Video captioning aims to automatically generate natural language descriptions of video content, which has drawn a lot attention recent years. Generating accurate and fine-grained captions needs not only understand the global content video, but also capture detailed object information. Meanwhile, representations have great impact on quality generated captions. Thus, it is important for salient objects with their temporal dynamics, represent them using discriminative spatio-temporal...

In recent years, the rapid expansion of subway construction has brought increasing challenges related to crossing new and existing lines. This study focuses on Nanjing Metro line 11 project, where crosses 3. A numerical simulation analysis tunnel intersection area is conducted using ABAQUS software investigate deformation mechanism shield segments when a constructed at close distance oblique angle an tunnel. During tunnel, experience greatest settlement point, with pattern gradually evolving...

Video classification is highly important and has widespread applications, such as video search intelligent surveillance. naturally contains both static motion information, which can be represented by frames optical flow, respectively. Recently, researchers have generally adopted deep networks to capture the information separately, two main limitations. First, coexistence relationship between spatial temporal attention ignored, although they should jointly modeled evolutions of learn...

The interferometry technique is commonly used to obtain the phase information of an object in optical metrology. obtained wrapped subject a 2π ambiguity. To remove ambiguity and correct phase, unwrapping essential. Conventional approaches are time-consuming noise sensitive. address those issues, we propose new approach, where transfer task into multi-class classification problem introduce efficient segmentation network identify classes. Moreover, noise-to-noise denoised integrated preprocess...

Division of focal plane (DoFP) polarimeters operate by integrating micro-polarizer elements with a plane. These polarization imaging sensors reduce spatial resolution output and each pixel has varying instantaneous field view (IFoV). drawbacks can be mitigated applying proper interpolation methods. In this paper, we present new method for DoFP using intensity correlation. We employ the correlation measurements in different orientations to detect edges then implement along edges. The...

Spaceborne synthetic aperture radar (SAR) real-time imaging is especially important for disaster emergencies and monitoring applications with highly desired requirements. Therefore, the continuous improvement of efficiency an development trend. At present, traditional algorithms based on constant pulse repetition frequency (PRF) have low accuracy when processing spaceborne SAR data nonlinear trajectory. For this problem, existing methods usually introduce some complex signal steps, such as...

Image fusion is the key step to improve performance of object detection in polarization images. We propose an unsupervised deep network address image issue. The learns end-to-end mapping for fused images from intensity and degree linear images, without ground truth Customized architecture loss function are designed boost performance. Experimental results show that our proposed outperforms other state-of-the-art methods terms visual quality quantitative measurement.

We propose a polarization demosaicing convolutional neural network to address the image issue, last unsolved issue in microgrid polarimeters. This learns an end-to-end mapping between mosaic images and full-resolution ones. Skip connections customized loss function are used boost performance. Experimental results show that our proposed outperforms other state-of-the-art methods by large margin terms of quantitative measures visual quality.

Division of focal plane (DoFP) polarization image sensors capture properties light at every imaging frame. However, these only partial information, resulting in reduced spatial resolution output and a varying instantaneous field overview (IFoV). Interpolation methods are used to reduce the drawbacks recover missing information. In this paper, we propose residual interpolation as an alternative normal for division sensors, where is difference between observed tentatively estimated pixel...

Abstract Background Elymus sibiricus is an ecologically and economically important perennial, self-pollinated, allotetraploid (StStHH) grass, widely used for forage production animal husbandry in Western Northern China. However, it has low seed yield mainly caused by shattering, which makes difficult this species. The goals of study were to construct the high-density genetic linkage map, identify QTLs candidate genes seed-yield related traits. Results An F 2 mapping population 200...

Video captioning is a significant challenging task in computer vision and natural language processing, aiming to automatically describe video content by sentences. Comprehensive understanding of the key for accurate captioning, which needs not only capture global salient objects video, but also understand spatio-temporal relations objects, including their temporal trajectories spatial relationships. Thus, it important objects' relationships both within across frames. Therefore, this paper,...

Benign prostatic hyperplasia (BPH) is a common disease in elderly men, mainly resulted from an imbalance between cell proliferation and death. Glutathione peroxidase 3 (GPX3) was one of the differentially expressed genes BPH identified by transcriptome sequencing 5 hyperplastic normal prostate specimens, which had not been elucidated prostate. This study aimed to ascertain mechanism GPX3 involved proliferation, apoptosis, autophagy ferroptosis BPH.Human tissues, silencing overexpression...

Elymus L. is the largest genus in tribe Triticeae Dumort., encompassing approximately 150 polyploid perennial species widely distributed temperate regions of world. It considered to be an important gene pool for improving cereal crops. However, a shortage molecular marker limits efficiency and accuracy genetic breeding species. High-throughput transcriptome sequencing data essential discovery development.We obtained dataset E. sibiricus, type Elymus, identified total 8871 putative EST-SSRs...

Abstract The photocatalytic reduction of CO 2 is a promising strategy to generate chemical fuels. However, this reaction usually suffers from low photoactivity because insufficient light absorption and rapid charge recombination. Defect engineering has become an effective approach improve the activity. Herein, ultra-thin (~ 4.1 nm) carbon-doped Bi WO 6 nanosheets were prepared via hydrothermal treatment followed by calcination. nanosheet structure catalyst not only provides more active sites...

To address color polarization demosaicking problems in imaging with a camera, we propose convolutional neural network (CPDCNN), which has two-branch structure to ensure the fidelity of signatures and enhance image resolution. train network, built unique dual-camera system captured pairwise dataset. Experimental results show that CPDCNN outperformances other methods by large margin contrast

Division of focal plane (DoFP) polarimeters are composed interlaced linear polarizers overlaid upon a array sensor. The interpolation is essential to reconstruct polarization information. However, current methods based on the unrealistic assumption noise-free images. Thus, it advantageous carry out denoising before interpolation. In this paper, we propose principle component analysis (PCA) method, which works directly DoFP Both simulated and real images used evaluate performance....

To address the key image interpolation issue in microgrid polarimeters, we propose a machine learning model based on sparse representation. The sparsity and non-local self-similarity priors are used as regularization terms to enhance stability of an model. Moreover, make best correlation among different polarization orientations, patches channels joined learn adaptive sub-dictionary. Synthetic real images evaluate interpolated performance. experimental results demonstrate that our proposed...

We propose a deep learning-based restoration method to remove honeycomb patterns and improve resolution for fiber bundle (FB) images. By building calibrating dual-sensor imaging system, we capture FB images corresponding ground truth data train the network. Images without fixed are restored from raw as direct inputs, spatial is significantly enhanced trained sample type. also construct brightness mapping between two image types effective use of all data, providing ability output expected...

Phase unwrapping is a very important step in fringe projection 3D imaging. In this paper, we propose new neural network for accurate phase to address the special needs Instead of labeling wrapped with integers directly, two-step training process same configuration proposed. first step, (network I) trained label only four key features phase. second another II) segments. The advantages are that dimension can be much larger from data, and serious Gaussian noise correctly unwrapped. We...