Ke Wu

Ke Wu

Auto-generated profile, based on publicly available information

Ke Wu

School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

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

Representative Publications

View All Publications
28 September 2020 in IEEE Sensors Journal
Verified
Journal Article
Closed access
KWKe Wu
Ke Wu
JTJie Tan
Jie Tan
HXHai Lun Xia
Hai Lun Xia
CLCheng Bao Liu
Cheng Bao Liu
Reflection happening on a specular surface severely degrades the performance of structured light-based 3D measurements. Traditional structured light techniques remain limited to ideal scenes. Inspired by the multiple exposure fusion technique in photography, which fuses a number of exposure sequences into a well-exposed image, we propose an exposure fusion (EF)-based structured light method in this work. With well-exposed images, the brightness difference of the specular surface can be reduced. Binary gray code is adopted as the primary structured light pattern. With the help of the EF technique, the generated well-exposed structured light images are used for 3D reconstruction of the specular surface. To evaluate the performance of our method, the defective batteries with reflective properties are tested in this experiment. The high dynamic range (HDR)-based structured light method, auto-HDR and ABFDP methods are introduced for comparison. In these experiments, an average error of 0.088 mm and a standard deviation of 0.1921 mm are obtained. These results show that the EF method performs better than other methods for the reconstruction of specular surfaces and can be used to distinguish defects on the surface of a battery.

22 March 2022 in IEEE Sensors Journal
Verified
Journal Article
Closed access
KWKe Wu
Ke Wu
JTJie Tan
Jie Tan
CLChengbao Liu
Chengbao Liu
It is difficult to detect the surface defects of a lithium battery with an aluminum/steel shell. The reflectivity, lack of 3D information on the battery surface, and the shortage of many datasets make the 2D detection method hard to apply in this field. In this paper, a cross-domain few-shot learning (FSL) approach for lithium-ion battery defect classification using an improved siamese network (BSR-SNet) is proposed. To obtain the critical 3D surface of the lithium-ion battery, a multiexposure-based structured light method is utilized. Then, the heights of the 3D cloud points are transferred to grayscale information and are saved as 8-bit 2D images. For the FSL task, the DAGM 2007 datasets are used as the source domain to pre-train the improved siamese model. To avoid negative mitigation in the target domain, batch spectral regularization (BSR) is added as a penalizer in the loss function. The accuracies of the experimental results are 93.3% for 10-shot batteries and 91.0% for 5-shot batteries, which means that our method can be used to classify the surface defects of lithium batteries well.

1 January 2022 in Neural Networks
Verified
Journal Article
Closed access
JLJingwei Li
Jingwei Li
HWHuanjie Wang
Huanjie Wang
KWKe Wu
Ke Wu
CLChengbao Liu
Chengbao Liu
JTJie Tan
Jie Tan

7 June 2021 in Measurement Science and Technology
Verified
Journal Article
Closed access
PDF file
KWKe Wu
Ke Wu
JTJie Tan
Jie Tan
CLChengbao Liu
Chengbao Liu
The approach to choose the exposure sequence substantially affects the performance of the multiexposure structured light method when reconstructing the 3D shape of high dynamic range (HDR) targets. Previous researchers have conducted numerous works regarding how to select exposure time automatically, but most of them lack the consideration of the time efficiency of the 3D measurement when selecting the exposure time. In this paper, a novel approach to obtain optimal exposure for HDR targets is proposed. We developed a concept, namely, exposure style, to help us find the optimal exposure sequence. The area-weighted method is proposed as a guide to find the preliminary exposure sequence with the demanded exposure style. By uniform sampling, the preliminary exposure sequence is sampled into several alternate sequences with different lengths. By utilizing the exposure fusion method, the fused images generated by the captured images, which is under the exposures of these alternate sequences, are used to reconstruct the 3D shape of the HDR target. Then, the running time of 3D measurement and the number of generated 3D cloud points are recorded. By evaluating these two indicators comprehensively, the time efficiency of the selected exposure sequence is well-considered, and the sequence with the highest score is selected as the final optimal exposure sequence. Three HDR targets with different grayscale distributions are chosen to testify to the practicability of our method. The auto-HDR method and Zhang’s method are introduced for comparison. The experimental results show that the optimal exposure sequence selected by our auto-exposure strategy performs better than other methods and the 3D shape of HDR targets can be reconstructed completely and accurately.

1 August 2019
Verified
Conference Paper
Closed access
HXHailun Xia
Hailun Xia
JTJie Tan
Jie Tan
KWKe Wu
Ke Wu
Gray code is a commonly used coding strategy in structured light techniques. However, projecting the large number of patterns required by it slows down the reconstruction. In this paper, a colored Gray code was generated to reduce the number of projected patterns and cut down the projecting time, which appeared to be a major time-consuming part especially for low-cost projectors. Meanwhile, two cameras were used to improve the performance of color stripes matching. Therefore, the complex considerations about the spectral characteristics of devices and the calibration of projectors could be avoided, making the system easy to use. Inverse colored patterns were also projected to assist the decoding procedure. The effects of the combination of colored line-shifting methods and different threshold parameters were also studied. A 130mm*100mm statue was reconstructed into a dense point cloud with 214,4848 points using our approach. Evaluations showed that the system had an average deviation of 0.02mm, and its speed can be further elevated with GPUs.