Wei Li

Wei Li

Auto-generated profile, based on publicly available information

Wei Li

Computer Science, Xidian University, Xi’an, Shaanxi, 710126, China

Research Focus & Metrics

Academic Career Length

14Years (2013-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
9 July 2020 in Scientific Reports
Verified
Journal Article
Open access
PDF file
WLWei Li
Wei Li
KLKai Liu
Kai Liu
LZLizhe Zhang
Lizhe Zhang
FCFei Cheng
Fei Cheng
Object detection is an important component of computer vision. Most of the recent successful object detection methods are based on convolutional neural networks (CNNs). To improve the performance of these networks, researchers have designed many different architectures. They found that the CNN performance benefits from carefully increasing the depth and width of their structures with respect to the spatial dimension. Some researchers have exploited the cardinality dimension. Others have found that skip and dense connections were also of benefit to performance. Recently, attention mechanisms on the channel dimension have gained popularity with researchers. Global average pooling is used in SENet to generate the input feature vector of the channel-wise attention unit. In this work, we argue that channel-wise attention can benefit from both global average pooling and global max pooling. We designed three novel attention units, namely, an adaptive channel-wise attention unit, an adaptive spatial-wise attention unit and an adaptive domain attention unit, to improve the performance of a CNN. Instead of concatenating the output of the two attention vectors generated by the two channel-wise attention sub-units, we weight the two attention vectors based on the output data of the two channel-wise attention sub-units. We integrated the proposed mechanism with the YOLOv3 and MobileNetv2 framework and tested the proposed network on the KITTI and Pascal VOC datasets. The experimental results show that YOLOv3 with the proposed attention mechanism outperforms the original YOLOv3 by mAP values of 2.9 and 1.2% on the KITTI and Pascal VOC datasets, respectively. MobileNetv2 with the proposed attention mechanism outperforms the original MobileNetv2 by a mAP value of 1.7% on the Pascal VOC dataset.

1 May 2013 in Journal of Biomedical Optics
Verified
Journal Article
Closed access
PDF file
HYHuangjian Yi
Huangjian Yi
DCDuofang Chen
Duofang Chen
WLWei Li
Wei Li
SZSHOUPING ZHU
SHOUPING ZHU
XWXiaorui Wang
Xiaorui Wang
JTJie Tian
Jie Tian
Fluorescence molecular tomography (FMT) is an important imaging technique of optical imaging. The major challenge of the reconstruction method for FMT is the ill-posed and underdetermined nature of the inverse problem. In past years, various regularization methods have been employed for fluorescence target reconstruction. A comparative study between the reconstruction algorithms based on l 1 -norm and l 2 -norm for two imaging models of FMT is presented. The first imaging model is adopted by most researchers, where the fluorescent target is of small size to mimic small tissue with fluorescent substance, as demonstrated by the early detection of a tumor. The second model is the reconstruction of distribution of the fluorescent substance in organs, which is essential to drug pharmacokinetics. Apart from numerical experiments, in vivo experiments were conducted on a dual-modality FMT/micro-computed tomography imaging system. The experimental results indicated that l 1norm regularization is more suitable for reconstructing the small fluorescent target, while l 2-norm regularization performs better for the reconstruction of the distribution of fluorescent substance.

1 June 2022 in Journal of Physics: Conference Series
Verified
Conference Paper
Open access
WLWei Li
Wei Li
Vehicle detection is the key to driverless technology. For safety, driverless technology requires extremely high accuracy and real-time for vehicle detection in different situations. In this paper, we study an enhanced YOLO -based algorithm for vehicle detection in foggy weather conditions. We add a dehazing module in the YOLO model for more information restoration, which is built by the multi-scale retinex with color restoration (MSRCR). And the enhanced model is trained with the augmentation data processed MSRCR for more stable performance. We evaluate our method in the public dataset, the results show the enhanced YOLO model has better performance than conventional YOLO in vehicle detection in foggy weather.