Kai Zeng's research while affiliated with Kunming University of Science and Technology and other places

To address the limitations of the Raft consensus algorithm, such as the lack of support for Byzantine fault tolerance, performance bottleneck of the leader single node, and high leader election delay, an improved Byzantine fault tolerance consensus algorithm called RaBFT based on Raft is proposed. The distribution process of log messages is optimized by utilizing the secret sharing technique to make it Byzantine fault tolerance, and the role of the committee is introduced to share the communication pressure of the leader, thereby resolving the performance bottleneck issue of the leader single node. The leader election algorithm based on a dynamic committee improves the speed of leader election and reduces the time required for leader election. The experimental results show that RaBFT algorithm has a significant improvement in throughput and consensus delay in the log replication phase, and has a lower leader election delay, RaBFT algorithm can improve the efficiency and performance of the system, it is a safe and efficient consensus algorithm.

Binarization efficiently compresses full-precision convolutional neural networks (CNNs) to achieve accelerated inference but with substantial performance degradations. Self-knowledge distillation (SKD) can significantly improve the performance of a network by inheriting its own advanced knowledge. However, SKD for binary neural networks (BNNs) remains underexplored because the binary characteristics of weak BNNs limit their ability to act as effective teachers, hindering their ability to learn as students. In this study, a novel SKD-BNN framework is proposed by using two pieces of underutilized information. Full-precision weights, which are applied for gradient transfer, concurrently distill the feature knowledge of the teacher with high-level semantics. A value-swapping strategy minimizes the knowledge capacity gap, while the channel-spatial difference distillation loss promotes feature transfer. Moreover, historical output predictions generate a concentrated soft-label bank, providing abundant intra- and inter-category similarity knowledge. Dynamic filtering ensures the correctness of the soft labels during training, and the label-cluster loss enhances the summarization ability of the soft-label bank within the same category. The developed methods excel in extensive experiments, achieving state-of-the-art accuracy of 93.0% on the CIFAR-10 dataset, which is equivalent to that of full-precision CNNs. On the ImageNet dataset, the accuracy improves by 1.6% with the widely adopted IR-Net. It is emphasized that for the first time, the proposed method fully explores the underutilized information contained in BNNs and conducts an effective SKD process, enabling weak BNNs to serve as competent self-teachers and proficient students.

The multitude of airborne point clouds limits the point cloud processing efficiency. Superpoints are grouped based on similar points, which can effectively alleviate the demand for computing resources and improve processing efficiency. However, existing superpoint segmentation methods focus only on local geometric structures, resulting in inconsistent spectral features of points within a superpoint. Such feature inconsistencies degrade the performance of subsequent tasks. Thus, this study proposes a novel Superpoint Segmentation method that jointly utilizes spatial Geometric and Spectral Information for multispectral point cloud superpoint segmentation (GSI-SS). Specifically, a similarity metric that combines spatial geometry and spectral information is proposed to facilitate the consistency of geometric structures and object attributes within segmented superpoints. Following the formation of the primary superpoints, an intersuperpoint pointexchange mechanism that maximizes feature consistency within the final superpoints is proposed. Experiments are conducted on two real multispectral point cloud datasets, and the proposed method achieved higher recall, precision, F score, and lower global consistency and feature classification errors. The experimental results demonstrate the superiority of the proposed GSI-SS over several state-of-the-art methods.

Point cloud data generated by LiDAR sensors play a critical role in 3D sensing systems, with applications encompassing object classification, part segmentation, and point cloud recognition. Leveraging the global learning capacity of dot product attention, transformers have recently exhibited outstanding performance in point cloud learning tasks. Nevertheless, existing transformer models inadequately address the challenges posed by uncertainty features in point clouds, which can introduce errors in the dot product attention mechanism. In response to this, our study introduces a novel global guidance approach to tolerate uncertainty and provide a more reliable guidance. We redefine the granulation and lower-approximation operators based on neighborhood rough set theory. Furthermore, we introduce a rough set-based attention mechanism tailored for point cloud data and present the rough set transformer (RST) network. Our approach utilizes granulation concepts derived from token clusters, enabling us to explore relationships between concepts from an approximation perspective, rather than relying on specific dot product functions. Empirically, our work represents the pioneering fusion of rough set theory and transformer networks for point cloud learning. Our experimental results, including point cloud classification and segmentation tasks, demonstrate the superior performance of our method. Our method establishes concepts based on granulation generated from clusters of tokens. Subsequently, relationships between concepts can be explored from an approximation perspective, instead of relying on specific dot product or addition functions. Empirically, our work represents the pioneering fusion of rough set theory and transformer networks for point cloud learning. Our experimental results, including point cloud classification and segmentation tasks, demonstrate the superior performance of our method.

As an important part of autonomous driving intelligence perception, pedestrian detection has high requirements for parameter size, real-time, and model performance. Firstly, a novel multiplexed connection residual block is proposed to construct the lightweight network for improving the ability to extract pedestrian features. Secondly, the lightweight scalable attention module is investigated to expand the local perceptual field of the model based on dilated convolution that can maintain the most important feature channels. Finally, we verify the proposed model on the Caltech pedestrian dataset and BDD 100 K datasets. The results show that the proposed method is superior to existing lightweight pedestrian detection methods in terms of model size and detection performance.

Remote sensing object detection based on the combination of infrared and visible images can effectively adapt to the around-the-clock and changeable illumination conditions. However, most of the existing infrared and visible object detection networks need two backbone networks to extract the features of two modalities, respectively. Compared with the single modality detection network, this greatly increases the amount of calculation, which limits its real-time processing on the vehicle and unmanned aerial vehicle (UAV) platforms. Therefore, this paper proposes a local adaptive illumination-driven input-level fusion module (LAIIFusion). The previous methods for illumination perception only focus on the global illumination, ignoring the local differences. In this regard, we design a new illumination perception submodule, and newly define the value of illumination. With more accurate area selection and label design, the module can more effectively perceive the scene illumination condition. In addition, aiming at the problem of incomplete alignment between infrared and visible images, a submodule is designed for the rapid estimation of slight shifts. The experimental results show that the single modality detection algorithm based on LAIIFusion can ensure a large improvement in accuracy with a small loss of speed. On the DroneVehicle dataset, our module combined with YOLOv5L could achieve the best performance.

Sensor fusion is an important component of the perception system in autonomous driving, and the fusion of radar point cloud information and camera visual information can improve the perception capability of autonomous vehicles. However, most of the existing studies ignore the extraction of local neighborhood information and only consider shallow fusion between the two modalities based on the extracted global information, which cannot perform a deep fusion of cross-modal contextual information interaction. Meanwhile, in data preprocessing, the noise in radar data is usually only filtered by the depth information derived from image feature prediction, and such methods affect the accuracy of radar branching to generate regions of interest and cannot effectively filter out irrelevant information of radar points. This paper proposes the CenterTransFuser model that makes full use of millimeter-wave radar point cloud information and visual information to enable cross-modal fusion of the two heterogeneous information. Specifically, a new interaction called cross-transformer is explored, which cooperatively exploits cross-modal cross-multiple attention and joint cross-multiple attention to mine radar and image complementary information. Meanwhile, an adaptive depth thresholding filtering method is designed to reduce the noise of radar modality-independent information projected onto the image. The CenterTransFuser model is evaluated on the challenging nuScenes dataset, and it achieves excellent performance. Particularly, the detection accuracy is significantly improved for pedestrians, motorcycles, and bicycles, showing the superiority and effectiveness of the proposed model.