Search

Accurate prediction of short-term origin-destination (OD) matrix is crucial for operations in metro systems. Recently, some deep learning-based models have been proposed for OD matrix forecasting in ride-hailing or high-way scenarios. However, the metro OD matrix forecasting receives less attention, and it has different prior knowledge and complex spatiotemporal contextual setting; for example, the sparse destination distribution and the incomplete OD matrices collection in recent time intervals due to unfinished trips before the predicted time interval. This paper designs a deep learning approach for metro OD matrix prediction by addressing the recent destination distribution availability, augmenting the flow presentation for each station, and digging out the global spatial dependency and multiple temporal scale correlations in the mobility patterns of metro passengers. Specifically, it first proposes to complete the recent OD matrices by combining some empirical knowledge including the historical mobility pattern and arrival time distribution. Then, it learns the complementary spatiotemporal contextual features by embedding methods to enrich the station representation. Finally, it captures global mobility trend of metro passengers at each origin station through aggregating the trend of all other origin stations by self-attention mechanism since the mobility synchronizes among stations from spatial perspective. Three temporal convolutional networks are leveraged to extract three temporal trends in passenger mobility data, i.e., recent trend, daily trend, and weekly trend. Smart card data from Shenzhen and Hangzhou metro systems are utilized to demonstrate the superiority of our model over other competitors. [ABSTRACT FROM AUTHOR]

Segmentation of biomedical images is the method of semiautomatic and automatic detection of boundaries within 2D and 3D images. The major challenge of medical image segmentation is the high variability of shape, location, size and texture of the medical images. Manual segmentation is a time-consuming and monotonous process; therefore, a fully automated segmentation process is highly desirable. UNet is deployed as one of the most popular and generic architectures for medical image segmentation. This paper proposes and deploys two variants based on UNet architecture, namely 2DGA-UNet and 3DGA-UNet for the segmentation of 2D and 3D medical images, respectively. The first variant increases the impact of the 2DGA-UNet framework performance by applying transfer learning with the UNet architecture. We use simple convolutional neural networks from the VGG family known as VGG 16 as an encoder in the 2DGA-UNet network. The critical concept of 3DGA-UNet is to supplement a comprehensive contracting network by successive layers, where upsampling operators replace pooling operators. Such layers, therefore, improve the resolution of the output and further be trained end-to-end from very few images, outperforming the state-of-the-art methods. The proposed models are evaluated for 2D and 3D medical images on five benchmark datasets including brain tumor segmentation (BRATS 2018 and BRATS 2019), brain lesion segmentation (MICCAI 2008 multiple sclerosis challenge), lung segmentation (NIH tuberculosis chest X-ray dataset, Shenzhen No. 3 Hospital X-ray set, RSNA pneumonia detection challenge), liver segmentation (3D-IRCADb-01 database). The comprehensive results show remarkable performance considering 14 different evaluation parameters for the segmentation of medical images. Besides, the GA-UNet outperforms traditional methods in terms of ACC, i.e., 97.0% and the DSC of 91.8%. [ABSTRACT FROM AUTHOR]