2022 Vol. 31, No. 5

A Note for Estimation About Average Differential Entropy of Continuous Bounded Space-Time Random Field
SONG Zhanjie, ZHANG Jiaxing
2022, 31(5): 793-803. doi: 10.1049/cje.2021.00.213
Abstract(698) HTML (337) PDF(138)
In this paper, we mainly study the discrete approximation about average differential entropy of continuous bounded space-time random field. The estimation of differential entropy on random variable is a classic problem, and there are many related studies. Space-time random field is a theoretical extension of adding random variables to space-time parameters, but studies on discrete estimation of entropy on space-time random field are relatively few. The differential entropy forms of continuous bounded space-time random field and discrete estimations are discussed, and three estimation forms of differential entropy in the case of random variables are generated in this paper. Furthermore, it is concluded that under the condition that the entropy estimation formula after space-time segmentation converges with probability 1, the average entropy in the bounded space-time region can also converge with probability 1, and three generalized entropies are verified respectively. In addition, we also carried out numerical experiments on the convergence of average entropy estimation based on parameters, and the numerical results are consistent with the theoretical results, which indicting further study of the average entropy estimation problem of space-time random fields is significant in the future.
Variance-SNR Based Noise Suppression on Linear Canonical Choi-Williams Distribution of LFM Signals
ZHANG Zhichao
2022, 31(5): 804-820. doi: 10.1049/cje.2020.00.367
Abstract(445) HTML (210) PDF(61)
By solving the existing expectation-signal-to-noise ratio (expectation-SNR) based inequality model of the closed-form instantaneous cross-correlation function type of Choi-Williams distribution (CICFCWD), the linear canonical transform (LCT) free parameters selection strategies obtained are usually unsatisfactory. Since the second-order moment variance outperforms the first-order moment expectation in accurately characterizing output SNRs, this paper uses the variance analysis technique to improve parameters selection strategies. The CICFCWD’s average variance of deterministic signals embedded in additive zero-mean stationary circular Gaussian noise processes is first obtained. Then the so-called variance-SNRs are defined and applied to model a variance-SNR based inequality. A stronger inequalities system is also formulated by integrating expectation-SNR and variance-SNR based inequality models. Finally, a direct application of the system in noisy one-component and bi-component linear frequency-modulated (LFM) signals detection is studied. Analytical algebraic constraints on LCT free parameters newly derived seem more accurate than the existing ones, achieving better noise suppression effects. Our methods have potential applications in optical, radar, communication and medical signal processing.
Multipath Suppressing Method Based on Pseudorange Model Using Modified Teaching-Learning Based Optimization Algorithm
CHENG Lan, ZHANG Jing, NI Zihang, YAN Gaowei
2022, 31(5): 821-831. doi: 10.1049/cje.2020.00.168
Abstract(380) HTML (178) PDF(67)
Satellites based positioning has been widely applied to many areas in our daily lives and thus become indispensable, which also leads to increasing demand for high-positioning accuracy. In some complex environments (such as dense urban, valley), multipath interference is one of the main error sources deteriorating positioning accuracy, and it is difficult to eliminate via differential techniques due to its uncertainty of occurrence and irrelevance in different instants. To address this problem, we propose a positioning method for global navigation satellite systems (GNSS) by adopting a modified teaching-learning based optimization (TLBO) algorithm after the positioning problem is formulated as an optimization problem. Experiments are conducted by using actual satellite data. The results show that the proposed positioning algorithm outperforms other algorithms, such as particle swarm optimization based positioning algorithm, differential evolution based positioning algorithm, variable projection method, and TLBO algorithm, in terms of accuracy and stability.
Hyperspectral Image Classification Based on A Multi-Scale Weighted Kernel Network
SUN Le, XU Bin, LU Zhenyu
2022, 31(5): 832-843. doi: 10.1049/cje.2021.00.130
Abstract(906) HTML (436) PDF(71)
Recently, many deep learning models have shown excellent performance in hyperspectral image (HSI) classification. Among them, networks with multiple convolution kernels of different sizes have been proved to achieve richer receptive fields and extract more representative features than those with a single convolution kernel. However, in most networks, different-sized convolution kernels are usually used directly on multi-branch structures, and the image features extracted from them are fused directly and simply. In this paper, to fully and adaptively explore the multiscale information in both spectral and spatial domains of HSI, a novel multi-scale weighted kernel network (MSWKNet) based on an adaptive receptive field is proposed. First, the original HSI cubic patches are transformed to the input features by combining the principal component analysis and one-dimensional spectral convolution. Then, a three-branch network with different convolution kernels is designed to convolve the input features, and adaptively adjust the size of the receptive field through the attention mechanism of each branch. Finally, the features extracted from each branch are fused together for the task of classification. Experiments on three well-known hyperspectral data sets show that MSWKNet outperforms many deep learning networks in HSI classification.
Clustering for Topological Interference Management
JIANG Xue, ZHENG Baoyu, WANG Lei, HOU Xiaoyun
2022, 31(5): 844-850. doi: 10.1049/cje.2021.00.277
Abstract(307) HTML (153) PDF(41)
To reduce the overhead and complexity of channel state information acquisition in interference alignment, the topological interference management (TIM) was proposed to manage interference, which only relied on the network topology information. The previous research on topological interference management via the low-rank matrix completion approach is known to be NP-hard. This paper considers the clustering method for the topological interference management problem, namely, the low-rank matrix completion for TIM is applied within each cluster. Based on the clustering result, we solve the low-rank matrix completion problem via nuclear norm minimization and Frobenius norm minimization function. Simulation results demonstrate that the proposed clustering method combined with TIM leads to significant gain on the achievable degrees of freedom.
An Accurate Near-Field Distance Estimation Differential Algorithm
ZHAO Yan, TAO Haihong, CHANG Xin
2022, 31(5): 851-859. doi: 10.1049/cje.2021.00.174
Abstract(304) HTML (136) PDF(40)
The triangular geometry is the basis of near field array accurate distance estimation algorithms. The Fisher expression of traditional distance estimation is derived by utilizing the Taylor series. To improve convergence rate and estimation accuracy, a novel iterative distance estimation algorithm is proposed with differential equations based on the triangular geometry. Firstly, its convergence performance is analysed in detail. Secondly, the selection of the initial value and the number of iterations are respectively studied. Thirdly, compared with the traditional estimation algorithms by utilizing the Fisher approximation, the proposed algorithm has a higher convergence rate and estimation accuracy. Moreover, its pseudocode is presented. Finally, the experiment results and performance analysis are provided to verify the effectiveness of the proposed algorithm.
A Cross-Domain Ontology Semantic Representation Based on NCBI-BlueBERT Embedding
ZHAO Lingling, WANG Junjie, WANG Chunyu, GUO Maozu
2022, 31(5): 860-869. doi: 10.1049/cje.2020.00.326
Abstract(506) HTML (244) PDF(60)
A common but critical task in biological ontologies data analysis is to compare the difference between ontologies. There have been numerous ontology-based semantic-similarity measures proposed in specific ontology domain, but it still remains a challenge for cross-domain ontologies comparison. An ontology contains the scientific natural language description for the corresponding biological aspect. Therefore, we develop a new method based on natural language processing (NLP) representation model bidirectional encoder representations from transformers (BERT) for cross-domain semantic representation of biological ontologies. This article uses the BERT model to represent the word-level of the ontologies as a set of vectors, facilitating the semantic analysis or comparing the biomedical entities named in an ontology or associated with ontology terms. We evaluated the ability of our method in two experiments: calculating similarities of pair-wise disease ontology and human phenotype ontology terms and predicting the pair-wise of proteins interaction. The experimental results demonstrated the comparative performance. This gives promise to the development of NLP methods in biological data analysis.
AttentionSplice: An Interpretable Multi-Head Self-Attention Based Hybrid Deep Learning Model in Splice Site Prediction
YAN Wenjing, ZHANG Baoyu, ZUO Min, ZHANG Qingchuan, WANG Hong, MAO Da
2022, 31(5): 870-887. doi: 10.1049/cje.2021.00.221
Abstract(646) HTML (297) PDF(45)
Pre-mRNA splicing is an essential procedure for gene transcription. Through the cutting of introns and exons, the DNA sequence can be decoded into different proteins related to different biological functions. The cutting boundaries are defined by the donor and acceptor splice sites. Characterizing the nucleotides patterns in detecting splice sites is sophisticated and challenges the conventional methods. Recently, the deep learning frame has been introduced in predicting splice sites and exhibits high performance. It extracts high dimension features from the DNA sequence automatically rather than infers the splice sites with prior knowledge of the relationships, dependencies, and characteristics of nucleotides in the DNA sequence. This paper proposes the AttentionSplice model, a hybrid construction combined with multi-head self-attention, convolutional neural network, bidirectional long short-term memory network. The performance of AttentionSplice is evaluated on the Homo sapiens (Human) and Caenorhabditis Elegans (Worm) datasets. Our model outperforms state-of-the-art models in the classification of splice sites. To provide interpretability of AttentionSplice models, we extract important positions and key motifs which could be essential for splice site detection through the attention learned by the model. Our result could offer novel insights into the underlying biological roles and molecular mechanisms of gene expression.
Prediction of Protein Subcellular Localization Based on Microscopic Images via Multi-Task Multi-Instance Learning
ZHANG Pingyue, ZHANG Mengtian, LIU Hui, YANG Yang
2022, 31(5): 888-896. doi: 10.1049/cje.2020.00.330
Abstract(394) HTML (180) PDF(36)
Protein localization information is essential for understanding protein functions and their roles in various biological processes. The image-based prediction methods of protein subcellular localization have emerged in recent years because of the advantages of microscopic images in revealing spatial expression and distribution of proteins in cells. However, the image-based prediction is a very challenging task, due to the multi-instance nature of the task and low quality of images. In this paper, we propose a multi-task learning strategy and mask generation to enhance the prediction performance. Furthermore, we also investigate effective multi-instance learning schemes. We collect a large-scale dataset from the Human Protein Atlas database, and the experimental results show that the proposed multi-task multi-instance learning model outperforms both single-instance learning and common multi-instance learning methods by large margins.
Research on Virtual Coupled Train Control Method Based on GPC & VAPF
CAO Yuan, YANG Yaran, MA Lianchuan, WEN Jiakun
2022, 31(5): 897-905. doi: 10.1049/cje.2021.00.241
Abstract(522) HTML (250) PDF(39)
Improving transportation efficiency is an eternal research hotspot in rail transit system. In recent years, the train operation control method based on virtual coupling has attracted the attention of many scholars. The method of train coordination and anti-collision control is not only the key to realize the virtual coupling of train, but also the key to ensure the safety of train operation. Therefore, based on the existing research, a virtual coupled train dynamics model with nonlinear dynamics is established. Then, the parameters of the operation process model of the nonlinear virtual coupled train are identified by the recursive least squares method based on real-time data, which is applied to the variable parameter artificial potential field (VAPF) for parameter identification. A fusion controller based on feature-based generalized model prediction (GPC) and VAPF is used to control the virtual coupled train and prevent collision. Finally, the validity of the proposed method is verified by using real high-speed railway data.
Time Optimal Trajectory Planning Algorithm for Robotic Manipulator Based on Locally Chaotic Particle Swarm Optimization
DU Yuxiao, CHEN Yihang
2022, 31(5): 906-914. doi: 10.1049/cje.2021.00.373
Abstract(378) HTML (193) PDF(42)
Optimal trajectory planning is a fundamental problem in the area of robotic research. On the time-optimal trajectory planning problem during the motion of a robotic arm, the method based on segmented polynomial interpolation function with a locally chaotic particle swarm optimization (LCPSO) algorithm is proposed in this paper. While completing the convergence in the early or middle part of the search, the algorithm steps forward on the problem of local convergence of traditional particle swarm optimization (PSO) and improved learning factor PSO (IFPSO) algorithms. Finally, simulation experiments are executed in joint space to obtain the optimal time and smooth motion trajectory of each joint, which shows that the method can effectively shorten the running time of the robotic manipulator and ensure the stability of the motion as well.
Research on Global Clock Synchronization Mechanism in Software-Defined Control Architecture
LYU Shuyu, DAI Xinfa, MA Zhong, GAO Yi, HU Zhekun
2022, 31(5): 915-929. doi: 10.1049/cje.2021.00.059
Abstract(494) HTML (255) PDF(63)
Adopt software-definition technology to decouple the functional components of the industrial control system (ICS) in a service-oriented and distributed form is an important way for the industrial Internet of things to integrate information technology, communication technology, and operation technology. Therefore, this paper presents the concept of software-defined control architecture and describes the time consistency requirements under the paradigm shift of ICS architecture. By analyzing the physical clock and virtual clock mechanism models, the global clock synchronization space is logically divided into the physical and virtual clock synchronization domains, and a formal description of the global clock synchronization space is proposed. According to the fundamental analysis of the clock state model, the physical clock linear filtering synchronization model is derived, and a distributed observation fusion filtering model is constructed by considering the two observation modes of the virtual clock to realize the time synchronization of the global clock space by way of timestamp layer-by-layer transfer and fusion estimation. Finally, the simulation results show that the proposed model can significantly improve the accuracy and stability of clock synchronization.
Intelligent Orchestrating of IoT Microservices Based on Reinforcement Learning
WU Yuqin, SHEN Congqi, CHEN Shuhan, WU Chunming, LI Shunbin, WEI Ruan
2022, 31(5): 930-937. doi: 10.1049/cje.2020.00.417
Abstract(360) HTML (166) PDF(36)
With the recent increase in the number of Internet of things (IoT) services, an intelligent scheduling strategy is needed to manage these services. In this paper, the problem of automatic choreography of microservices in IoT is explored. A type of reinforcement learning (RL) algorithm called TD3 is used to generate the optimal choreography policy under the framework of a softwaredefined network. The optimal policy is gradually reached during the learning procedure to achieve the goal, despite the dynamic characteristics of the network environment. The simulation results show that compared with other methods, the TD3 algorithm converges faster after a certain number of iterations, and it performs better than other non-RL algorithms by obtaining the highest reward. The TD3 algorithm can effciently adjust the traffic transmission path and provide qualified IoT services.
Learning to Combine Answer Boundary Detection and Answer Re-ranking for Phrase-Indexed Question Answering
WEN Liang, SHI Haibo, ZHANG Xiaodong, SUN Xin, WEI Xiaochi, WANG Junfeng, CHENG Zhicong, YIN Dawei, WANG Xiaolin, LUO Yingwei, WANG Houfeng
2022, 31(5): 938-948. doi: 10.1049/cje.2021.00.079
Abstract(632) HTML (309) PDF(62)
Phrase-indexed question answering (PIQA) seeks to improve the inference speed of question answering (QA) models by enforcing complete independence of the document encoder from the question encoder, and it shows that the constrained model can achieve significant efficiency at the cost of its accuracy. In this paper, we aim to build a model under the PIQA constraint while reducing its accuracy gap with the unconstrained QA models. We propose a novel framework—AnsDR, which consists of an answer boundary detector (AnsD) and an answer candidate ranker (AnsR). More specifically, AnsD is a QA model under the PIQA architecture and it is designed to identify the rough answer boundaries; and AnsR is a lightweight ranking model to finely re-rank the potential candidates without losing the efficiency. We perform the extensive experiments on public datasets. The experimental results show that the proposed method achieves the state of the art on the PIQA task.
Lexicon-Augmented Cross-Domain Chinese Word Segmentation with Graph Convolutional Network
YU Hao, HUANG Kaiyu, WANG Yu, HUANG Degen
2022, 31(5): 949-957. doi: 10.1049/cje.2021.00.363
Abstract(316) HTML (158) PDF(31)
Existing neural approaches have achieved significant progress for Chinese word segmentation (CWS). The performances of these methods tend to drop dramatically in the cross-domain scenarios due to the data distribution mismatch across domains and the out of vocabulary words problem. To address these two issues, proposes a lexicon-augmented graph convolutional network for cross-domain CWS. The novel model can capture the information of word boundaries from all candidate words and utilize domain lexicons to alleviate the distribution gap across domains. Experimental results on the cross-domain CWS datasets (SIGHAN-2010 and TCM) show that the proposed method successfully models information of domain lexicons for neural CWS approaches and helps to achieve competitive performance for cross-domain CWS. The two problems of cross-domain CWS can be effectively solved through various interactions between characters and candidate words based on graphs. Further, experiments on the CWS benchmarks (Bakeoff-2005) also demonstrate the robustness and efficiency of the proposed method.
DeepHGNN: A Novel Deep Hypergraph Neural Network
LIN Jingjing, YE Zhonglin, ZHAO Haixing, FANG Lusheng
2022, 31(5): 958-968. doi: 10.1049/cje.2021.00.108
Abstract(881) HTML (428) PDF(103)
With the development of deep learning, graph neural networks (GNNs) have yielded substantial results in various application fields. GNNs mainly consider the pair-wise connections and deal with graph-structured data. In many real-world networks, the relations between objects are complex and go beyond pair-wise. Hypergraph is a flexible modeling tool to describe intricate and higher-order correlations. The researchers have been concerned how to develop hypergraph-based neural network model. The existing hypergraph neural networks show better performance in node classification tasks and so on, while they are shallow network because of over-smoothing, over-fitting and gradient vanishment. To tackle these issues, we present a novel deep hypergraph neural network (DeepHGNN). We design DeepHGNN by using the technologies of sampling hyperedge, residual connection and identity mapping, residual connection and identity mapping bring from graph convolutional neural networks. We evaluate DeepHGNN on two visual object datasets. The experiments show the positive effects of DeepHGNN, and it works better in visual object classification tasks.
A Novel Neighborhood-Weighted Sampling Method for Imbalanced Datasets
GUANG Mingjian, YAN Chungang, LIU Guanjun, WANG Junli, JIANG Changjun
2022, 31(5): 969-979. doi: 10.1049/cje.2021.00.121
Abstract(555) HTML (258) PDF(51)
The weighted sampling methods based on k-nearest neighbors have been demonstrated to be effective in solving the class imbalance problem. However, they usually ignore the positional relationship between a sample and the heterogeneous samples in its neighborhood when calculating sample weight. This paper proposes a novel neighborhood-weighted based sampling method named NWBBagging to improve the Bagging algorithm’s performance on imbalanced datasets. It considers the positional relationship between the center sample and the heterogeneous samples in its neighborhood when identifying critical samples. And a parameter reduction method is proposed and combined into the ensemble learning framework, which reduces the parameters and increases the classifier’s diversity. We compare NWBBagging with some state-of-the-art ensemble learning algorithms on 34 imbalanced datasets, and the result shows that NWBBagging achieves better performance.
Combination for Conflicting Interval-Valued Belief Structures with CSUI-DST Method
LI Shuangming, GUAN Xin, YI Xiao, SUN Guidong
2022, 31(5): 980-990. doi: 10.1049/cje.2021.00.214
Abstract(330) HTML (148) PDF(35)
Since the basic probability of an interval-valued belief structure (IBS) is assigned as interval number, its combination becomes difficult. Especially, when dealing with highly conflicting IBSs, most of the existing combination methods may cause counter-intuitive results, which can bring extra heavy computational burden due to nonlinear optimization model, and lose the good property of associativity and commutativity in Dempster-Shafer theory (DST). To address these problems, a novel conflicting IBSs combination method named CSUI (conflict, similarity, uncertainty, intuitionistic fuzzy sets)-DST method is proposed by introducing a similarity measurement to measure the degree of conflict among IBSs, and an uncertainty measurement to measure the degree of discord, non-specificity and fuzziness of IBSs. Considering these two measures at the same time, the weight of each IBS is determined according to the modified reliability degree. From the perspective of intuitionistic fuzzy sets, we propose the weighted average IBSs combination rule by the addition and number multiplication operators. The effectiveness and rationality of this combination method are validated with two numerical examples and its application in target recognition.