Weichao Wu

Dr. Weichao Wu


Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

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

Representative Publications

View All Publications
28 January 2025 in Signal, Image and Video Processing
Verified
Journal Article
Closed access
PDF file
YSYONG SUN
YONG SUN
JWJianzhong Wang
Jianzhong Wang
YYYou Yu
You Yu
ZYZibo Yu
Zibo Yu
SBShaobo Bian
Shaobo Bian
WWWeichao Wu
Weichao Wu

1 August 2024 in IEEE Access
Verified
Journal Article
Open access
YYYu You
Yu You
JWJianzhong Wang
Jianzhong Wang
ZYZibo Yu
Zibo Yu
YSYong Sun
Yong Sun
YPYiguo Peng
Yiguo Peng
WWWeichao Wu
Weichao Wu
Owing to its ability to provide more accurate and detailed battlefield situational information, fine-grained detection research on soldier targets is of significant importance for military decision-making and firepower threat assessment. To address the issues of low detection accuracy and inaccurate classification in the fine-grained detection of soldier targets, we propose a fine-gain soldier target detection model based on the improved YOLOv8 (You Only Look Once v8). First, we developed a multi-branch feature fusion module to effectively fuse multi-scale feature information and used a dynamic deformable attention mechanism to help the detection model focus on key areas in deep-level features. Second, we proposed a decoupled lightweight dynamic head to extract the position and category information of soldier targets separately, effectively solving the problem of misclassification of soldier targets’ attack actions under different poses. Finally, we used the Inner Minimum Points Distance Intersection over Union (Inner-MPDIoU) to further improve the convergence speed and accuracy of the network model. The proposed improvements are evaluated through comparative experiments conducted in published twenty-six test groups, and the effectiveness of the proposed method is demonstrated. Compared with the original model, our method achieved a detection precision of 78.9%, a 6.91% improvement; the mAP@50 (mean Average Precision at 50) was 79.6%, a 3.51% increase; and an mAP@50-95 of 63.8%, a gain of 5.28%. The proposed method achieves high precision and recall while reducing the computational complexity of the model, thereby enhancing its efficiency and robustness for fine-grained soldier target detection.

18 July 2024 in IEEE Access
Verified
Journal Article
Open access
EWEndi Wang
Endi Wang
JWJianzhong Wang
Jianzhong Wang
SZSheng Zhang
Sheng Zhang
YSYong Sun
Yong Sun
YYYu You
Yu You
WWWeichao Wu
Weichao Wu
In urban warfare and rescue scenarios, the localization of soldiers in staircase environments presents substantial challenges. Addressing this issue, this study introduces the “Target Localization in Staircase Environments-Proximal Policy Optimization” module, which combines object detection with reinforcement learning algorithms. Utilizing a bespoke dataset, “Soldier-Staircase for Tracked Robots”, and design standards tailored for swing-arm tracked robots, the module meets the requirements for autonomous precise localization in staircase environments. A series of experiments conducted in simulated environments verify the module’s efficacy in autonomously and accurately localizing soldier targets in staircase environments, laying the groundwork for further research into the application of reinforcement learning in autonomous robot control.

27 February 2024 in Research Square
Verified
Preprint
Open access
PDF file
YHYongchen Han
Yongchen Han
WWWeichao Wu
Weichao Wu
HLHongyu Lan
Hongyu Lan
CBChen Bai
Chen Bai
GWGuoqiang Wu
Guoqiang Wu
ZGZhiming Guo
Zhiming Guo
Feature-based visual odometry has difficulty in feature extraction and matching in weak texture environment, resulting in substantial inter-frame pose resolution errors. Meanwhile, the computation and matching of feature point descriptors can be time-consuming and computationally inefficient. To address these issues encountered by traditional ORB-SLAM odometry in texture-lacking regions, an enhanced method for visual odometry estimation is proposed. First, the quadtree technique is employed to extract ORB feature points with a uniform distribution and an adequate number. Subsequently, when processing non-critical frames, the optical flow method is utilized to predict the precise locations of the feature points, circumventing the need for feature matching. Following this, the random sampling consistency method is applied to eliminate mismatched points in optical flow tracking, ensuring that only high-quality internal points are retained. Afterwards, a system of nonlinear equations is solved using AP3P method to estimate the precise position of the camera. Finally, the trajectory is optimized by Dogleg algorithm to achieve accurate and stable tracking and positioning. The experimental results demonstrate that the improved algorithm outperforms mainstream ORB-SLAM3 algorithm in terms of operation efficiency and positioning accuracy across multiple experimental scenarios. This method effectively addresses the challenges of low tracking accuracy and poor real-time performance commonly encountered by traditional visual odometers operating in weak texture environments. As a result, the method combining the feature-based method and the optical flow method significantly enhances the application of visual odometry in complex environments by improving the tracking stability, motion estimation accuracy, and real-time performance.

26 January 2026 in The Visual Computer
Verified
Journal Article
Closed access
PDF file
SBShaoBo Bian
ShaoBo Bian
JWJianzhong Wang
Jianzhong Wang
YYYou Yu
You Yu
ZYZibo Yu
Zibo Yu
YSYONG SUN
YONG SUN
WWWeichao Wu
Weichao Wu

3 August 2024
Verified
Book Chapter
Closed access
PDF file
CLChenglin Li
Chenglin Li
CLCheng Liu
Cheng Liu
WWWeichao Wu
Weichao Wu