Search

Aiming at the initial alignment problem of low-cost INS/GNSS in the case of moving base, this paper proposes a set of GNSS-assisted low-cost INS combined navigation initial alignment schemes, which is designed, optimized, and verified. Firstly, this paper analyses the impact of accelerometer zero bias on the accuracy of multi-vector construction and proposes an improved vector construction method, which approximates the suppression of the error caused by the accelerometer zero bias accumulating over time by vector subtraction instead of sliding window integration and also overcomes the impact of the OBA algorithm’s computational increase and memory occupation caused by the fact that the sliding window integration needs to store and integrate the data in the window. Secondly, this paper also designs an improved Kalman filter model to reconstruct the state equation and measurement equation of the system, which can more accurately estimate the zero bias of the gyroscope online and compensate for the misalignment angle of the carrier system by using the velocity and position information of the GNSS, eliminating the influence of the zero bias of the gyroscope on the accuracy of the multi-vector, and improving the accuracy of the initial alignment of the moving base and the robustness of the algorithm. Finally, the algorithm proposed in this paper and other algorithms are compared and analyzed in simulation and vehicle test experiments. The simulation experiments verify the feasibility of the initial alignment method proposed in this paper; the vehicle test experiments use vector subtraction and sliding window integration to construct the multivector, and then estimate the zero bias of the gyroscope and compensate for the misalignment angle of the carrier system by using Kalman filtering model proposed in this paper and other Kalman filtering models, respectively. The measured results show that the proposed method has better alignment accuracy and algorithm robustness than the existing Optimization-based alignment and their improvement algorithms for the moving base of low-cost INS/GNSS combined navigation.

Text classification is one of the most fundamental tasks in text data analysis. Generally, textual data are extensive in scale and complex in the inherent relationship, leading to low accuracy in traditional classification models. This paper proposes a multi-label classification model based on multi-layer neural network architecture. We first construct a dual-attention mechanism graph neural network (named DGAT for short) to fuse the typological and informational features of the target node and the connected nodes. Secondly, we also build a multi-layer network architecture with multiple DGATs to expand the range of neighborhood nodes participating in the feature fusion to meet the needs of classifying different datasets. To ensure the learning ability of the model, we also use the residual network to solve the problem of error rise and gradient descent caused by multi-layer network architecture. Finally, we conducted a large number of experiments on five benchmark datasets. The results show that the accuracy of the proposed model is significantly better than that of the traditional models, and that there is a noticeable improvement when compared with other deep learning methods.

The issues of mean-square admissibility and synthesis of It $\hat {o}$ -type stochastic singular systems (SSSs) under Brownian parameter perturbations are introduced in this article. For ease of computation, a novel sufficient condition is given to guarantee autonomous systems are mean-square admissible via strict linear matrix inequalities(LMIs). Furthermore, own to the measurable of the system states, both state feedback controller and observer-based controller (OBC) for It $\hat {o}$ -type SSSs are investigated. However, in It $\hat {o}$ -type SSSs, because the state of the system and the observer can be affected by Brownian fluctuation, it is not feasible that the observer and control gains design are completely separate. To this end, an innovative design approach is also proposed to solve the controller and observer parameters simultaneously in form of strict LMIs. Finally, three examples are introduced to demonstrate the effectiveness of the proposed method.

This paper proposes a modified relevance vector machine with slow feature analysis fault classification for industrial processes. Traditional support vector machine classification does not work well when there are insufficient training samples. A relevance vector machine, which is a Bayesian learning-based probabilistic sparse model, is developed to determine the probabilistic prediction and sparse solutions for the fault category. This approach has the benefits of good generalization ability and robustness to small training samples. To maximize the dynamic separability between classes and reduce the computational complexity, slow feature analysis is used to extract the inner dynamic features and reduce the dimension. Experiments comparing the proposed method, relevance vector machine and support vector machine classification are performed using the Tennessee Eastman process. For all faults, relevance vector machine has a classification rate of 39%, while the proposed algorithm has an overall classification rate of 76.1%. This shows the efficiency and advantages of the proposed method.

Many deep learning architectures have been employed to model the semantic compositionality for text sequences, requiring a huge amount of supervised data for parameters training, making it unfeasible in situations where numerous annotated samples are not available or even do not exist. Different from data-hungry deep models, lightweight word embedding-based models could represent text sequences in a plug-and-play way due to their parameter-free property. In this paper, a modified hierarchical pooling strategy over pre-trained word embeddings is proposed for text classification in a few-shot transfer learning way. The model leverages and transfers knowledge obtained from some source domains to recognize and classify the unseen text sequences with just a handful of support examples in the target problem domain. The extensive experiments on five datasets including both English and Chinese text demonstrate that the simple word embedding-based models (SWEMs) with parameter-free pooling operations are able to abstract and represent the semantic text. The proposed modified hierarchical pooling method exhibits significant classification performance in the few-shot transfer learning tasks compared with other alternative methods.

Human activity recognition (HAR) has received a lot of attention due to its wide applications in recent years, while the improvement of recognition accuracy is seemingly considered to be one of the great challenges in this field. In this paper, a novel wearable device for improving the activity recognition accuracy is proposed based on the different multiple sensors, which simultaneously collects the muscle activity and motion information. The muscular activity is monitored by measuring the air pressure in an air bladder contacting the targeted muscle, while the motion information, such as three-axis accelerations and angular velocities of body movement, is collected via the on-body inertial measurement unit (IMU) sensor. The performance of the air-pressure sensor is verified by comparing with the electromyography and the IMU sensors. To implement our device, we collect the labeled activities data from eight subjects as they perform 11 daily activities. Some commonly used features from raw data are calculated, and five popular classification techniques are evaluated in terms of the accuracy, recall, precision, and F-measure. The experimental results indicate that the proposed wearable device can improve the performance of HAR system. Particularly, the usage of air-pressure sensor can eliminate the confusions among dynamic activities, such as walking and going upstairs, which is an open problem in HAR.

This paper is devoted to the distributed optimization problem of heterogeneous multi-agent systems, where the communication topology is jointly strongly connected and the dynamics of each agent is the first-order or second-order integrator. A new distributed algorithm is first designed for each agent based on the local objective function and the local neighbors' information that each agent can access. By a model transformation, the original closed-loop system is converted into a time-varying system and the system matrix of which is a stochastic matrix at any time. Then, by the properties of the stochastic matrix, it is proven that all agents' position states can converge to the optimal solution of a team objective function provided the union communication topology is strongly connected. Finally, the simulation results are provided to verify the effectiveness of the distributed algorithm proposed in this paper.

The walking state monitoring is indispensable during the robot-aided walking of people with lower limb dysfunctions. In this paper, the existence of human-robot coordination state is first statistically verified in the process of using a walking-aid cane-type robot during walking. Based on this coordination, a new walking state monitoring method is proposed by using the principal component analysis (PCA). The abnormal or emergency walking state is promptly detected if the new sample data are found to deviate from an off-line PCA model, which is generated from plentiful normal walking data of different subjects. Furthermore, a state diagnosis algorithm based on the contribution plot is also developed for the walking state recognition and diagnosis. In this way, typical abnormal states like the leg restrictions can be distinguished from the emergency states including falls and the stumbling. Moreover, the human-robot coordination analysis can be performed using less sensors built-in the robot without needing the posture information of full human body. The effectiveness of the proposed method is proven by experiments. Better recognition rate and real-time performance of the method are also verified by comparing with conventional center of pressure based monitoring method.

The firefly algorithm has been successfully used in many optimization problems. However, the standard firefly algorithm uses a fixed randomization parameter in the optimization, which emphasizes more on exploration than exploitation, and hence impacts its convergence. This paper proposes a switchmode firefly algorithm, which first focuses on exploration and then switches to exploitation. A fixed randomization parameter is used in exploration, and a gradually decreasing random randomization parameter is used in exploitation. The condition for the switching from exploration to exploitation is identified automatically. Extensive experiments on 15 benchmark functions were performed to verify the effectiveness of the proposed approach.

Optional course selection is a critical activity for college students due to a large number of available but unfamiliar optional courses. Improper selection of optional courses would seriously affect the students' optional course achievements, which enforces students to drop out the improperly selected optional courses. Therefore, there is an urgent need to develop an optional course recommendation system. In this paper, we develop an optional course recommendation system based on score prediction. In particular, a novel cross-user-domain collaborative filtering algorithm is designed to accurately predict the score of the optional course for each student by using the course score distribution of the most similar senior students. After generating the predicted scores of all optional courses, the top $t$ optional courses with the highest predicted scores without time conflict will be recommended to the student. The extensive experiments have been conducted to evaluate the effectiveness of the proposed method, and the results show that the proposed method is able to accurately recommend optional courses to students who will achieve relatively high scores.

Recently, the low-dimensional embedding of nodes has received a large amount of attention in the field of network analysis. While the existing methods mostly focus on the network embedding of the entire network, there are also some situations, where people may only be interested in some nodes (i.e., partial nodes) rather than all nodes, especially in large-scale networks. Although there are some approaches dealing with the large-scale network, most of them require that all nodes are present during the optimization procedure. The necessity to generate the embedding results for redundant non-interested nodes makes these methods quite inefficient. In this paper, we present a novel node representation framework termed partial nodes representation learning (PartNRL), which is capable of preserving the local similarity of the interested node pairs and generating the embedding results efficiently. Two phases are carefully designed in PartNRL. The first phase is to use local random walk to capture the t -step local similarity for memory-efficiency. By modeling the inherent properties of nodes through the first phase, we design the second phase to learn the node representations by maximizing a likelihood function based on the skip-gram model. To make the optimization procedure more efficient, the negative sampling strategy is applied. The extensive experiments have been conducted on three tasks: node clustering task, single-labeled classification task, and multi-labeled classification task. The experimental results confirm the superior performance of the proposed model over the state-of-the-art network embedding methods.

Person re-identification (Re-ID), which is for matching pedestrians across disjoint camera views in surveillance, has made great progress in supervised learning. However, requirement of a large number of labelled identities leads to high cost for large-scale Re-ID systems. Consequently, it is significant to study learning Re-ID with unlabelled data and limited labelled data, that is, semi-supervised person re-identification. When labelled data is limited, the learned model tends to overfit the data and cannot generalize well. Moreover, the scene variations between cameras lead to domain shift in the feature space, which makes mining auxiliary supervision information from unlabelled data more difficult. To address these problems, we propose a Distilled Camera-Aware Self Training framework for semi-supervised person re-identification. To alleviate the overfitting problem for learning from limited labelled data, we propose a Multi-Teacher Selective Similarity Distillation Loss to selectively aggregate the knowledge of multiple weak teacher models trained with different subsets and distill a stronger student model. Then, we exploit the unlabelled data by learning pseudo labels by clustering based on the student model for self training. To alleviate the effect of scene variations between cameras, we propose a Camera-Aware Hierarchical Clustering (CAHC) algorithm to perform intra-camera clustering and cross-camera clustering hierarchically. Experiments show that our method outperformed the state-of-the-art semi-supervised person re-identification methods.

Network community detection with higher-order features has become a hot research topic recently since higher-order features that are captured at the level of small network subgraphs can help to gain new insights into the higher-order organizations of the overall network. However, most of the existing higher-order community detection methods only leverage the structural information of networks, yet fail to consider the rich information of node attributes, which typically makes them incapable of discovering semantically meaningful communities. To address this problem, we propose a novel method for community detection based on a newly designed higher-order feature termed Attribute Homogenous Motif (AHMotif), which integrates both node attributes and higher-order structure of the network in a seamless way. Specifically, the statistically significant motifs are identified for the network in terms of the topological structure, and homogeneity values are computed for structural motifs according to the attributes of the nodes that are involved in the motif. Then, the given attributed network can be further represented via a newly designed AHMotif adjacency matrix. The AHMotif adjacency matrix encodes both structures and attribute information and can characterize the network from a higher-order perspective. Afterward, several proximity based methods are utilized to obtain the final community detection results. Extensive experiments on several real-world data sets have shown the superiority of our AHMotif-based method in community detection.