MR image have the characteristics of small volume, large morphological difference, and blur edge, which make the precise segmentation more difficult. To address these challenges, in the first step of our proposed method, we used AR to estimate subcortical locations in MRI. We constructed a probabilistic atlas for deep structures by AR and then obtained coarse results with initial shapes of structures. Candidate boundary regions were obtained in this stage. Then, in the second step, the kernel SRC was linearized and was applied to refine results around boundaries. The labels of voxels in candidate boundary regions were predicted in this stage, and this refinement was utilized to smooth the boundary. Finally, combining initial shapes and refined boundaries, we obtained a final segmentation of the subcortical brain. The experiments were conducted on IBSR, LPBA40, and SATA MICCAI 2013 challenge datasets. The results showed that our method outperformed other methods including AR, linearized SRC, and deep learning models with mean DSC of 0.843, 0.83, and 0.827, respectively. The time-consumption was relatively lower than the comparison methods. The proposed method has great potential for other segmentation tasks.

Reliable representations of global urban extent remain limited, hindering scientific progress across a range of disciplines that study functionality of sustainable cities. We present an efficient and low-cost machine-learning approach for pixel-based image classification of built-up areas at a large geographic scale using Landsat data. Our methodology combines nighttime-lights data and Landsat 8 and overcomes the lack of extensive ground-reference data. We demonstrate the effectiveness of our methodology, which is implemented in Google Earth Engine, through the development of accurate 30 m resolution maps that characterize built-up land cover in three geographically diverse countries: India, Mexico, and the US. Our approach highlights the usefulness of data fusion techniques for studying the built environment and is a first step towards the creation of an accurate global-scale map of urban land cover over time.

Automatic segmentation of hippocampus in MRI plays an important role in computer aided diagnosis, which is helpful for early treatments of many diseases such as Alzheimers disease and epilepsy. In this paper, a coarse to fine hippocampus segmentation method is proposed. In coarse segmentation stage, atlas registration is applied to construct probabilistic atlases. In fine segmentation stage, atlas label images and coarse segmentation results are applied to delineate ROI. Furthermore, training samples extracted from ROI are used to train the broad learning model. Experimental results show that the proposed method is competitive with deep learning. The average Dice value achieves 0.83 on hippocampus segmentation, which indicates the superiority of the proposed method.

Vehicle re-identification is an important part of intelligent transportation. Although much work has been done on this subject in recent years, vehicle re-identification is still a challenging task due to its obvious illumination change, high similarity between inter-class and great changes under different views. As discriminatory local areas and vehicle view information is the key to improving the above issues, it is desirable to create a model which considers both local details and cross-view situation. In this paper, we built a vehicle re-identification framework based on unsupervised local area detection and view discrimination. First, we combine the convolution features of multiple vehicle images by channel to generate joint channel features, and constructs a discriminating region detector unsupervised by clustering the channel covariance vectors generated between the joint channel features. In the next stage, the irregular shape detection results are converted to rectangular discriminative region by using a novel local region integration algorithm. These rectangular discriminative regions are fed into a multi-branch network to extract the local–global features which contains rich detail information. Furthermore, we generate the view features by regarding the detected discriminative areas as keypoints and construct the unsupervised view discriminator. By using the view information of the vehicle, we designed a view-discrimination based reranking algorithm, which effectively reduces the error rate of identification due to view variant. In order to prove the validity of our method, we have done extensive experiments on VehicleID and VERI-Wild dataset. Experimental results show that our method is superior to other existing vehicle re-identification algorithms.

This paper deals with the multi-degree cyclic robotic cell scheduling problem with multiple parallel flow shop lines. The objective is to find the optimal sequence of robot operations aimed at maximizing the throughput. Considering our practical scenario, a robotic cell with a single dual-gripper robot and reentrant workstations is taken into consideration. We present a mixed-integer linear programming model for this problem and propose a hybrid particle swarm optimization algorithm (HPSO) with local search operators. Computational experiments are conducted by comparing with other two meta-heuristic algorithms. The results validate the effectiveness of the proposed algorithm. Based on the numerical analysis, we obtain the optimal assignment of parallel lines under the same operating conditions and help manufacturers achieve a good balance between processing capacity of workstations and robot serving capability.

In order to deal with the dynamic production environment with frequent fluctuation of processing time, robotic cell needs an efficient scheduling strategy which meets the real-time requirements. This paper proposes an adaptive scheduling method based on pattern classification algorithm to guide the online scheduling process. The method obtains the scheduling knowledge of manufacturing system from the production data and establishes an adaptive scheduler, which can adjust the scheduling rules according to the current production status. In the process of establishing scheduler, how to choose essential attributes is the main difficulty. In order to solve the low performance and low efficiency problem of embedded feature selection method, based on the application of Extreme Gradient Boosting model (XGBoost) to obtain the adaptive scheduler, an improved hybrid optimization algorithm which integrates Gini impurity of XGBoost model into Particle Swarm Optimization (PSO) is employed to acquire the optimal subset of features. The results based on simulated robotic cell system show that the proposed PSO-XGBoost algorithm outperforms existing pattern classification algorithms and the newly learned adaptive model can improve the basic dispatching rules. At the same time, it can meet the demand of real-time scheduling.