A5102983151- Advanced Measurement and Detection Methods
- Optical Systems and Laser Technology
- Infrared Target Detection Methodologies
- Advanced Computational Techniques and Applications
- Lightning and Electromagnetic Phenomena
- Remote Sensing and Land Use
- Human Pose and Action Recognition
- Image Processing Techniques and Applications
- Metal and Thin Film Mechanics
- Tribology and Wear Analysis
- Metal Alloys Wear and Properties
- Icing and De-icing Technologies
- Gastrointestinal Bleeding Diagnosis and Treatment
- AI and Multimedia in Education
- Image Enhancement Techniques
- Transportation Safety and Impact Analysis
- Multimodal Machine Learning Applications
- Advanced Decision-Making Techniques
- Metallurgy and Material Forming
- Video Analysis and Summarization
- AI in cancer detection
- Advanced Image Fusion Techniques
- Colorectal Cancer Screening and Detection
- Tribology and Lubrication Engineering
- Mechanical stress and fatigue analysis
Shaanxi Normal University
2020-2022
House of Representatives
2013
Southwest University
2010
Xi'an Jiaotong University
2009
Wuxi Industrial Equipment Installation (China)
2006
The number of mitotic cells present in histopathological slides is an important predictor tumor proliferation the diagnosis breast cancer. However, current approaches can hardly perform precise pixel-level prediction for mitosis datasets with only weak labels (i.e., provide centroid location cells), and take no account large domain gap across from different pathology laboratories. In this work, we propose a Domain adaptive Box-supervised Instance segmentation Network (DBIN) to address above...
In the recent past, Wireless endoscopy (WE) helps physicians to study digestive tract at cost of a wide range images without surgery. The major challenge arises from complication robust image characterization in computer-aided WE pictorial diagnostics. purpose research is provide biased definition and enable medical professionals automatically identify polyp images. this paper, learning feature approach called Deep Stacked Auto Encoder with Constraint Image (DSAECI) for recognizing polyps...
Using lightning surges generated by impulse voltage generator which conform to the International Standard IEC,we made an experiment of strikes on buildings' plastic-steel windows and doors for studying diffusing mechanism side struck exterior wall. By combining this diffusion with manufacture installation technology windows, we found a fundamental principle in protection: connection metal components or between protection devices must be equipotential connection. In addition, construction...
Because of the huge structure and special position, bridge is vulnerable to lightning stroke. As most important bearing member cable supported bridge, once suffering stroke, structural safety will be affected. The used actually was sampled carried out stroke simulation experiment in laboratory, influence stoke on analyzed through five experimental models, truth that can cause high-tensile steel wire melting loss discovered. tensile strength change wires before after test, result shows...
In allusion to the security of people in a straddle-type monorail vehicle when is struck by lightning, this article carries out calculation and analysis two cases, i.e. are not contact with metallic components which directly connecting housing vehicle, concludes that former case, completely safe; latter will feel tingle or causalgia, but do threaten safe life.
SCIT is a classic algorithm in storm identification, tracking and forecasting. In this paper, we try to employ disjointed set ADT improve the efficiency of identification firstly, then performance forecasting strategy by optical flow methodology.
深度学习在检测领域高速发展,但受限于训练数据和计算效率,在基于嵌入式平台的边缘计算领域,尤其是实时跟踪应用中深度学习的智能化算法应用并不广泛。针对这一现象,同时为满足现阶段国产化、智能化的技术需求,提出了一种改进的孪生网络深度学习跟踪算法。在特征网络加入微调网络,解决了网络模型无法在线更新的问题,提升了跟踪的准确性;在IoUNet损失函数中加入中心距离惩罚项,解决了IoUNet当IoU相同时位置跳跃,存在收敛盲区和收敛速度慢的问题;将训练后的网络通过通道剪枝,缩减网络模型尺寸,提升了模型加载和运行的速度。在华为Atlas200NPU平台上实现了实时运行,算法准确率高达0.90(IoU>0.7),帧率达到66 Hz。
针对红外与可见光图像融合的实时性要求,介绍了一种解决目前高清或超高清多源图像拉普拉斯金字塔图像融合算法的实时实现方法。基于视频数据流设计了拉普拉斯金字塔并行流水线处理结构,分析了各流水线之间的时延及优化思路。通过片上缓存的方式补偿时延时间差,实现了算法的流水等长,保证了处理数据的完整性。该方法可以在赛灵思7系列及以上的可编程门阵列(field programmable gate array,FPGA)芯片上实现双通道1080×1920@60 Hz视频图像的5层拉普拉斯金字塔融合。实验结果显示,该实时并行处理方法融合效果良好,一帧图像的融合仅需10.535 ms,处理时延小于1 ms。