A5080483119- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
- Spectroscopy and Chemometric Analyses
- Video Surveillance and Tracking Methods
- Robotics and Sensor-Based Localization
- Advanced Image Processing Techniques
- Advanced Vision and Imaging
- Domain Adaptation and Few-Shot Learning
- Analytical Chemistry and Chromatography
- Neural dynamics and brain function
- Advanced Image Fusion Techniques
- Image Retrieval and Classification Techniques
- Fractal and DNA sequence analysis
- Image and Signal Denoising Methods
- Toxic Organic Pollutants Impact
- Biometric Identification and Security
- Functional Brain Connectivity Studies
- Advanced Algorithms and Applications
- Remote Sensing and LiDAR Applications
- Image Processing Techniques and Applications
- Phytochemicals and Antioxidant Activities
- Spectroscopy Techniques in Biomedical and Chemical Research
- Advanced Chemical Sensor Technologies
- Remote Sensing and Land Use
- 3D Surveying and Cultural Heritage
China Jiliang University
2024
Zhoukou Normal University
2008-2022
Northwest University
2022
Chengdu Technological University
2022
Qingdao University of Technology
2022
Sichuan University of Science and Engineering
2021-2022
Northeastern University
2009-2021
Beijing Information Science & Technology University
2021
Xi'an Jiaotong University
2019-2020
University of Hong Kong
2014-2019
Video object detection has drawn great attention recently. The Vision Meets Drone Object Detection in Challenge 2019 (VisDrone-VID2019) is held to advance the state-of-the-art video for videos captured by drones. Specifically, there are 13 teams participating challenge. We also report results of 6 detectors on collected dataset. A short description provided appendix each detector. present analysis and discussion challenge results. Both dataset publicly available at website: http://www.aiskyeye.com/.
This paper proposes a novel point-cloud-based place recognition system that adopts deep learning approach for feature extraction. By using convolutional neural network pre-trained on color images to extract features from range image without fine-tuning extra images, significant improvement has been observed when compared hand-crafted features. The resulting is illumination invariant, rotation invariant and robust against moving objects are unrelated the identity. Apart itself, we also bring...
Moving objects can greatly jeopardize the performance of a visual simultaneous localization and mapping (vSLAM) system which relies on static-world assumption. Motion removal have seen successful solving this problem. Two main streams solutions are based either geometry constraints or deep semantic segmentation neural network. The former rely static majority assumption, latter require labor-intensive pixel-wise annotations. In paper we propose to adopt novel weakly-supervised method. mask is...
Abstract BACKGROUND The alteration of chromatin accessibility plays an important role in plant responses to abiotic stress. Carbon‐based nanomaterials (CBNMs) have attracted increasing interest agriculture due their potential impact on crop productivity, showcasing effects biological processes at transcriptional levels; however, remains unknown. RESULTS This study found that fullerenol can penetrate the seed coat pea mitigate reduction germination caused by osmotic RNA sequencing (RNA‐seq)...
The moisture content and distribution of Daqu significantly influences the quality products. This work presents visualization in using a combination spectral spatial information from hyperspectral imaging. least squares support vector machine (LS-SVM) partial regression (PLSR) methods were adopted to establish predictive models based on full wavelengths 29 feature combined with color features, respectively. best prediction model was PLSR (Rp2=0.9823, RMSEP=0.0109) wavelengths. results showed...
Abstract Cereal crops play an important role in preventing chronic diseases and regulating human functions due to their rich phytochemicals. However, increasing number of cases cereal‐crop adulteration are occurring, affecting the nutrition food threatening safety. Therefore, this study used hyperspectral imaging (HSI) a back‐propagation neural network (BPNN) establish variety identification model different grain shapes, colors, sizes. This can conduct quantitative qualitative analysis...
The drum water level of industrial boiler is very critical as one the monitoring parameters. intelligence great significance for traditional system with local bicolor gauge and remote TV. In this article, Wiener filtering algorithm first used to process image morphology applied segment extract information. Then, recognized digital value obtained. Furthermore, a video measurement designed based on CCD DSP.
This paper aims to verify the effectiveness of parabolic approximate models for solid phase diffusion process under batteries actual discharge conditions. Choosing 1C, 4C constant current and FUDS conditions, two-parameter, three-parameter model theory numerical algorithm are conducted simulate surface concentration curve in Maple environment after obtaining reaction ion flux density jp particle at separator/cathode interface by COMSOL. The result shows that is effective when applied...
At present, a person-specific 3D model generation technology has been an active research topic for scientists in the field of computer vision and graphics. The paper proposes modeling method based on single photograph, by analysis individual difference like geometric features which present shape locations facial components. An extension basic Active Appearance Model is proposed to localize some feature points frontal image all these are aligned fit generic face specialized one reflect given...
Most of the face hallucination methods are designed for complete inputs. They will not work well if inputs very tiny or contaminated by large occlusion. Inspired this fact, we propose an obscured network(OFHNet). The OFHNet consists four parts: inpainting network, upsampling a discriminative and fixed facial landmark detection network. network restores low-resolution(LR) images. following is to upsample output In order ensure generated high-resolution(HR) images more photo-realistic, utilize...
In this paper, the correlation analysis of EEG signals and ECG between epileptic normal population was carried out based on improved synchronization algorithm IRC Kendall. clinical animal experiments, it found that when epilepsy sample attacks, irregular heart rate changes often occur. experiment, an used to analyze study whether patients different from people. Experimental results show using Kendall improvement can significantly distinguish in left forehead anterior temporal regions...
This paper proposes a novel weakly-supervised semantic segmentation method using image-level label only. The class-specific activation maps from the well-trained classifiers are used as cues to train network. well-known defects of these coarseness and incompleteness. We use super-pixel refine them, fuse extracted both color image trained classifier gray compensate for their conditional random field is adapted regulate training process outputs further. Besides initializing network, previously...
Film grain noise (FGN) is generated by the procedure of capturing pictures using photographic film. Images with FGN are subjectively pleasing. However, difficult to compress and its pleasant features preserve when images resized. So in literature, extracted first, then regenerated for processed noise-free images. In this paper a new method proposed generate FGN. contrast some other models which spatial domain, our captures statistic feature frequency domain. signal dependent independent...
One-bit transform (1BT), followed by binary motion estimation, is an effective alternative for accelerating traditional 8-bit estimation (ME) in video coding. The underlining assumption the design of 1BT methods that natural videos contain noise. For screen content videos, however, special characteristics (e.g. typically noise-free) can be exploited to further improve accuracy. In this paper we propose a ME method which specifically designed videos. particular, performed on selected...
At present, in the large-scale book management scene, sorting, daily maintenance and retrieval are very common, but information is complicated efficiency of relying on manual extremely poor. Although there have been many self-service systems based optics or vision, they mostly traditional computer vision algorithms such as boundary extraction. Due to fact that more artificial experience thresholds, some shortcomings low detection accuracy, poor robustness, inability systematically deploy a...
High resolution (HR) magnetic resonance images (MRI) or computed tomography (CT) can provide clearer anatomical details of human body, which facilitates early diagnosis the diseases. However, due to imaging system, environment and factors, it is difficult obtain clear high-resolution images. In this paper, we proposed a novel medical image super (SR) reconstruction method via multi-scale information distillation (MSID) network in non-subsampled shearlet transform (NSST) domain, namely...