2023 Vol. 32, No. 3

IMAGE AND SIGNAL PROCESSING
Quaternion Quasi-Chebyshev Non-local Means for Color Image Denoising
XU Xudong, ZHANG Zhihua, M. James C. Crabbe
2023, 32(3): 397-414. doi: 10.23919/cje.2022.00.138
Abstract(1323) HTML (274) PDF(104)
Abstract:
Quaternion non-local means (QNLM) denoising algorithm makes full use of high degree self-similarities inside images to suppress the noise, so the similarity metric plays a key role in its denoising performance. In this study, two improvements have been made for the QNLM: 1) For low level noise, the use of quaternion quasi-Chebyshev distance is proposed to measure the similarity of image patches and it has been used to replace the Euclidean distance in the QNLM algorithm. Since the quasi-Chebyshev distance measures the maximal distance in all color channels, the similarity of color images measured by quasi-Chebyshev distance can capture the structural similarity uniformly for each color channel; 2) For high level noise, quaternion bilateral filtering has been proposed as the preprocessing step in the QNLM algorithm. Denoising simulations were performed on 110 images of landscape, people, and architecture at different noise levels. Compared with QNLM, quaternion non-local total variation (QNLTV), and non-local means (NLM) variants (NLTV, NLM after wavelet threshold preprocessing, and the color adaptation of NLM), our novel algorithm not only improved PSNR/SSIM (peak signal to noise rate/structural similarity) and figure of merit values by an average of 2.77 dB/8.96% and 0.0491 respectively, but also reduced processing time.
Hyperspectral Image Super-Resolution Based on Spatial-Spectral Feature Extraction Network
LI Yanshan, CHEN Shifu, LUO Wenhan, ZHOU Li, XIE Weixin
2023, 32(3): 415-428. doi: 10.23919/cje.2021.00.081
Abstract(916) HTML (458) PDF(68)
Abstract:
Constrained by the physics of hyperspectral sensors, the spatial resolution of hyperspectral images (HSI) is low. Hyperspectral image super-resolution (HSI SR) is a task to obtain high-resolution hyperspectral images from low-resolution hyperspectral images. Existing algorithms have the problem of losing important spectral information while improving spatial resolution. To handle this problem, a spatial-spectral feature extraction network (SSFEN) for HSI SR is proposed in this paper. It enhances the spatial resolution of the HSI while preserving the spectral information. The SSFEN is composed of three parts: spatial-spectral mapping network, spatial reconstruction network, and spatial-spectral fusing network. And a joint loss function with spatial and spectral constraints is designed to guide the training of the SSFEN. Experiment results show that the proposed method improves the spatial resolution of the HSI and effectively preserves the spectral information simultaneously.
A Low Complexity Distributed Multitarget Detection and Tracking Algorithm
FAN Jiande, XIE Weixin, LIU Zongxiang
2023, 32(3): 429-437. doi: 10.23919/cje.2021.00.282
Abstract(2199) HTML (963) PDF(97)
Abstract:
In this paper, we propose a low complexity distributed approach to address the multitarget detection/tracking problem in the presence of noisy and missing data. The proposed approach consists of two components: a distributed flooding scheme for measurements exchanging among sensors and a sampling-based clustering approach for target detection/tracking from the aggregated measurements. The main advantage of the proposed approach over the prevailing Markov-Bayes-based distributed filters is that it does not require any priori information and all the information required is the measurement set from multiple sensors. A comparison of the proposed approach with the available distributed clustering approaches and the cutting edge distributed multi-Bernoulli filters that are modeled with appropriate parameters confirms the effectiveness and the reliability of the proposed approach.
Necessary Condition for the Success of Synchronous GNSS Spoofing
WANG Yiwei, KOU Yanhong, HUANG Zhigang
2023, 32(3): 438-452. doi: 10.23919/cje.2021.00.307
Abstract(594) HTML (294) PDF(82)
Abstract:
A synchronous GNSS generator spoofer aims at directly taking over the tracking loops of the receiver with the lowest possible spoofing to signal ratio (SSR) without forcing it to lose lock. This paper investigates the factors that affect spoofing success and their relationships. The necessary conditions for successful spoofing are obtained by deriving the code tracking error in the presence of spoofing and analyzing the effects of SSR, spoofing synchronization errors, and receiver settings on the S-curve ambiguity and code tracking trajectory. The minimum SSRs for a successful spoofing calculated from the theoretical formulation agree with Monte Carlo simulations at digital intermediate frequency signal level within 1 dB when the spoofer pulls the code phase in the same direction as the code phase synchronization error, and the required SSRs can be much lower when pulling in the opposite direction. The maximum spoofing code phase error for a successful spoofing is tested by using TEXBAT datasets, which coincides with the theoretical results within 0.1 chip. This study reveals the mechanism of covert spoofing and can play a constructive role in the future development of spoofing and anti-spoofing methods.
Unique Parameters Selection Strategy of Linear Canonical Wigner Distribution via Multiobjective Optimization Modeling
SHI Xiya, WU Anyang, SUN Yun, QIANG Shengzhou, JIANG Xian, HAN Puyu, CHEN Yunjie, ZHANG Zhichao
2023, 32(3): 453-464. doi: 10.23919/cje.2021.00.338
Abstract(2723) HTML (1081) PDF(75)
Abstract:
There are many kinds of linear canonical transform (LCT)-based Wigner distributions (WDs), which are very effective in detecting noisy linear frequency-modulated (LFM) signals. Among WDs in LCT domains, the instantaneous cross-correlation function type of Wigner distribution (ICFWD) attracts much attention from scholars, because it achieves not only low computational complexity but also good detection performance. However, the existing LCT free parameters selection strategy, namely a solution of the expectation-based output signal-to-noise ratio (SNR) optimization model, is not unique. In this paper, by introducing the variance-based output SNR optimization model, a multiobjective optimization model is established. Then the existence and uniqueness of the optimal parameters of ICFWD are investigated. The solution of the multiobjective optimization model with respect to one-component LFM signal added with zero-mean stationary circular Gaussian noise is derived. A comparison of the unique parameters selection strategy and the previous one is carried out. The theoretical results are also verified by numerical simulations.
A Novel Re-weighted CTC Loss for Data Imbalance in Speech Keyword Spotting
LAN Xiaotian, HE Qianhua, YAN Haikang, LI Yanxiong
2023, 32(3): 465-473. doi: 10.23919/cje.2021.00.198
Abstract(734) HTML (393) PDF(82)
Abstract:
Speech keyword spotting system is a critical component of human-computer interfaces. And connectionist temporal classifier (CTC) has been proven to be an effective tool for that task. However, the standard training process of speech keyword spotting faces a data imbalance issue where positive samples are usually far less than negative samples. Numerous easy-training negative examples overwhelm the training, resulting in a degenerated model. To deal with it, this paper tries to reshape the standard CTC loss and proposes a novel re-weighted CTC loss. It evaluates the sample importance by its number of detection errors during training and automatically down-weights the contribution of easy examples, the majorities of which are negatives, making the training focus on samples deserving more training. The proposed method can alleviate the imbalance naturally and make use of all available data efficiently. Evaluation on several sets of keywords selected from AISHELL-1 and AISHELL-2 achieves 16%–38% relative reductions in false rejection rates over standard CTC loss at 0.5 false alarms per keyword per hour in experiments.
A Directly Readable Halftone Multifunctional Color QR Code
HUANG Yuan, CAO Peng, LYU Guangwu
2023, 32(3): 474-484. doi: 10.23919/cje.2021.00.366
Abstract(858) HTML (425) PDF(62)
Abstract:
Color quick response (QR) code is an important direction for the future development of QR code, which has become a research hotspot due to the additional functional characteristics of its colors as the wide application of QR code technology. The existing color QR code has solved the problem of information storage capacity, but it requires an enormous hardware and software support system, making how to achieve its direct readability an urgent issue. This paper proposes a novel color QR code that combines multiple types of different identification information. This code combines multiplexing and color-coding technology to present the publicly encoded information (such as advertisements, public query information) as plain code, and traceability, blockchain, anti-counterfeiting authentication and other information concealed in the form of hidden code. We elaborate the basic principle of this code, construct its mathematical model and supply a set of algorithm design processes, which breakthrough key technology of halftone printout. The experimental results show that the proposed color quick response code realizes the multi-code integration and can be read directly without special scanning equipment, which has unique advantages in the field of printing anti-counterfeiting labels.
Convolution Theorem Associated with the QWFRFT
MEI Yinyin, FENG Qiang, GAO Xiuxiu, ZHAO Yanbo
2023, 32(3): 485-492. doi: 10.23919/cje.2021.00.225
Abstract(830) HTML (418) PDF(77)
Abstract:
The quaternion windowed fractional Fourier transform (QWFRFT) is a generalized form of the quaternion fractional Fourier transform (QFRFT), it plays a crucial role in signal processing for the analysis of multidimensional signals. In this paper, we first give the definition of the two-sided QWFRFT and some fundamental properties. Then, the quaternion convolution is proposed, and the relation between the quaternion convolution and the classical convolution is also given. Based on the quaternion convolution of the QWFRFT, relevant convolution theorems for the QWFRFT are studied. Moreover, the fast algorithm for QWFRFT is discussed. Finally, the complexity of QWFRFT and the quaternion windowed fractional convolution are given.
Monaural Speech Separation Using Dual-Output Deep Neural Network with Multiple Joint Constraint
SUN Linhui, LIANG Wenqing, ZHANG Meng, LI Ping’an
2023, 32(3): 493-506. doi: 10.23919/cje.2022.00.110
Abstract(347) HTML (173) PDF(29)
Abstract:
Monaural speech separation is a significant research field in speech signal processing. To achieve a better separation performance, we propose three novel joint-constraint loss functions and a multiple joint-constraint loss function for monaural speech separation based on dual-output deep neural network (DNN). The multiple joint-constraint loss function for DNN separation model not only restricts the ideal ratio mask (IRM) errors of the two outputs, but also constrains the relationship of the estimated IRMs and the magnitude spectrograms of the clean speech signals, the relationship of the estimated IRMs of the two outputs, and the relationship of the estimated IRMs and the magnitude spectrogram of the mixed signal. The constraint strength is adjusted through three parameters to improve the accuracy of the speech separation model. Furthermore, we solve the optimal weighting coefficients of the multiple joint-constraint loss function based on the optimization idea, which further improves the performance of the separation system. We conduct a series of speech separation experiments on the GRID corpus to validate the superiority performance of the proposed method. The results show that using perceptual evaluation of speech quality, the short-time objective intelligibility, source to distortion ratio, signal to interference ratio and source to artifact ratio as the evaluation metrics, the proposed method outperforms the conventional DNN separation model. Taking the gender into consideration, we carry out experiments among Female-Female, Male-Male and Male-Female cases, which show that our method improves the robustness and performance of the separation system compared with some previous approaches.
Unsupervised Video Object Segmentation via Weak User Interaction and Temporal Modulation
FAN Jiaqing, ZHANG Kaihua, ZHAO Yaqian, LIU Qingshan
2023, 32(3): 507-518. doi: 10.23919/cje.2022.00.139
Abstract(662) HTML (335) PDF(26)
Abstract:
In unsupervised video object segmentation (UVOS), the whole video might segment the wrong target due to the lack of initial prior information. Also, in semi-supervised video object segmentation (SVOS), the initial video frame with a fine-grained pixel-level mask is essential to good segmentation accuracy. It is expensive and laborious to provide the accurate pixel-level masks for each training sequence. To address this issue, We present a weak user interactive UVOS approach guided by a simple human-made rectangle annotation in the initial frame. We first interactively draw the region of interest by a rectangle, and then we leverage the mask RCNN (region-based convolutional neural networks) method to generate a set of coarse reference labels for subsequent mask propagations. To establish the temporal correspondence between the coherent frames, we further design two novel temporal modulation modules to enhance the target representations. We compute the earth mover’s distance (EMD)-based similarity between coherent frames to mine the co-occurrent objects in the two images, which is used to modulate the target representation to highlight the foreground target. We design a cross-squeeze temporal modulation module to emphasize the co-occurrent features across frames, which further helps to enhance the foreground target representation. We augment the temporally modulated representations with the original representation and obtain the compositive spatio-temporal information, producing a more accurate video object segmentation (VOS) model. The experimental results on both UVOS and SVOS datasets including Davis2016, FBMS, Youtube-VOS, and Davis2017, show that our method yields favorable accuracy and complexity. The related code is available.
INFORMATION SECURITY AND CRYPTOLOGY
A Semi-shared Hierarchical Joint Model for Sequence Labeling
LIU Gongshen, DU Wei, ZHOU Jie, LI Jing, CHENG Jie
2023, 32(3): 519-530. doi: 10.23919/cje.2020.00.363
Abstract(906) HTML (453) PDF(19)
Abstract:
Multi-task learning is an essential yet practical mechanism for improving overall performance in various machine learning fields. Owing to the linguistic hierarchy, the hierarchical joint model is a common architecture used in natural language processing. However, in the state-of-the-art hierarchical joint models, higher-level tasks only share bottom layers or latent representations with lower-level tasks thus ignoring correlations between tasks at different levels, i.e., lower-level tasks cannot be instructed by the higher features. This paper investigates how to advance the correlations among various tasks supervised at different layers in an end-to-end hierarchical joint learning model. We propose a semi-shared hierarchical model that contains cross-layer shared modules and layer-specific modules. To fully leverage the mutual information between various tasks at different levels, we design four different dataflows of latent representations between the shared and layer-specific modules. Extensive experiments on CTB-7 and CONLL-2009 show that our semi-shared approach outperforms basic hierarchical joint models on sequence tagging while having much fewer parameters. It inspires us that the proper implementation of the cross-layer sharing mechanism and residual shortcuts is promising to improve the performance of hierarchical joint natural language processing models while reducing the model complexity.
A Hybrid Entropy and Blockchain Approach for Network Security Defense in SDN-Based IIoT
SU Jian, JIANG Mengnan
2023, 32(3): 531-541. doi: 10.23919/cje.2022.00.103
Abstract(368) HTML (185) PDF(34)
Abstract:
In the industrial Internet of things (IIoT), various applications generate a large number of interactions and are vulnerable to various attacks, which are difficult to be monitored in a sophisticated way by traditional network architectures. Therefore, deploying software-defined network (SDN) in IIoT is essential to defend against various attacks. However, SDN has a drawback: there is a security problem of distributed denial-of-service (DDoS) attacks at the control layer. This paper proposes an effective solution: DDoS detection within the domain using tri-entropy in information theory. The detected attacks are then uploaded to a smart contract in the blockchain, so that the attacks can be quickly cut off even if the same attack occurs in different domains. Experimental validation was conducted under different attack strengths and multiple identical attacks, and the results show that the method has better detection ability under different attack strengths and can quickly block the same attacks.
ESE: Efficient Security Enhancement Method for the Secure Aggregation Protocol in Federated Learning
TIAN Haibo, LI Maonan, REN Shuangyin
2023, 32(3): 542-555. doi: 10.23919/cje.2021.00.370
Abstract(572) HTML (288) PDF(35)
Abstract:
In federated learning, a parameter server may actively infer sensitive data of users and a user may arbitrarily drop out of a learning process. Bonawitz et al. propose a secure aggregation protocol for federated learning against a semi-honest adversary and a security enhancement method against an active adversary at ACM CCS 2017. The purpose of this paper is to analyze their security enhancement method and to design an alternative. We point out that their security enhancement method has the risk of Eclipse attack and that the consistency check round in their method could be removed. We give a new efficient security enhancement method by redesigning an authentication message and by adjusting the authentication timing. The new method produces an secure aggregation protocol against an active adversary with less communication and computation costs.
A Verifiable Multi-Secret Sharing Scheme Based on Short Integer Solution
LI Fulin, YAN Jiayun, ZHU Shixin, HU Hang
2023, 32(3): 556-563. doi: 10.23919/cje.2021.00.062
Abstract(369) HTML (181) PDF(38)
Abstract:
With the possible birth of the quantum computer, traditional secret sharing schemes have been unable to meet security requirements. We proposed a new verifiable multi-secret sharing scheme based on the short integer solution problem. By utilizing a symmetric binary polynomial, ${\boldsymbol{k}}$ secrets and secret shares can be generated, and then we convert the secret shares into binary string on $\mathbb{Z}_{\boldsymbol{q}}$ , which can be identified by one-way anti-collision hash function on the lattice, so that multiple secrets can be reconstructed safely. The advantages mainly focus on verifiability without interaction in the distribution phase and less memory requirement. In a secret sharing scheme, verifiability prevents the dealer to share the wrong shares and forces the participants to submit their shares correctly. Meanwhile, the interaction can be reduced, which means the security is improved. In a multi-secret sharing scheme, releasing the public values is inevitable, this paper has less public values and less size of shares per secret size to reduce the pressure of memory consumption in the proper parameters. In the end, it can also effectively resist the quantum attack.
A Novel Construction of Updatable Identity-Based Hash Proof System and Its Applications
QIAO Zirui, ZHOU Yanwei, YANG Bo, ZHANG Wenzheng, ZHANG Mingwu
2023, 32(3): 564-576. doi: 10.23919/cje.2022.00.203
Abstract(354) HTML (178) PDF(27)
Abstract:
In the previous works, to further provide the continuous leakage resilience for the identity-based encryption scheme, a new cryptography primitive, called updatable identity-based hash proof system (U-IB-HPS), was proposed. However, most of the existing constructions have some deficiencies, they either do not have perfect key update function or the corresponding security with tight reduction relies on a non-static complexity assumption. To address the above problems, a new construction of U-IB-HPS is created, and the corresponding security of our system is proved based on the static complexity assumption. Also, the corresponding comparisons and analysis of performances show that our proposal not only achieves the perfect key update function and the anonymity, but also has the tight security reduction. In additional, our proposal achieves the same computational efficiency as other previous systems. To further illustrate the practical function of U-IB-HPS, a generic method of non-interactive data authorization protocol with continuous leakage resilience is designed by employing U-IB-HPS as an underlying tool, which can provide continuous leakage-resilient data authorization function for the cloud computing. Hence, the application field of U-IB-HPS is further extended through our study.
RESS: A Reliable and Effcient Storage Scheme for Bitcoin Blockchain Based on Raptor Code
SHI Dongxian, WANG Xiaoqing, XU Ming, KOU Liang, CHENG Hongbing
2023, 32(3): 577-586. doi: 10.23919/cje.2022.00.343
Abstract(222) HTML (110) PDF(37)
Abstract:
The Bitcoin system uses a fully replicated data storage mechanism in which each node keeps a full copy of the blockchain. As the number of nodes in the system increases and transactions get more complex, more and more storage space are needed to store block data. The scalability of storage has become a bottleneck, limiting the practical application of blockchain. This paper proposes a node storage scheme, called RESS, to integrate erasure coding technology into the blockchain to encode multiple blocks. Under the proposed block grouping method, nodes can reduce the times of coded block decoding. In addition, the coding scheme based on Raptor codes proposed in this paper has linear coding and decoding complexity. The rateless feature of Raptor code helps to achieve high decentralization and scalability of the Bitcoin network. RESS ensures data availability, efficiency and blockchain robustness based on achieving storage space scalability. Experimental results show that the proposed scheme reduces the storage requirements of nodes by nearly an order of magnitude.
Zero-Cerd: A Self-Blindable Anonymous Authentication System Based on Blockchain
YANG Kunwei, YANG Bo, WANG Tao, ZHOU Yanwei
2023, 32(3): 587-596. doi: 10.23919/cje.2022.00.047
Abstract(483) HTML (241) PDF(40)
Abstract:
While the Internet of things brings convenience to people’s lives, it will also bring people hidden worries about data security. As an important barrier to protect data security, identity authentication is widely used in the Internet of things. However, it is necessary to protect users’ identity privacy while authenticating their identity. Anonymous authentication technology is often used to solve the contradiction between legitimacy and privacy in the authentication process. The existing anonymous authentication scheme has many problems in practical application such as the inability to achieve complete anonymity, the high computational complexity of the algorithm, and the corruption of the central authority. Aiming at the privacy of authentication, we propose Zero-Cerd, a self-blindable anonymous authentication system based on blockchain and dynamic accumulator. The self-blinding properties of the credential enable the users themselves to generate a new validly pseudonymous credential. With the help of zero-knowledge proof technology, users can prove the validity of their credentials without disclosing any information. Security analysis shows that our scheme has achieved the expected security objectives. Compared with the existing schemes, our scheme has the advantages of complete anonymity and high efficiency, and is more suitable for IoT applications with privacy protection requirements.
Recover the Secret Components in a ForkCipher
HOU Tao, ZHANG Jiyan, CUI Ting
2023, 32(3): 597-602. doi: 10.23919/cje.2021.00.368
Abstract(360) HTML (184) PDF(27)
Abstract:
Recently, a new cryptographic primitive has been proposed called ForkCiphers. This paper aims at proposing new generic cryptanalysis against such constructions. We give a generic method to apply existing decompositions againt the underlying block cipher ${{\cal{{E}}}}^{\boldsymbol{r}}$ on the forking variant Fork ${{\cal{E}}}$ -(r−1)-r0-(r+1−r0). As application, we consider the security of ForkSPN and ForkFN with secret inner functions. We provide a generic attack against ForkSPN-2-r0-(4−r0) based on the decomposition of SASAS. And also we extend the decomposition of Biryukov et al. against Feistel networks in SAC 2015 to get all the unknown round functions in ForkFN-r-r0-r1 for r≤ 6 and r0+r1≤ 8. Therefore, compared with the original block cipher, the forking version requires more iteration rounds to resist the recovery attack.
A New Edge Perturbation Mechanism for Privacy-Preserving Data Collection in IOT
CHEN Qiuling, YE Ayong, ZHANG Qiang, HUANG Chuan
2023, 32(3): 603-612. doi: 10.23919/cje.2021.00.411
Abstract(588) HTML (298) PDF(37)
Abstract:
A growing amount of data containing the sensitive information of users is being collected by emerging smart connected devices to the center server in Internet of things (IoT) era, which raises serious privacy concerns for millions of users. However, existing perturbation methods are not effective because of increased disclosure risk and reduced data utility, especially for small data sets. To overcome this issue, we propose a new edge perturbation mechanism based on the concept of global sensitivity to protect the sensitive information in IoT data collection. The edge server is used to mask users’ sensitive data, which can not only avoid the data leakage caused by centralized perturbation, but also achieve better data utility than local perturbation. In addition, we present a global noise generation algorithm based on edge perturbation. Each edge server utilizes the global noise generated by the center server to perturb users’ sensitive data. It can minimize the disclosure risk while ensuring that the results of commonly performed statistical analyses are identical and equal for both the raw and the perturbed data. Finally, theoretical and experimental evaluations indicate that the proposed mechanism is private and accurate for small data sets.
ARTIFICIAL INTELLIGENCE
Semantic Comprehension Method for Chinese Sentences Based on Minimal Semantic Structures and Its Application
WEN Hao, HE Qianru
2023, 32(3): 613-624. doi: 10.23919/cje.2021.00.161
Abstract(124) HTML (63) PDF(16)
Abstract:
The importance of small Chinese sentences is no less than that of sentences, which is an inherent feature of Chinese itself. According to this characteristic, this paper proposes a sentence semantic understanding method for Chinese scientific and technological abstracts based on the minimum semantic structure. Firstly, a conceptual model was established for identifying the minimum semantic structure of a sentence based on a corpus of verbs, relative words, prepositions and markers based on Language Technology Planform (LTP) tools. Secondly, the model was used to extract the minimum semantic structure of abstract sentence. Finally, three experiments were carried out, namely, the classification of the abstract sentences, knowledge graph generation and automatic semantic inference discovery. Our study confirmed the practical value of the small Chinese sentence. The experimental results show that the effect of using small sentences to understand the semantics of Chinese text is better than that of the full stop sentence, and the minimum semantic structure can be used as the basic unit of the Chinese sentence semantic comprehension. This method is conducive in the automatic understanding of the basic semantics of sentences in unstructured Chinese science and technology text sentences.
Explainable Business Process Remaining Time Prediction Using Reachability Graph
CAO Rui, ZENG Qingtian, NI Weijian, LU Faming, LIU Cong, DUAN Hua
2023, 32(3): 625-639. doi: 10.23919/cje.2021.00.170
Abstract(527) HTML (233) PDF(53)
Abstract:
With the recent advances in the field of deep learning, an increasing number of deep neural networks have been applied to business process prediction tasks, remaining time prediction, to obtain more accurate predictive results. However, existing time prediction methods based on deep learning have poor interpretability, an explainable business process remaining time prediction method is proposed using reachability graph, which consists of prediction model construction and visualization. For prediction models, a Petri net is mined and the reachability graph is constructed to obtain the transition occurrence vector. Then, prefixes and corresponding suffixes are generated to cluster into different transition partitions according to transition occurrence vector. Next, the bidirectional recurrent neural network with attention is applied to each transition partition to encode the prefixes, and the deep transfer learning between different transition partitions is performed. For the visualization of prediction models, the evaluation values are added to the sub-processes of a Petri net to realize the visualization of the prediction models. Finally, the proposed method is validated by publicly available event logs.
Improved Cross-Corpus Speech Emotion Recognition Using Deep Local Domain Adaptation
ZHAO Huijuan, YE Ning, WANG Ruchuan
2023, 32(3): 640-646. doi: 10.23919/cje.2021.00.196
Abstract(911) HTML (449) PDF(110)
Abstract:
Due to the scarcity of high-quality labeled speech emotion data, it is natural to apply transfer learning to emotion recognition. However, transfer learning-based speech emotion recognition becomes more challenging because of the complexity and ambiguity of emotion. Domain adaptation based on maximum mean discrepancy considers marginal alignment of source domain and target domain, but not pay regard to class prior distribution in both domains, which results in the reduction of transfer efficiency. In order to address the problem, this study proposes a novel cross-corpus speech emotion recognition framework based on local domain adaption. A category-grained discrepancy is used to evaluate the distance between two relevant domains. According to research findings, the generalization ability of the model is enhanced by using the local adaptive method. Compared with global adaptive and non-adaptive methods, the effectiveness of cross-corpus speech emotion recognition is significantly improved.
Representation of Semantic Word Embeddings Based on SLDA and Word2vec Model
TANG Huanling, ZHU Hui, WEI Hongmin, ZHENG Han, MAO Xueli, LU Mingyu, GUO Jin
2023, 32(3): 647-654. doi: 10.23919/cje.2021.00.113
Abstract(818) HTML (395) PDF(46)
Abstract:
To solve the problem of semantic loss in text representation, this paper proposes a new embedding method of word representation in semantic space called wt2svec based on supervised latent Dirichlet allocation (SLDA) and Word2vec. It generates the global topic embedding word vector utilizing SLDA which can discover the global semantic information through the latent topics on the whole document set. It gets the local semantic embedding word vector based on the Word2vec. The new semantic word vector is obtained by combining the global semantic information with the local semantic information. Additionally, the document semantic vector named doc2svec is generated. The experimental results on different datasets show that wt2svec model can obviously promote the accuracy of the semantic similarity of words, and improve the performance of text categorization compared with Word2vec.
Convolutional Neural Networks of Whole Jujube Fruits Prediction Model Based on Multi-Spectral Imaging Method
WANG Jing, FAN Xiaofei, SHI Nan, ZHAO Zhihui, SUN Lei, SUO Xuesong
2023, 32(3): 655-662. doi: 10.23919/cje.2021.00.149
Abstract(479) HTML (236) PDF(49)
Abstract:
Soluble sugar is an important index to determine the quality of jujube, and also an important factor to influence the taste of jujube. The acquisition of the soluble sugar content of jujube mainly relies on manual chemical measurement which is time-consuming and labor-intensive. In this study, the feasibility of multispectral imaging combined with deep learning for rapid nondestructive testing of fruit internal quality was analyzed. Support vector machine regression model, partial least squares regression model, and convolutional neural networks (CNNs) model were established by multispectral imaging method to predict the soluble sugar content of the whole jujube fruit, and the optimal model was selected to predict the content of three kinds of soluble sugar. The study showed that the sucrose prediction model of the whole jujube had the best performance after CNNs training, and the correlation coefficient of verification set was 0.88, which proved the feasibility of using CNNs for prediction of the soluble sugar content of jujube fruits.