Xuyang Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Xuyang Wang has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Chao Feng
Qilu University of Technology (Shandong Academy of Sciences)
3 shared publications - Kaiyue Liu
International School for Optoelectronic Engineering, Qilu University of Technology (Shandong Academy of Sciences),Jinan,China
2 shared publications - Kun Zhang
Key Laboratory of Ocean Observation and Forecasting & Laboratory of Ocean Circulation and Waves Institute of Oceanology Chinese Academy of Sciences Qingdao China
2 shared publications - Mengyao Zhang
School of Information and Automation, Qilu University of Technology (Shandong Academy of Sciences),Jinan,China
2 shared publications - Qinghua Li
School of Information and Automation, Qilu University of Technology (Shandong Academy of Sciences),Jinan,China
2 shared publications Baocai Pei
School of Information and Automation, Qilu University of Technology (Shandong Academy of Sciences),Jinan,China
1 shared publication
Latest Publications
- 17 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 5674-5679
SPANet—Sparse Convolutional Pyramid Attention Network for Grasping Detection in Low-Light Conditions
This paper presents a novel grasping detection network - the Sparse Convolutional Pyramid Attention Network (SPANet), designed to efficiently detect grasping in low-light conditions. The proposed network is particularly suitable for performing grasping tasks in low-light environments, including dimming enhancement and object detection. Furthermore, we propose a method based on inverse curve adjustment, which simulates low-light image algorithms with controllable dimming levels. The SPANet was trained and tested on the Cornell and Jacquard datasets, achieving detection accuracies of 94.38% and 90.76%, respectively. Moreover, we tested the SPANet on the Yuejiang CR5 robot in actual low-light environments, achieving a grasping success rate of 92.69 %. These results demonstrate the potential of the proposed approach to provide reliable and efficient solutions for robots operating in low-light environments.
- 17 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 5674-5679
SPANet—Sparse Convolutional Pyramid Attention Network for Grasping Detection in Low-Light Conditions
This paper presents a novel grasping detection network - the Sparse Convolutional Pyramid Attention Network (SPANet), designed to efficiently detect grasping in low-light conditions. The proposed network is particularly suitable for performing grasping tasks in low-light environments, including dimming enhancement and object detection. Furthermore, we propose a method based on inverse curve adjustment, which simulates low-light image algorithms with controllable dimming levels. The SPANet was trained and tested on the Cornell and Jacquard datasets, achieving detection accuracies of 94.38% and 90.76%, respectively. Moreover, we tested the SPANet on the Yuejiang CR5 robot in actual low-light environments, achieving a grasping success rate of 92.69 %. These results demonstrate the potential of the proposed approach to provide reliable and efficient solutions for robots operating in low-light environments.
- 17 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2287-2292
A Grasping Pose Detection Algorithm for Occluded Objects
To solve the problem of difficult grasping detection of occluded objects, this paper proposes a Channel Attention Grasp Convolution Neural Network (CA-GraspNet) based on convolution neural networks and channel attention mechanisms. CA-GraspNet integrates residual module and the SENet module. This module enables CNN to focus more on the no occluded area and less on the occluded area, overcoming the low accuracy of the existing methods in identifying occluded objects. The occluded datasets are obtained through data enhancement of Cornell and Jacquard grasp datasets, and the occluded datasets are used for algorithm verification. The experimental results show that CA-GraspNet can obtain 97% detection accuracy on Cornell occluded datasets, and 88% on Jacquard occluded datasets, which is higher than similar algorithms. Furthermore, CA-GraspNet also shows approximate performance in grasping detection.
- 17 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2287-2292
A Grasping Pose Detection Algorithm for Occluded Objects
To solve the problem of difficult grasping detection of occluded objects, this paper proposes a Channel Attention Grasp Convolution Neural Network (CA-GraspNet) based on convolution neural networks and channel attention mechanisms. CA-GraspNet integrates residual module and the SENet module. This module enables CNN to focus more on the no occluded area and less on the occluded area, overcoming the low accuracy of the existing methods in identifying occluded objects. The occluded datasets are obtained through data enhancement of Cornell and Jacquard grasp datasets, and the occluded datasets are used for algorithm verification. The experimental results show that CA-GraspNet can obtain 97% detection accuracy on Cornell occluded datasets, and 88% on Jacquard occluded datasets, which is higher than similar algorithms. Furthermore, CA-GraspNet also shows approximate performance in grasping detection.
- 17 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2388-2393
Robust Human Upper Limbs Trajectory Predictor for Human-Robot Collaboration
Accurately sensing the human posture information of collaborators is the critical technology to achieve human-robot collaboration, which can effectively improve the security of collaborative tasks. However, the pose of the collaborator, especially the motion trajectory of the human upper limb has the characteristics of randomness and sudden change. To solve the instability of prediction probability caused by the sudden change of trajectory and improve the prediction accuracy, this paper proposes a novel motion trajectory predictor for the human upper limb (GMM-LSTM). The algorithm can be described as follows: (1) according to the statistical feature of Human Upper Limbs motion, a joint probability distribution is inferred as its cluster model. A human upper limbs trajectory dataset can be clustered using this distribution as a sub-dataset. (2) an improved long and short-term memory network is proposed to predicate the future motion in three directions under the spatial coordinate system conditioned on the observed motion. (3) in the end, one fuse the predicted values of the three directions to generate the future trajectories. The experimental results show that the algorithm's overall prediction error consistently falls within a range of 1cm, indicating a high degree of robustness.
- 17 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2388-2393
Robust Human Upper Limbs Trajectory Predictor for Human-Robot Collaboration
Accurately sensing the human posture information of collaborators is the critical technology to achieve human-robot collaboration, which can effectively improve the security of collaborative tasks. However, the pose of the collaborator, especially the motion trajectory of the human upper limb has the characteristics of randomness and sudden change. To solve the instability of prediction probability caused by the sudden change of trajectory and improve the prediction accuracy, this paper proposes a novel motion trajectory predictor for the human upper limb (GMM-LSTM). The algorithm can be described as follows: (1) according to the statistical feature of Human Upper Limbs motion, a joint probability distribution is inferred as its cluster model. A human upper limbs trajectory dataset can be clustered using this distribution as a sub-dataset. (2) an improved long and short-term memory network is proposed to predicate the future motion in three directions under the spatial coordinate system conditioned on the observed motion. (3) in the end, one fuse the predicted values of the three directions to generate the future trajectories. The experimental results show that the algorithm's overall prediction error consistently falls within a range of 1cm, indicating a high degree of robustness.
- Open Access
Generating Adversarial Samples For Training Wake-up Word Detection Systems Against Confusing Words
Wake-up word detection models are widely used in real life, but suffer from severe performance degradation when encountering adversarial samples. In this paper we discuss the concept of confusing words in adversarial samples. Confusing words are commonly encountered, which are various kinds of words that sound similar to the predefined keywords. To enhance the wake word detection system's robustness against confusing words, we propose several methods to generate the adversarial confusing samples for simulating real confusing words scenarios in which we usually do not have any real confusing samples in the training set. The generated samples include concatenated audio, synthesized data, and partially masked keywords. Moreover, we use a domain embedding concatenated system to improve the performance. Experimental results show that the adversarial samples generated in our approach help improve the system's robustness in both the common scenario and the confusing words scenario. In addition, we release the confusing words testing database called HI-MIA-CW for future research.
Open AccessGenerating Adversarial Samples For Training Wake-up Word Detection Systems Against Confusing Words
Wake-up word detection models are widely used in real life, but suffer from severe performance degradation when encountering adversarial samples. In this paper we discuss the concept of confusing words in adversarial samples. Confusing words are commonly encountered, which are various kinds of words that sound similar to the predefined keywords. To enhance the wake word detection system's robustness against confusing words, we propose several methods to generate the adversarial confusing samples for simulating real confusing words scenarios in which we usually do not have any real confusing samples in the training set. The generated samples include concatenated audio, synthesized data, and partially masked keywords. Moreover, we use a domain embedding concatenated system to improve the performance. Experimental results show that the adversarial samples generated in our approach help improve the system's robustness in both the common scenario and the confusing words scenario. In addition, we release the confusing words testing database called HI-MIA-CW for future research.