Table of Content

30 September 2024, Volume 42 Issue 5

Previous Issue    Next Issue

Signal and Information Processing

Review of Steganalysis for Digital Images

WANG Zichi, LI Bin, FENG Guorui, ZHANG Xinpeng

2024, 42(5):  723-732.  doi:10.3969/j.issn.0255-8297.2024.05.001

Asbtract ( 391 )   PDF (2695KB) ( 438 )  

References | Related Articles | Metrics

Digital steganography plays a crucial role in securely transmitting confidential data by concealing it within common multimedia, such as images, videos, and audio, to facilitate covert communication. To discover the covert communication of steganography, the technique of steganalysis can be employed. Steganalysis determines whether a given multimedia object contains secret data according to the statistical anomaly of stego data caused by steganography. In recent years, both steganography and steganalysis have made significant progress and development in their mutual confrontation, particularly in the context of the growing prevalence of digital images on social networks. Focusing on digital images, this paper sorts out the development of steganalysis in the past decade, and reviews the traditional steganalysis and deep learning based-steganalysis. Then, the limitations of each approach are discussed. Finally, the study offers insights into the prospective development trends in steganalysis.

Change Detection of 3D Mural Surface Based on Multi-view Contour Points of LiDAR Data

XIAO Kairong, MENG Qingxiang, YANG He, GONG Yuanfu

2024, 42(5):  733-746.  doi:10.3969/j.issn.0255-8297.2024.05.002

Asbtract ( 156 )   PDF (7260KB) ( 76 )  

References | Related Articles | Metrics

In this paper, we propose a novel method for detecting changes in 3D mural surfaces. This method processes the mural point cloud data by point cloud alignment and principal component analysis (PCA) algorithm. It extracts the feature contour lines using Gaussian sphere mapping in the direction of multiple lines of sight. Combined with voxel grid shifting and four-dimensional surface fitting techniques, accurate detection of geometric changes of frescoes is achieved. The practical application results demonstrate significant advantages in detection accuracy and efficiency, which is of great value for the protection and restoration of ancient cultural relics.

Plural Adversarial Sample Generation Method for SAR-ATR System

ZHANG Mengjun, XIONG Bangshu

2024, 42(5):  747-756.  doi:10.3969/j.issn.0255-8297.2024.05.003

Asbtract ( 144 )   PDF (4752KB) ( 64 )  

References | Related Articles | Metrics

Existing adversarial attack methods are limited to real-valued convolutional neural networks. To address this limitation, this paper proposes a complex adversarial sample generation method based on generative adversarial networks. Firstly, a complex model for generating effective adversarial samples is designed by introducing complex computation modules. Secondly, a pre-trained complex network is used as the discriminator in adversarial training, with a residual neural network serving as the basic framework, to enhance the attack capability of adversarial samples. Finally, transferable adversarial attacks are achieved through substitute models, resulting in higher attack success rates. Experimental results demonstrate that the success rates of the proposed method in targeted and untargeted attack tasks reach 76.338% and 87.841%, respectively, with higher transferability and closer resemblance between adversarial and original clean samples. By extending adversarial attacks to complex neural networks, this method preserves synthetic aperture radar (SAR) target information and accuracy, providing a reference solution for the security and robustness of practical synthetic aperture radar automatic target recognition (SAR-ATR) systems.

Casting Defect Detection Based on Local and Global Features

LI Sha, WANG Yongxiong, WANG Zhe, CHEN Xu, HE Jiaxin

2024, 42(5):  757-768.  doi:10.3969/j.issn.0255-8297.2024.05.004

Asbtract ( 159 )   PDF (4221KB) ( 65 )  

References | Related Articles | Metrics

Aluminum alloy casting has been widely used in automobile, aircraft and other important industries, where its quality directly affects the safety of mechanical parts. Aiming at the diversification and minuteness of defects in the surface and interior of the X-ray images of aluminum alloy casting, a casting defect detection method based on local and global features was proposed. Firstly, an efficient channel attention module efficient channel attention is fused with the classical network resnet-50 to form a new basic convolutional neural network, which serves as the backbone for constructing a double-branch network model. Then, a detailed information location and extraction (DILE) module is proposed, which located the local area containing rich discriminant information. Finally, combining the local image obtained by DILE with the original image as the input to the network, a double branch network model integrating local and global features is constructed. The global region learning aids in extracting meaningful subtle information in complex background, while the learning of local region further improves the classification effectiveness. The method was tested and trained on an X-ray image data set of real automobile castings, achieving a test set accuracy of 98.3%. Experimental results show that this method is more effective than conventional methods.

Fuzzy Generalized Deduplication Based on Lightweight Secure Cloud Storage Method for Images

CHEN Haixin, TANG Xin, JIN Luchao, FU Yaowen, ZHOU Yiteng

2024, 42(5):  769-781.  doi:10.3969/j.issn.0255-8297.2024.05.005

Asbtract ( 158 )   PDF (3100KB) ( 68 )  

References | Related Articles | Metrics

Generalized deduplication is an effective technique to achieve secure deduplication for cloud images. However, the existing generalized deduplication technique only supports precise deduplication and cannot be integrated with encryption technique for images. Furthermore, image encryption techniques impose substantial computational overhead on clients. To deal with the above challenges, we propose a fuzzy generalized deduplicationbased lightweight secure cloud storage method for images. Firstly, the integer wavelet transform is applied to extract the low-frequency components as bases and high-frequency ones as deviations. By proposing a lightweight encryption scheme based on XOR, the confidentiality protection for images is effectively integrated with the generalized deduplication technique. In addition, the proposed scheme also supports fuzzy deduplication for deviations, ensuring that the cloud service provider only stored a single copy of highly similar deviations, thereby achieving fuzzy generalized deduplication for cloud images. Finally, we conduct experiments on related image datasets. The results show that the proposed scheme significantly improves both communication and storage efficiency while ensuring security.

Attack Towards Speaker Identification Using Deep Conversion Networks for Voiceprint Features

TAO Ziyu, SU Zhaopin, LIAN Chensi, WANG Niansong, ZHANG Guofu

2024, 42(5):  782-794.  doi:10.3969/j.issn.0255-8297.2024.05.006

Asbtract ( 133 )   PDF (7730KB) ( 88 )  

References | Related Articles | Metrics

In the field of speaker identification (SID) systems, attacks often rely on fast gradient descent and mapping gradient descent algorithms, which suffer from unstable attack performance and poor auditory quality of generated attack samples. This paper proposes an advanced attack method against SID systems using deep neural networks to generate attack speeches with the target speaker’s voiceprint. Specifically, the attack process on SID system is first analyzed to determine the approach to generating attack speeches. Then, a two-dimensional convolutional neural network is designed as a generator to effectively integrate the speech content of the source speaker and the voiceprint features of the target speaker. A discriminator is designed based on adversarial learning to improve the quality of the attack speeches. Finally, comparative experiments are conducted on two automatic SID systems based on generalized end-to-end loss and AMSoftmax loss, respectively. Experimental results demonstrate that the proposed method not only improves the stability of attack performance, but also enhances the auditory quality of attack speeches. Moreover, the proposed method is applicable to short samples, making it suitable for practical attack scenarios.

Computer Science and Applications

Autonomous Driving Algorithm Based on Meta-Learning and Reinforcement Learning

JIN Yanliang, FAN Baorong, GAO Yuan, WANG Xiaoyong, GU Chenjie

2024, 42(5):  795-809.  doi:10.3969/j.issn.0255-8297.2024.05.007

Asbtract ( 175 )   PDF (5419KB) ( 101 )  

References | Related Articles | Metrics

To address the problems of convergence difficulty, unsatisfactory training effect and poor generalization performance of autonomous driving algorithms based on reinforcement learning, an autonomous driving system based on meta-learning and reinforcement learning is proposed in this paper. The system first combines variational auto encoder (VAE) with Wasserstein generative adversarial network incorporating gradient penalty (WGAN-GP) to form the VWG (VAE-WGAN-GP) model, which improves the quality of extracted feature. Then, the meta learning algorithm Reptile is used to train the VWG feature extraction model, yielding the MVWG (Meta-VWG) feature extraction model. This approach accelerates the training speed. Finally, the feature extraction model is combined with the proximal policy optimization (PPO) decision algorithm, and the reward function in the PPO algorithm is refined to enhance the convergence speed of the decision model, resulting in the MVWG-PPO autonomous driving model. Experimental results show that compared with VAE, VW (VAE-WGAN) and VWG benchmark models, the MVWG feature extraction model proposed in this paper reduces reconstruction loss by 60.82%, 44.73%, and 29.09%, respectively. The convergence rate increases approximately fivefold, achieving clearer reconstructed images and superior performance in automatic driving tasks. It can provide higher-quality feature information for autonomous vehicles. Meanwhile, compared with the benchmark decision model, the improved reward function model exhibits an 11.33% increase in convergence rate, which fully demonstrating the superiority of the proposed method.

News Recommendation Algorithm Incorporating Headline Sentiment and Topic Characteristics

AI Jun, HONG Xingqi

2024, 42(5):  810-822.  doi:10.3969/j.issn.0255-8297.2024.05.008

Asbtract ( 187 )   PDF (1748KB) ( 271 )  

References | Related Articles | Metrics

Traditional lexicon-based news recommendation algorithms often ignore the emotional nuances present in words beyond the confines of the dictionary. This oversight can lead to issues such as diminished prediction accuracy and subpar sorting performance. To address these challenges, this paper introduces a heuristic approach to deduce the sentiment of unfamiliar words and devises a news recommendation algorithm to verify its feasibility. A tripartite graph model is constructed to propagate sentiment from a sentiment dictionary to individual words and obtain the headline sentiment. In addition, the bag-of-words model is used to extract topic features from the headlines. The sentiment similarity and topic similarity between headlines are calculated, consolidating these into a comprehensive similarity evaluation index. The news with higher similarity to the target news is then selected as the neighbor. The algorithm predicts the hourly average click volume of the target news by considering the hourly average click volume of neighbors, treating this as the predicted score for the target news. Finally, users receive a selection of high-scoring news articles. Validation using real data from NetEase News confirms the feasibility and effectiveness of our algorithm. Compared with other algorithms, our algorithm has shown improvements in the optimal accuracy of mean absolute error (MAE) by 2.2% to 3.4%, root mean square error (RMSE) by 2.3% to 2.9%, and the mean score of normalized discounted cumulative gain (NDCG) by 0.7% to 1.8%, respectively.

Filter Rod Quality Inspection Algorithm Based on Multi-scale Feature Fusing Attention Mechanism

DIAO Yueqin, LI Zhiwen, SHAN Ziqi, LI Fan

2024, 42(5):  823-836.  doi:10.3969/j.issn.0255-8297.2024.05.009

Asbtract ( 170 )   PDF (2494KB) ( 68 )  

References | Related Articles | Metrics

To address the high cost and low efficiency caused by manual monitoring and adjustment of equipment parameters in the production of cigarette filter rods, a filter rod quality inspection algorithm with multi-scale feature fusing attention mechanism is proposed. The algorithm uses one-dimensional convolutions of various sizes to obtain the multi-scale features of the filter rod, which greatly reduces the possibility of missed detection of local features. To further enhance the feature representation, this paper introduces an attention mechanism, enabling the model to focus on regions with richer feature information. Experimental results show that compared with five existing methods such as 1D convolutional neural network (1DCNN), back propagation (BP) neural network and recurrent neural network (RNN), the proposed algorithm shows superior performance on the test set, especially compared with the three methods of 1DCNN, BP neural network and extreme gradient boosting. Specifically, the accuracy of the model increases by 3.27%, 4.28% and 5.70%; the precision improves by 3.12%, 4.68% and 5.10%; the recall rate rises by 3.28%, 4.57% and 2.97%; and the F1-Score enhances by 3.20%, 4.13% and 4.33%, respectively. In general, the algorithm proposed in this paper not only increases the production efficiency of filter rods, but also reduces labor costs, demonstrating practical engineering value.

NBATMAN-ADV Routing Protocol for Large-Scale Flying Ad Hoc Networks

WANG Cong, ZHAO Jihang, WU Xia, MA Wenfeng, TIAN Hui

2024, 42(5):  837-846.  doi:10.3969/j.issn.0255-8297.2024.05.010

Asbtract ( 161 )   PDF (3107KB) ( 119 )  

References | Related Articles | Metrics

Flying ad hoc network is a hot topic in current research, particularly concerning the design of routing mechanisms. The primary challenge lies in managing routing overhead, which can lead to network collapse as the number of UAV nodes increases. To address this issue in large-scale UAVs scenarios, a virtual backbone network is constructed using the unifying connected dominating set algorithm, thereby reducing the number of nodes in route flooding. Next, the NBATMAN-ADV (new better approach to mobile ad-hoc networking-advanced) routing protocol is deployed on the backbone nodes. This protocol evaluates link quality using the received signal strength index and signal-to-noise ratio of the physical layer data, enabling rapid detection of link changes while reducing the routing overhead. Simulation results show that the proposed routing protocol has significantly improved packet delivery rate, end-to-end delay and throughput compared with traditional proactive routing protocols such as optimized link state routing and destination-sequenced distance vector. Experimental results on communication module show that the proposed routing protocol exhibits superior performance in terms of multi-hop delay.

RUL Prediction Model Combined with Transformer

ZHENG Hong, LIU Wen, QIU Junjie, YU Jinhao

2024, 42(5):  847-856.  doi:10.3969/j.issn.0255-8297.2024.05.011

Asbtract ( 223 )   PDF (711KB) ( 145 )  

References | Related Articles | Metrics

Remaining useful life (RUL) prediction is crucial for prognostics and health management of large equipment. However, nonlinear characteristics such as high dimensionality, large scale, strong coupling, and time-varying parameters in monitoring data of some devices can lead to low accuracy in RUL prediction. To solve this problem, this paper introduces a neural network model that combines a transformer decoder with a multiscale bi-directional long and short-term memory network. This model improves prediction accuracy of the model by integrating global information through a multi-head attention mechanism. Using aviation engines as the research focus, comparative experiments were conducted employing various models on NASA’s C-MPASS dataset. The results show that the proposed multi-scale bi-directional long and short-term memory network fused with Transformer model (MSBiLSTM-Transformer) outperforms other benchmark models, demonstrating superior performance in both accuracy and root mean square error metrics.

Highway Toll Evasion Patterns Identification Based on RFE-OPTUNA-XGBoost Model

MA Feihu, LEI Haoan, SUN Cuiyu, LUO Jiajie

2024, 42(5):  857-870.  doi:10.3969/j.issn.0255-8297.2024.05.012

Asbtract ( 134 )   PDF (1156KB) ( 92 )  

References | Related Articles | Metrics

Driven by economic benefits, highway toll evasion behavior occurs frequently in China. This study utilizes anonymized toll data from a specific region in 2020 to address this issue. Through data mining to analyze the behavior characteristics of the evading vehicles, we propose a toll evasion pattern recognition model based on RFE-OPTUNA-XGBoost. The accuracy of this recognition model reaches 0.945, with average AUC values for different evasion methods as follows: large vehicle misclassification at 0.997, U/J-turn evasion at 0.980, fake green pass at 0.969, and gate crashing at 0.924. The results demonstrate that the RFE-OPTUNA-XGBoost model achieves higher accuracy in toll evasion pattern recognition and higher AUC values for each evasion method. In summary, the proposed model can accurately identify toll evasion patterns, offering significant practical value for highway management departments in conducting inspections and preventing toll evasion.

Lane Line Detection Based on CNN and Transformer Hybrid Network

TANG Hong, DENG Feng, ZHANG Kai, NIE Xuefang, LI Guanghui

2024, 42(5):  871-883.  doi:10.3969/j.issn.0255-8297.2024.05.013

Asbtract ( 195 )   PDF (1647KB) ( 75 )  

References | Related Articles | Metrics

Lane detection technology plays a crucial role in autonomous driving systems. Currently, deep learning-based methods for lane detection typically involve extracting features from a backbone network, followed by confidence estimation of key points on the lane lines and their offsets relative to a starting point. However, existing backbone networks struggle to effectively capture features of elongated lanes, and offset networks face challenges in regressing the offsets of key points along the lane line. In this paper, we propose a hybrid network model called CTNet (CNN-Transformer hybrid network) based on a pointbased lane detection approach. CTNet enhances feature representation through a feature pyramid network and an augmented coordinate attention mechanism. Additionally, it employs a vision transformer-based offset network to regress crucial offsets. Consequently, CTNet extracts elongated lane line features, captures long-range offsets between points, and significantly improves the accuracy of lane detection. Experiments conducted on the TuSimple and CULane datasets demonstrate that CTNet outperforms six commonly used lane detection algorithms across various accuracy metrics. Specifically, CTNet achieves superior results on TuSimple across all evaluation metrics. Furthermore, when tested across nine different lane scenarios in the CULane dataset, CTNet achieves the highest accuracy in six scenarios.

Cross-Modal Person Re-identification Driven by Cross-Channel Interactive Attention Mechanism in Dual-Stream Networks

HE Lei, LI Fengyong, QIN Chuan

2024, 42(5):  884-892.  doi:10.3969/j.issn.0255-8297.2024.05.014

Asbtract ( 197 )   PDF (4390KB) ( 391 )  

References | Related Articles | Metrics

Existing cross-modal person re-identification methods often fail to take into account the difference of target person between modes and within modes, making it difficult to further improve the retrieval accuracy. To solve this problem, this paper introduces the cross-channel interaction attention mechanism to enhance the robust extraction of person features, effectively suppresses the extraction of irrelevant features and achieves more discriminative feature expression. Furthermore, hetero-center triplet loss, triplet loss and identity loss are combined for supervised learning, effectively integrating the intermodal and intra-class differences in person features. Experimental results demonstrate the effectiveness of the proposed method, which outperforms seven existing methods on two standard datasets, RegDB and SYSU-MM01.

Electronic Engineering

Design of Planar Scanning System for Electromagnetic Near-Field Testing

JIA Hongchuan, CHENG Xin, WAN Fayu, RAVELO Blaise

2024, 42(5):  893-902.  doi:10.3969/j.issn.0255-8297.2024.05.015

Asbtract ( 199 )   PDF (6118KB) ( 130 )  

References | Related Articles | Metrics

This paper develops a design of near-field scanning system for measuring the electromagnetic (EM) field radiated by electronic devices for EM compatibility. Firstly, a magnetic field probe implemented on four-layer printed circuit board (PCB) structure is designed, with a working frequency of up to 12 GHz and a spatial resolution of 2 mm. The simulation results match well with the experimental measurements, and the probe is calibrated accordingly. Secondly, automation of the near-field scanning is achieved by designing a position machine using LabVIEW, with the STM32 serving as the motion control core. The STM32 receives the positioning data through the serial port, and controls the stepper motor to drive the probe for fixed-point scanning. The position machine communicates with the vector network analyzer through the local area network to read and save data. Finally, the data is visualized by the position machine. The results are calibrated upon completion of the scanning process, producing a real-time visualization of the field distribution map of the tested object. The measured field strength results show good agreement with electromagnetic simulation results, demonstrating the system’s suitability for analyzing electromagnetic coupling paths and transitioning between near-field and farfield regions.