Zeyuan Liu’s research while affiliated with Changchun University of Technology and other places

For the challenges of low learning efficiency, slow convergence speed and slow inference speed in robot path planning. This paper proposes an improved deep reinforcement learning algorithm for robot path planning. Firstly, the Dueling DQN network architecture is employed, combined with a priority experience replay strategy, to more effectively learn from and utilize experience data. Secondly, the mobility space of the robot is expanded, enhancing the diversity and flexibility of the action space. Additionally, in the action selection process, the Artificial Potential Field (APF) algorithm is introduced to intervene in the action selection with a certain probability, thereby accelerating the convergence process of the network. Simultaneously, the $\varepsilon$ -greedy strategy is employed to balance exploration and exploitation, facilitating better exploration of the environment and utilization of existing knowledge. Furthermore, this paper devises novel composite reward functions that comprehensively integrate multiple reward mechanisms to enhance the convergence performance of the algorithm and the quality of path planning. Finally, the effectiveness and superiority of the proposed method are validated through detailed comparative simulations. Compared to traditional DQN algorithms, Double DQN, and Double DQN with the APF strategy, the method proposed in this paper demonstrates higher learning efficiency and faster convergence speed, enabling more effective planning of shorter paths.

In recent years, transfer learning (TL) approaches have seen extensive application in diagnosing bearing faults due to their exceptional performance. However, mechanical noise, equipment aging, and wear lead to notable disparities and differences in the multi-level feature distributions across the source and target domain signals. The issue is addressed by proposing a TL model based on a texture loss strategy and nuclear norm regularization method. First, a feature-enhanced network is designed, which significantly improves the ability to capture local details and long-range dependencies by combining a multi-scale feature extraction module with a dilated residual module. Next, a texture loss strategy is proposed to align multi-scale features across domains by minimizing the Gram matrix of signal features. Finally, a nuclear norm regularization method is proposed to perform low-rank approximation on the signal matrix, facilitating the extraction of more robust feature data and mitigating the risk of overfitting. The experimental results demonstrate that the proposed method achieved an average accuracy of 98.58% on the University of Ottawa bearing fault dataset and 98.11% on the Jiangnan University bearing dataset, surpassing eight other algorithms in bearing fault diagnosis.

With the development of 3D matching technology, point cloud registration (PCR) based on corresponding points has received increasing attention in the field of computer vision. Unfortunately, 3D keypoint technology inevitably produces a large number of outliers. To solve the problems of poor stability, low efficiency and the high number of iterations required to calculate the accepted solution of random sampling consistency (RANSAC) and its variants under a high outlier rate, a streamlined progressive sample consensus (SPROSAC) algorithm is proposed in this paper. SPROSAC is an improved estimator of progressive sample consensus that guides the sampling process by increasing the use of 3D point cloud surface information and optimizes the model verification process based on registration error decision acceptance. Compared to classic RANSAC-family algorithms, SPROSAC has a greater probability of obtaining an accepted solution more quickly. The experiments demonstrate that SPROSAC achieves significantly smaller and more stable registration errors with fewer iterations across three datasets. In the performance experiments based on evaluation metrics such as recall, 1-precision, and F1 score for inlier classification, SPROSAC demonstrates the best performance across the three datasets, with outlier rates exceeding 95%. Furthermore, we propose a coarse–fine PCR algorithm based on SPROSAC and ICP to address the issues of high initialization requirements, susceptibility to local optima, and low efficiency in traditional ICP algorithms. The experimental results of coarse–fine registration show that our algorithm provides initial values for the ICP, which can reduce the number of iterations of the ICP by 50%, 64.4%, and 57.4%.

Keypoint detection plays a pivotal role in three-dimensional computer vision, with widespread applications in improving registration precision and efficiency. However, current keypoint detection methods often suffer from poor robustness and low discriminability. In this study, a novel keypoint detection approach based on the local variation of surface (LVS) is proposed. The LVS keypoint detection method comprises three main steps. Firstly, the surface variation index for each point is calculated using the local coordinate system. Subsequently, points with a surface variation index lower than the local average are identified as initial keypoints. Lastly, the final keypoints are determined by selecting the minimum value within the neighborhood from the initial keypoints. Additionally, a sampling consensus correspondence estimation algorithm based on geometric constraints (SAC-GC) for efficient and robust estimation of optimal transformations in correspondences is proposed. By combining LVS and SAC-GC, we propose a coarse-to-fine point cloud registration algorithm. Experimental results on four public datasets demonstrate that the LVS keypoint detection algorithm offers improved repeatability and robustness, particularly when dealing with noisy, occluded, or cluttered point clouds. The proposed coarse-to-fine point cloud registration algorithm also exhibits enhanced robustness and computational efficiency.

In recent years, deep learning has been increasingly used in fault diagnosis of rotating machinery. However, the actual acquisition of rolling bearing fault signals often contains ambient noise, making it difficult to determine the optimal values of the parameters. In this paper, a sparrow search algorithm (LSSA) based on backward learning of lens imaging and Gaussian Cauchy variation is proposed. The lens imaging reverse learning strategy enhances the traversal capability of the algorithm and allows for a better balance of algorithm exploration and development. Then, the performance of the proposed LSSA was tested on the benchmark function. Finally, LSSA is used to find the optimal modal component K and the optimal penalty factor α in VMD-GRU, which in turn realizes the fault diagnosis of rolling bearings. The experimental results show that the model can achieve a 96.61% accuracy in rolling bearing fault diagnosis, which proves the effectiveness of the method.

3D point cloud registration has a wide range of applications in object shape detection, robot navigation and 3D reconstruction. Aiming at the problems of the traditional ICP registration algorithm, such as slow convergence speed and high requirements for the initial point cloud position, this paper proposes a coarse-fine point cloud registration method based on a fast and robust local point-pair feature (LPPF) and the ICP algorithm. The LPPF feature descriptor is an improved descriptor for the local application of classic point-pair features and is a histogram descriptor formed by counting the feature information of the local point cloud neighborhood. This paper completes point cloud registration through keypoint extraction, LPPF feature description, feature matching, coarse registration and fine registration. To verify the effectiveness of this method, under the evaluation indices of L1, RMSE and MAE, we analyzed the experimental results from the three aspects of descriptors, coarse registration and coarse-fine registration. Under Gaussian noise conditions, LPPF compared to the second-ranked descriptor, the L1 scores of LPPF on the Stanford, Kinect and Princeton datasets increased by 12%, 12.4% and 9.1%, respectively. The coarse registration experiment is compared with 5 classic descriptors on 3 commonly used datasets. The LPPF feature descriptor has higher registration accuracy and less registration time. Finally, the coarse-fine registration experiment shows that our method can reduce the number of iterations of the ICP algorithm by 77% under optimal conditions.