Table of Content

10 December 2024, Volume 44 Issue 12 Cover Download Catalog Download

Previous Issue   

2023 CCF China Blockchain Conference （CCF CBCC 2023）

Development， technologies and applications of blockchain 3.0

Peng FANG, Fan ZHAO, Baoquan WANG, Yi WANG, Tonghai JIANG

2024, 44(12):  3647-3657.  DOI: 10.11772/j.issn.1001-9081.2023121826

Asbtract ( )   HTML ( )   PDF (2294KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Blockchain 3.0 is the third stage of the development of blockchain technology and the core of building a value Internet. Its innovations in sharding， cross-chain and privacy protection have given it a wide range of application scenarios and research value. It is highly valued by relevant people in academia and industry. For the development， technologies and applications of blockchain 3.0， the relevant literature on blockchain 3.0 at home and abroad in the past five years were surveyed and reviewed. Firstly， the basic theory and technical characteristics of blockchain were introduced， laying the foundation for an in-depth understanding of the research progress of blockchain. Subsequently， based on the evolution trend of blockchain technology over time， the development process and various key development time nodes of blockchain 3.0， as well as the reasons of the division of different stages of development of blockchain using sharding and side-chain technologies as benchmarks， were given. Then， the current research status of key technologies of blockchain 3.0 was analyzed in detail， and typical applications of blockchain 3.0 in six major fields such as internet of things， medical care， and agriculture were summarized. Finally， the key challenges and future development opportunities faced by blockchain 3.0 in its development process were summed up.

Overview of on-chain and off-chain consistency protection technologies

Tingting GAO, Zhongyuan YAO, Miao JIA, Xueming SI

2024, 44(12):  3658-3668.  DOI: 10.11772/j.issn.1001-9081.2023121818

Asbtract ( )   HTML ( )   PDF (1452KB) ( )  

Figures and Tables | References | Related Articles | Metrics

The immutable characteristic of blockchain ensures the consistency of on-chain data， but off-chain data may be destroyed in the process of recording， storage， and transmission， causing inconsistency between on-chain and off-chain data， which greatly affects the implementation and development of blockchain. Therefore， it requires some mechanisms to ensure the consistency between on-chain and off-chain data. In view of the problem of inconsistency between on-chain and off-chain data， some current on-chain and off-chain consistency protection technologies were summarized. Firstly， the basic concept of on-chain and off-chain consistency was introduced， and the importance of consistency was pointed out. Secondly， the technologies of on-chain and off-chain consistency protection were summed up from three aspects： oracle mechanism， data integrity mechanism and on-chain and off-chain data collaboration mechanism， and some on-chain and off-chain consistency protection schemes were compared and analyzed. Finally， the technologies of on-chain and off-chain consistency protection were prospected from three aspects， providing a theoretical reference for blockchain practitioners and researchers to further discuss and study the on-chain and off-chain consistency protection methods， which could promote the implementation of blockchain applications.

Progress and prospect of zero-knowledge proof enabling blockchain

Miao JIA, Zhongyuan YAO, Weihua ZHU, Tingting GAO, Xueming SI, Xiang DENG

2024, 44(12):  3669-3677.  DOI: 10.11772/j.issn.1001-9081.2023121819

Asbtract ( )   HTML ( )   PDF (3019KB) ( )  

Figures and Tables | References | Related Articles | Metrics

With decentralized and immutable characteristics， blockchain technology has been widely applied in the fields such as social governance， education， and healthcare. However， as the applications deepen and the demands increase， a number of challenges such as security issues， insufficient storage， and silo of value are faced by blockchain systems. Consequently， the needs for blockchain in terms of privacy protection， expansion， and cross-chain interoperability are becoming prominent increasingly. Enabling blockchain with Zero-Knowledge Proof （ZKP） technology can enable advanced anonymity and transaction privacy protection. At the same time， KPZ-based effectiveness proof can replace complete data submitted by side-chain or off-chain roles to the main chain， and KPZ-based consensus proof can improve the performance of blockchain cross-chain protocols significantly. Aiming at the urgent need for comprehensive comparative analysis of the current situation of ZKP enabling blockchain， a research on the progress and prospect of ZKP enabling blockchain was carried out， and based on the related progress of ZKP technology enabling blockchain， representative solutions in recent years were summarized systematically. Firstly， the development context of zero-knowledge technology was introduced. Then， representative applications of blockchain based on ZKP technology were categorized and summed up， and the realization ideas and innovation points of these applications were introduced emphatically. At the same time， based on typical cases， these applications were analyzed in the performance under indicators such as block size， proof size and transaction cost. Finally， the applications of ZKP technology in the development of blockchain privacy protection， expansion and cross-chain development opportunities were prospected.

Review of blockchain consensus algorithms for internet of things

Kedi NIU, Min LI, Zhongyuan YAO, Xueming SI

2024, 44(12):  3678-3687.  DOI: 10.11772/j.issn.1001-9081.2023121820

Asbtract ( )   HTML ( )   PDF (1915KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Most of the current consensus algorithms require high computing power or specific communication environment and are not suitable for resource-constrained Internet of Things （IoT）. In view of the limitations of the traditional consensus algorithms in blockchain applied to IoT， blockchain consensus algorithms for IoT were reviewed. Firstly， the consensus algorithms for IoT were introduced and summarized from three categories： improved consensus algorithms based on Practical Byzantine Fault Tolerance （PBFT）， improved algorithms based on other consensus algorithms， and new blockchain consensus algorithms suitable for IoT. Secondly， a basic evaluation index system of consensus algorithms was established to compare consensus algorithms from five aspects： decentralization， scalability， security， latency， and energy consumption. Finally， the challenges and future research directions of consensus algorithms for IoT were analyzed. Analysis of the basic evaluation index system shows that the new blockchain consensus algorithms are more suitable for IoT than the improved consensus algorithms based on the traditional consensus algorithms， providing a reference for research on blockchain consensus algorithms for IoT.

Review of blockchain lightweight technology applied to internet of things

Ziqian CHEN, Kedi NIU, Zhongyuan YAO, Xueming SI

2024, 44(12):  3688-3698.  DOI: 10.11772/j.issn.1001-9081.2023121817

Asbtract ( )   HTML ( )   PDF (1709KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Blockchain technology is used in Internet of Things （IoT） due to its characteristics of decentralization and encryption. However， the traditional blockchain has problems such as poor scalability， high latency， high consumption of cryptographic components， complex consensus computing and large data storage scale in IoT environment. In view of the poor performance of traditional blockchain in IoT devices， the lightweight technology of blockchain was reviewed. Firstly， the blockchain architectures were divided into single-chain structure and Directed Acyclic Graph （DAG） structure， and the lightweight operations in the two types of blockchain architectures were compared. Secondly， the lightweight hash function was analyzed from the perspectives of iterative structure， compression function and hardware implementation. Thirdly， the consensus algorithms and the lightweight schemes in storage were introduced. Finally， the design ideas of blockchain lightweight technology were summarized based on the literature research results， and the future research directions were prospected.

Multi-authority attribute-based encryption scheme for private blockchain over public blockchain

Keshuo SUN, Haiying GAO, Yang SONG

2024, 44(12):  3699-3708.  DOI: 10.11772/j.issn.1001-9081.2023121816

Asbtract ( )   HTML ( )   PDF (1372KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Blockchain is an emerging data structure that combines distributed storage， consensus mechanisms， principles of cryptography， and other technologies. Aiming at the problem of data access control on public blockchain， a Multi-Authority Attribute-Based Encryption （MA-ABE） scheme for Private blockchain Over Public blockchain （POP） was proposed. Firstly， the private blockchain based on public blockchain was constructed， and a specific data privacy protection process was given. Secondly， a joint authority initialization algorithm and a joint key generation algorithm were designed. Finally， the static security of the proposed scheme was proved by using the Decisional Bilinear Diffie-Hellman （DBDH） assumption through simulating the challenge ciphertext and the user private key under the random oracle model. Experimental results show that the proposed scheme can resist collusion attacks under n-\mathrm{1} parties of malicious authorities targeting the master key.

Linkable ring signature scheme based on SM9 algorithm

Yiting WANG, Wunan WAN, Shibin ZHANG, Jinquan ZHANG, Zhi QIN

2024, 44(12):  3709-3716.  DOI: 10.11772/j.issn.1001-9081.2023121825

Asbtract ( )   HTML ( )   PDF (1254KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Aiming at the problem that the Identity-Based Linkable Ring Signature （IBLRS） scheme has excessive overhead and does not meet the requirements of technical autonomy， a Linkable Ring Signature （LRS） scheme based on SM9 algorithm was proposed. Firstly， the identifier of the signer in the ring was sent to the Key Generation Center （KGC） to generate the corresponding private key. Secondly， the private key was combined with SM9 algorithm to generate a signature， and this private key generation method was consistent with the private key generation method in SM9 algorithm. Finally， the signer's private key and the event identifier were bound to construct a linkable label without need of complex calculation operations， which improved the efficiency of the proposed algorithm. Under the random oracle model， it was proved that the proposed scheme has correctness， unforgeability， unconditional anonymity and linkability. At the same time， a multi-notary cross-chain scheme was designed on the basis of the proposed algorithm to achieve efficient and safe cross-chain interaction. Compared with the IBLRS algorithm， the proposed scheme only requires 4 bilinear pairing operations， which reduces the computational overhead and communication overhead by 39.06% and 51.61% respectively. Performance analysis of the scheme shows that the proposed scheme reduces computing overhead and communication overhead， and satisfies the autonomous controllability of the technology.

Delegated proof of stake consensus algorithm based on reputation value and strong blind signature algorithm

Zhenhao ZHAO, Shibin ZHANG, Wunan WAN, Jinquan ZHANG, zhi QIN

2024, 44(12):  3717-3722.  DOI: 10.11772/j.issn.1001-9081.2023121822

Asbtract ( )   HTML ( )   PDF (1330KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In order to address the issues of Delegated Proof of Stake （DPoS） algorithm， such as the growing centralization trend caused by high-weight nodes having a higher probability of accounting rights， low voting enthusiasm among nodes， and collusion attacks caused by node corruption， a DPoS consensus algorithm based on reputation value and strong blind signature algorithm was proposed. Firstly， the nodes were sorted into two types based on the initial conditions， and the initial selection of nodes was carried out to select the proxy nodes. Secondly， the vote for each other was performed among the proxy nodes， and the top 21 nodes were selected to form the witness node set based on the average of historical reputation value and final number of votes， while the remaining nodes were used to form the standby witness node set. During the voting process， an Elgamal-based strong blind signature algorithm was employed to ensure privacy for voting nodes. Finally， consensus process was achieved after block out of witness nodes. Experimental results demonstrate that compared to the original DPoS consensus algorithm， the proposed algorithm increases active node proportion by approximately 20 percentage points， and reduces malicious node proportion close to zero. It can be observed that the proposed algorithm enhances node enthusiasm in voting and protects privacy information of nodes.

Cross-chain identity management scheme based on identity-based proxy re-encryption

Xin ZHANG, Jinquan ZHANG, Deyuan LIU, Wunan WAN, Shibin ZHANG, Zhi QIN

2024, 44(12):  3723-3730.  DOI: 10.11772/j.issn.1001-9081.2023121823

Asbtract ( )   HTML ( )   PDF (2492KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In view of the current problems of low authentication efficiency， insufficient security performance and poor scalability in cross-chain identity management， a cross-chain identity management scheme based on Identity-Based Proxy Re-Encryption （IBPRE） was proposed. Firstly， an identity chain was built combining Decentralized IDentifier （DID）， and DIDs were provided as cross-chain identity identifiers and verifiable certificates were provided as access certificates to the users to build an access control policy based on certificate information. Secondly， the relay chain was combined with the cryptographic accumulator to achieve user identity authentication. Finally， by combining IBPRE and signature algorithm， a cross-chain communication model based on IBPRE was constructed. Experimental analysis and evaluation results show that compared with RSA （Rivest-Shamir-Adleman algorithm） and Elliptic Curve Cryptosystem （ECC）， the proposed scheme has the authentication time reduced by 66.9% and 4.8% respectively. It can be seen that relay chain and identity chain can realize identity management， improve decentralization and scalability， build cross-chain communication models and access policies based on certificate information， and ensure security in cross-chain identity management.

Cross-chain identity authentication scheme based on certificate-less signcryption

Deyuan LIU, Jingquan ZHANG, Xing ZHANG, Wunan WAN, Shibin ZHANG, Zhi QIN

2024, 44(12):  3731-3740.  DOI: 10.11772/j.issn.1001-9081.2023121824

Asbtract ( )   HTML ( )   PDF (2361KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In response to the issues of low decentralization， poor scalability， and high resource consumption in the current blockchain cross-chain identity authentication schemes， a Cross-chain Identity Authentication scheme based on Certificate-Less SignCryption （CIA-CLSC） was proposed. Firstly， Certificate-Less SignCryption （CLSC） was utilized to generate keys for cross-chain entities， realize communication encryption， and perform identity authentication. Secondly， secret sharing was employed for key management in the distributed system. Finally， decentralized identities were used to establish the association between entity keys and cross-chain identities. Under the premise of ensuring identity privacy and security， CIA-CLSC achieved cross-chain interactive identity authentication among different blockchain systems. Theoretical analysis and experimental results demonstrate that CIA-CLSC does not rely on centralized certificate authorities or third-party key management organizations， ensuring decentralization； the CIA-CLSC generated digital identities comply with the World Wide Web Consortium （W3C） standards， ensuring scalability. Furthermore， compared to the combination of ECC （Elliptic Curve Cryptography） and AES （Advanced Encryption Standard）， CIA-CLSC achieves approximately 34% reduction in time overhead； compared to the combination of RSA （Rivest-Shamir-Adleman algorithm） and AES， CIA-CLSC achieves approximately 38% reduction in time overhead while maintaining decentralization for cross-chain interactive identity authentication. It can be seen that CIA-CLSC can enhance the decentralization， scalability， and interaction efficiency of cross-chain systems in practical applications effectively.

Highway free-flow tolling method based on blockchain and zero-knowledge proof

Yifan WANG, Shaofu LIN, Yunjiang LI

2024, 44(12):  3741-3750.  DOI: 10.11772/j.issn.1001-9081.2023121830

Asbtract ( )   HTML ( )   PDF (2979KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In response to the issues of vehicle toll evasion caused by license plate cloning and potential user privacy leaking to centralized entity due to centralized data storage in the current intelligent transportation highway free-flow tolling schemes， a highway free-flow tolling method based on blockchain and zero-knowledge proof was developed. Initially， a video surveillance mechanism for toll evasion detection was designed to ensure the compliance of vehicles on highways. Subsequently， smart contracts in the blockchain were designed to encrypt and store vehicle Location Certificate （LC） and payment data in a distributed ledger， and zero-knowledge proof technology was introduced to ensure the correctness of payment while protecting privacy. At the same time， an algorithm for charging tolls based on the vehicle's mileage was designed within the zero-knowledge circuit. Theoretical analysis and simulation results demonstrate that under normal conditions， the proposed method can achieve correct tolling based on the actual mileage with zero-knowledge of location privacy， and in the event of exceptions， the proposed method can provide timely warnings and record anomalies on the blockchain； compared to traditional manual tolling method， the proposed method has the average tolling time reduced from 38.0 s to 1.8 s， and has the decrease of about 0.1 s compared to the tolling method combining 5G and electronic non-stop toll collection system ETC （Electronic Toll Collection） in average tolling time. For the same entry and exit， the lower the overlap ratio in the number of information network trusted third-party Information Collection Point （ICP） of different routes， the more accurate the mileage-based tolling.

Traceability storage model of charity system oriented to master-slave chain

Jing LIANG, Wunan WAN, Shibin ZHANG, Jinquan ZHANG, Zhi QIN

2024, 44(12):  3751-3758.  DOI: 10.11772/j.issn.1001-9081.2023121821

Asbtract ( )   HTML ( )   PDF (2966KB) ( )  

Figures and Tables | References | Related Articles | Metrics

The traceability data of single-chain storage charity system have huge storage pressure， and the charity data need to be shared， may leading to the problem of privacy leakage. Therefore， a charity system traceability storage model oriented to master-slave chain was proposed. Firstly， a master chain and several slave chains were designed in the model. The master chain was mainly responsible for the query of charity traceability data and the supervision of slave chains， and the slave chains were responsible for the storage of a large number of charity traceability data. Then， an intelligent contract for the classification of charity traceability data was designed to classify charity data into public data and private data according to privacy requirements. The public data were stored in the master chain directly， while the private data were encrypted with Ciphertext-Policy Attribute-Based Encryption （CP-ABE） and stored in the slave chains， which ensured data privacy， thus achieving storage scalability and intelligence. Finally， the storage structure of Merkle tree was improved. By designing a smart contract to mark duplicate data， the same block detection and duplicate data deletion of blockchain system were completed， which avoided data redundancy and reduced storage consumption. Experimental results show that compared to the single-chain model， with the increase of total number of data， the proposed model has the response time of the master-slave chain stabilized at 0.53 s and the throughput stabilized at 149 B. It can be seen that the master-slave chain model improves search efficiency， optimizes storage space， and realizes data privacy protection.

Artificial intelligence

Federated learning client selection method based on label classification

Zucuan ZHANG, Xuebin CHEN, Rui GAO, Yuanhuai ZOU

2024, 44(12):  3759-3765.  DOI: 10.11772/j.issn.1001-9081.2023121740

Asbtract ( )   HTML ( )   PDF (1918KB) ( )  

Figures and Tables | References | Related Articles | Metrics

As a distributed machine learning method， federated learning can fully exploit the value in the data while protecting data privacy. However， as the traditional federated learning training method only selects participating clients randomly， it is difficult to adapt to Not Identically and Independently Distributed （Non-IID） datasets. To solve the problems of low accuracy and slow convergence of federated learning models under Non-IID data， a Federated learning Client Selection method based on Label Classification （FedLCCS） was proposed. Firstly， the client dataset labels were classified and sorted according to the frequency statistics results. Then， clients with high-frequency labels were selected to participate in training. Finally， models with different accuracy were obtained by adjusting own parameters. Experimental results on MNIST， Fashion-MNIST and Cifar-10 datasets show that the two baseline methods， Federated Averaging （FedAvg） and Federated Proximal （FedProx）， after combining with FedLCCS are better than the original ones under the initial dataset label selection ratio. The minimum accuracy improvements are 9.13 and 6.53 percentage points， the minimum convergence speed improvements are 57.41% and 18.52%， and the minimum running time reductions are 7.60% and 17.62%. The above verifies that FedLCCS can optimize the accuracy， convergence speed and running efficiency of federated models， and can train models with different accuracy to meet diversified demands.

Unsupervised feature selection model with dictionary learning and sample correlation preservation

Jingxin LIU, Wenjing HUANG, Liangsheng XU, Chong HUANG, Jiansheng WU

2024, 44(12):  3766-3775.  DOI: 10.11772/j.issn.1001-9081.2023121783

Asbtract ( )   HTML ( )   PDF (7520KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Focusing on the issue that most unsupervised feature selection models based on dictionary learning cannot fully exploit the intrinsic correlations among data， which reduces the accuracy of feature importance judgment， an unsupervised feature selection model with Dictionary Learning and Sample Correlation Preservation （DLSCP） was proposed. Firstly， the original data were encoded by learning the dictionary atoms， and the latent representations to characterize data distribution were obtained in the dictionary space. Secondly， the intrinsic correlations among data were learned adaptively in the dictionary space to alleviate the influence of redundant and noisy features， thus obtaining accurate local structure among data. Finally， the intrinsic correlations among data were used to measure the relevance and importance of data features. Experimental results on TOX dataset show that， when selecting 50 features， DLSCP improves the Normalized Mutual Information （NMI） and clustering Accuracy （Acc） by 13.33 and 7.95 percentage points respectively compared to non negative spectral analysis model NDFS（Nonnegative Discriminative Feature Selection） and by 15.74 and 7.31 percentage points respectively compared to unsupervised feature selection model with hidden space embedding LSEUFS （Latent Space Embedding for Unsupervised Feature Selection via joint dictionary learning）， which verifies the effectiveness of DLSCP.

Image-text retrieval model based on intra-modal fine-grained feature relationship extraction

Zucheng WU, Xiaojun WU, Tianyang XU

2024, 44(12):  3776-3783.  DOI: 10.11772/j.issn.1001-9081.2023121860

Asbtract ( )   HTML ( )   PDF (3093KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In response to the diversity of relationships in cross-modal retrieval tasks， and the poor application effect of the traditional paradigm based on appearance in complex scenarios caused by inaccurately reflecting the relationships between significant objects in images， an image-text retrieval model based on intra-modal fine-grained feature relationship extraction was proposed. Firstly， to obtain more intuitive position information， the image was divided into grids， and position representations were established on the basis of the relationships between objects and grids. Then， to maintain the stability and independence of node information during the relationship modeling stage， a cross-modal information-guided feature fusion module was utilized. Finally， an adaptive triplet loss was proposed to balance the training weights of positive and negative samples dynamically. Experimental results demonstrate that compared with the model CHAN （Cross-modal Hard Aligning Network）， on the Flickr30K and MS-COCO 1K datasets， the proposed model improves 1.5% and 0.02% in R@sum metric （the sum of R@1， R@5， R@10 for image-to-text retrieval and text-to-image retrieval tasks）， respectively. The above results prove the effectiveness of the proposed model in retrieval recall.

Incomplete multi-view clustering algorithm based on attention mechanism

Chenghao YANG, Jie HU, Hongjun WANG, Bo PENG

2024, 44(12):  3784-3789.  DOI: 10.11772/j.issn.1001-9081.2023121866

Asbtract ( )   HTML ( )   PDF (1333KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In order to solve the problems of uncertainty in completing missing view data， lack of robustness of embedding learning and low model generalization in traditional deep incomplete multi-view clustering algorithms， an Incomplete Multi-View Clustering algorithm based on Attention Mechanism （IMVCAM） was proposed. Firstly， K-Nearest Neighbors （KNN） algorithm was used to complete the missing data in the view， making the training data complementary. Then， after passing the linear encoding layer， the obtained embedding was passed through the attention layer to improve the quality of the embedding. Finally， the embedding obtained from the training of each view was clustered using k-means clustering algorithm， and the weights of the views were determined by the Pearson correlation coefficient. Experimental results on five classic datasets show that， the optimal result was achieved by IMVCAM on Fashion dataset， compared with the sub-optimal Deep Safe Incomplete Multi-View Clustering （DSIMVC） algorithm， IMVCAM improves the clustering accuracy by 2.85 and 4.35 percentage points respectively when the data missing rate is 0.1 and 0.3. Besides， on Caltech101-20 dataset， the clustering accuracy of IMVCAM is increased by 7.68 and 3.48 percentage points respectively compared to that of the sub-optimal algorithm IMVCSAF （Incomplete Multi-View Clustering algorithm based on Self-Attention Fusion） when the missing rate is 0.1 and 0.3. The proposed algorithm can effectively deal with the incompleteness of multi-view data and the problem of model generalization.

Few-shot object detection method based on improved region proposal network and feature aggregation

Keyi FU, Gaocai WANG, Man WU

2024, 44(12):  3790-3797.  DOI: 10.11772/j.issn.1001-9081.2023121731

Asbtract ( )   HTML ( )   PDF (3013KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In the existing few-shot object detection， Region Proposal Network （RPN） is usually trained on base class data to generate new class anchor boxes. However， new class data are more sparse compared to the base class. Introducing new class data may lead to the presence of complex backgrounds different to the objects， causing RPN to misclassify the background as foreground， resulting in the omission of high Intersection over Union （IoU） value anchor boxes. To address the above issues， a Few-Shot Object Detection method based on Improved RPN and Feature Aggregation （IFA-FSOD） was proposed. Firstly， an improvement was made on the basis of RPN by incorporating a metric-based non-linear classifier within RPN. This classifier was designed to compute the similarity between features extracted by the backbone network and the features representing the new class， so as to increase the recall for anchor boxes of the new class， thereby filtering out high IoU value anchor boxes. Then， a Feature Aggregation Module （FAM） based on attention mechanism was introduced in Region of Interest Alignment （RoI Align）. And by designing grids of different scales， more comprehensive information and feature representation were obtained， which alleviated the lack of feature information caused by different scales. Experimental results show that compared with QA-FewDet （Query Adaptive Few-shot object Detection） method， IFA-FSOD method improves nAP50（Novel Average Precision at 50% IoU） by 4.5 percentage points under Novel Set 3's 10-shot on the new class of PASCAL VOC dataset； compared with FsDetView （Few-shot object Detection and Viewpoint estimation） method， under the settings of 10-shot and 30-shot， IFA-FSOD method has mean Average Precision （mAP） increased by 0.2 and 0.8 percentage points， respectively， on the new class of COCO dataset. It can be seen that Improved RPN and Feature Aggregation （IFA） can improve the detection performance of object classes in the case of few-shot effectively， and solve the problem of missing high IoU value anchor boxes and incomplete feature information capture.

Fast adversarial training method based on data augmentation and label noise

Yifei SONG, Yi LIU

2024, 44(12):  3798-3807.  DOI: 10.11772/j.issn.1001-9081.2023121835

Asbtract ( )   HTML ( )   PDF (1924KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Adversarial Training （AT） has been an effective defense approach for protecting classification models against adversarial attacks. However， high computational cost of the generation of strong adversarial samples during the training process may lead to significantly large extra training time. To overcome this limitation， Fast Adversarial Training （FAT） based on single-step attacks was explored. Previous work improves FAT from different perspectives， such as sample initialization， loss regularization， and training strategies. However， Catastrophic Overfitting （CO） will be encountered when dealing with large perturbation budgets. Therefore， an FAT method based on data augmentation and label noise was proposed. Firstly， multiple image transformations were performed to the original samples and random noise was introduced to implement data enhancement. Secondly， a small amount of label noise was injected. Thirdly， the augmented data were used to generate adversarial samples for model training. Finally， the label noise rate was adjusted adaptively according to the adversarial robustness test results. Comprehensive experimental results on CIFAR-10 and CIFAR-100 datasets show that compared to FGSM-MEP （Fast Gradient Sign Method with prior from the Momentum of all Previous Epoch） method， the proposed method improves 4.63 and 5.38 percentage points respectively on AA （AutoAttack） on the two datasets under the condition of large perturbation budget. The experimental results demonstrate that the proposed method can effectively handle the catastrophic overfitting problem under large perturbation budgets and enhance the adversarial robustness of model significantly.

Efficient active-set method for support vector data description problem with Gaussian kernel

Qiye ZHANG, Xinrui ZENG

2024, 44(12):  3808-3814.  DOI: 10.11772/j.issn.1001-9081.2023121809

Asbtract ( )   HTML ( )   PDF (4044KB) ( )  

Figures and Tables | References | Related Articles | Metrics

To address the large amount of calculation and low efficiency during each iteration in large-scale data scenarios when using active-set method to solve the problem of Support Vector Data Description （SVDD）， an efficient Active-Set Method for SVDD problem with Gaussian kernel （ASM-SVDD） was designed. Firstly， due to the peculiarity of constraint conditions in SVDD dual model， a dimension-reduced subproblem with equality constraints was solved in each iteration. Then， the active-set was updated through matrix manipulations. Each update calculation was only related to the existing support vectors and a single sample point， which reduced the amount of computation dramatically. In addition， since ASM-SVDD algorithm can be seen as a variant of the traditional active-set method， the limited termination of this algorithm was obtained by applying the theory of active-set method. Finally， simulation and real datasets were used to verify the performance of ASM-SVDD algorithm. The results show that ASM-SVDD algorithm can improve the model performance effectively as the number of training rounds increases. Compared to the fast incremental algorithm to solve SVDD problem — FISVDD （Fast Incremental SVDD）， ASM-SVDD algorithm has the objective value obtained by training reduced by 25.9% and the recognition ability of support vectors improved by 10.0% on the typical low-dimensional high-sample dataset shuttle. At the same time， ASM-SVDD algorithm obtains F1 scores on different datasets all higher than FISVDD algorithm with the maximum improvement of 0.07% on the super large-scale dataset criteo. It can be seen that ASM-SVDD algorithm can obtain more stable hypersphere through training， and obtain higher judgment accuracy of test samples while performing outlier detection. Therefore， ASM-SVDD algorithm is suitable for outlier detection in large-scale data scenarios.

Prompt learning method for ancient text sentence segmentation and punctuation based on span-extracted prototypical network

Yingjie GAO, Min LIN, Siriguleng, Bin LI, Shujun ZHANG

2024, 44(12):  3815-3822.  DOI: 10.11772/j.issn.1001-9081.2023121719

Asbtract ( )   HTML ( )   PDF (1509KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In view of the phenomenon that automatic sentence segmentation and punctuation task in ancient book information processing relies on large-scale annotated corpora， and considering that training high-quality， large-scale samples is expensive and these samples are difficult to obtain， a prompt learning method for ancient text sentence segmentation and punctuation based on span-extracted prototypical network was proposed. Firstly， structured prompt information was incorporated into the support set to form an effective prompt template， so as to improve the model's learning efficiency. Then， combined with a punctuation position extractor and a prototype network classifier， the misjudgment impact and the interference from non-punctuation labels in traditional sequence labeling method were effectively reduced. Experimental results show that on Records of the Grand Historian dataset， the F1 score of the proposed method is 2.47 percentage points higher than that of the Siku-BERT-BiGRU-CRF （Siku - Bidirectional Encoder Representation from Transformer - Bidirectional Gated Recurrent Unit - Conditional Random Field） method. In addition， on the public multi-domain ancient text dataset CCLUE， the precision and F1 score of this method reach 91.60% and 93.12% respectively， indicating that the method can perform sentence segmentation and punctuation in multi-domain ancient text effectively and automatically by using a small number of training samples. Therefore， the proposed method offers new thought and approach for conducting in-depth research on automatic sentence segmentation and punctuation， as well as for enhancing the model's learning efficiency， in multi-domain ancient text.

Cyber security

Integrated method of inference and taint analysis for nested branch breakthrough

Jinhui CAI, Zhongxu YIN, Guoxiao ZONG, Junru LI

2024, 44(12):  3823-3830.  DOI: 10.11772/j.issn.1001-9081.2023121738

Asbtract ( )   HTML ( )   PDF (1292KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In view of the problem that the current fuzzing based on taint inference mainly focuses on the analysis of a single code branch in the target code block， but does not fully consider the correlation between context branches， which leads to the inaccurate inference of the relevant byte position of the code branch in the face of nested branches， an integrated method of inference and taint analysis for nested branch breakthrough was proposed. Firstly， the stage coverage information was used to evaluate the obstacle points that needed to be broken， and the priorities of the obstacle points were evaluated according to the coverage information of the obstacle points during the execution of the test cases， so as to focus on the test cases with more potential. Secondly， the taint inference algorithm was optimized， which meant that combined with the control flow information， the position of input bytes related to the nested branch were inferred more accurately， and the pre-order branch inference information was reused to speed up the inference. Finally， a lightweight taint analysis was performed to the inferred obstacle point related positions to guide the mutation process， so as to avoid the nested branch unreachable problem caused by random mutation. The prototyping tool DTFuzz was evaluated in 6 popular applications. Experimental results show that DTFuzz's node coverage rate is 9.85% higher than that of the existing fuzzing tools REDQUEEN averagely， and 5 unknown vulnerabilities are found by this tool. At the same time， compared with the benchmark tool， all of different modules have the coverage rate improved， and the highest improvement is 29.23%. It can be seen that the proposed method can breakthrough the complex nested branches effectively and improve the test coverage rate， as well as improves the efficiency of vulnerability mining.

Dynamic social network privacy publishing method for partial graph updating

Rui GAO, Xuebin CHEN, Zucuan ZHANG

2024, 44(12):  3831-3838.  DOI: 10.11772/j.issn.1001-9081.2023111706

Asbtract ( )   HTML ( )   PDF (3239KB) ( )   PDF（mobile） (4356KB) ( 14 )  

Figures and Tables | References | Related Articles | Metrics

Aiming at the problems of excessive added noise scale and error accumulation during iteration in existing dynamic social network privacy protection， a method named PGU-DNDP （Partial Graph Updating in Dynamic social Network based on Differential Privacy） was proposed. Firstly， the update sequences in the network snapshot graph set were collected through a temporal trade-off dynamic community discovery algorithm. Secondly， a static graph publishing method was used to obtain the initial generated graph. Finally， based on the generated graph of the previous moment and the update sequence of the current moment， the partial graph update was completed. The partial update method could reduce the excessive noise caused by the full graph perturbation and optimize the time cost， thus avoiding the intensive situation of synthetic graph. In addition， an edge updating strategy was proposed in the partial update， which combined the adaptive perturbation with a downsampling mechanism to reduce the cumulative error in the iterative process through privacy amplification， thus improving the synthetic graph accuracy effectively. Experimental results on three synthetic datasets and two real-world dynamic datasets show that PGU-DNDP can ensure the privacy requirements of dynamic social networks while retaining higher data utility than mainstream static graph generation method PrivGraph （differentially Private Graph data publication by exploiting community information）.

Differential and linear characteristic analysis of full-round Shadow algorithm

Yong XIANG, Yanjun LI, Dingyun HUANG, Yu CHEN, Huiqin XIE

2024, 44(12):  3839-3843.  DOI: 10.11772/j.issn.1001-9081.2023121762

Asbtract ( )   HTML ( )   PDF (960KB) ( )  

Figures and Tables | References | Related Articles | Metrics

As Radio Frequency IDentification （RFID） technology and wireless sensors become increasingly common， the need of secure data transmitted and processed by such devices with limited resources leads to the emergence and growth of lightweight ciphers. Characterized by their small key sizes and limited number of encryption rounds， precise security evaluation of lightweight ciphers is needed before putting into service. The differential and linear characteristics of full-round Shadow algorithm were analyzed for lightweight ciphers’ security requirements. Firstly， a concept of second difference was proposed to describe the differential characteristic more clearly， the existence of a full-round differential characteristic with probability 1 in the algorithm was proved， and the correctness of differential characteristic was verified through experiments. Secondly， a full-round linear characteristic was provided. It was proved that with giving a set of Shadow-32 （or Shadow-64） plain ciphertexts， it is possible to obtain 8 （or 16） bits of key information， and its correctness was experimentally verified. Thirdly， based on the linear equation relationship between plaintexts， ciphertexts and round keys， the number of equations and independent variables of the quadratic Boolean function were estimated. After that， the computational complexity of solving the initial key was calculated to be {\mathrm{2}}^{\mathrm{63.4}}. Finally， the structural features of Shadow algorithm were summarized， and the focus of future research was provided. Besides， differential and linear characteristic analysis of full-round Shadow algorithm provides preference for the differential and linear analysis of other lightweight ciphers.

Advanced computing

Improved differential evolution algorithm based on dual-archive population size adaptive method

Yawei HUANG, Xuezhong QIAN, Wei SONG

2024, 44(12):  3844-3853.  DOI: 10.11772/j.issn.1001-9081.2023121744

Asbtract ( )   HTML ( )   PDF (1393KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Addressing the poor performance of population size improvement methods in the existing Differential Evolution （DE） algorithms when dealing with decreased population diversity and local optimum challenges， a dual-Archive Population Size adaptive Differential Evolution algorithm （APDE） was proposed on the basis of dual-Archive Population Size Adaptive method （APSA）. Firstly， two archives were constructed to record individuals that had been discarded in previous evolutions and experimental individuals respectively. Then， diversity changes were measured according to the variations in the population distribution state. And when population diversity decreased， the individuals from the archives were selected and added to the population to enhance the population diversity and the ability to escape from the local optimum. Finally， an improved DE algorithm based on APSA method， APDE， was proposed.Results of extensive tests on CEC2017 test set and Lennard-Jones potential problem show that APDE algorithm outperforms five other DE algorithms in the average ranking based on Friedman test on 30 benchmark functions， and significant improvements are obtained on at least 20% of these functions. At the same time， APDE algorithm also achieves the best performance in solving the minimization of potential energy.

Conflict-based search algorithm for large-scale warehousing environment

Fuqin DENG, Chaoen TAN, Junwei LI, Jiaming ZHONG, Lanhui FU, Jianmin ZHANG, Hongmin WANG, Nannan LI, Bingchun JIANG, Tin Lun LAM

2024, 44(12):  3854-3860.  DOI: 10.11772/j.issn.1001-9081.2023121858

Asbtract ( )   HTML ( )   PDF (2950KB) ( )  

Figures and Tables | References | Related Articles | Metrics

When multiple agents performing path finding in large-scale warehousing environment， the existing algorithms have problems that agents are prone to fall into congestion areas and it take a long time. In response to the above problem， an improved Conflict-Based Search （CBS） algorithm was proposed. Firstly， the existing single warehousing environment modeling method was optimized. Based on the traditional grid based modeling， which is easy to solve path conflicts， a hybrid modeling method of grid-heat map was proposed， and congestion areas in the warehouse were located through a heat map， thereby addressing the issue of multiple agents prone to falling into congestion areas. Then， an improved CBS algorithm was employed to solve the Multi-Agent Path Finding （MAPF） problems in large-scale warehousing environment. Finally， a Heat Map for Explicit Estimation Conflict-Based Search （HM-EECBS） algorithm was proposed. Experimental results show that on warehouse-20-40-10-2-2 large map set， when the number of agents is 500， compared with Explicit Estimation Conflict-Based Search （EECBS） algorithm and Lazy Constraints Addition for MAPF （LaCAM） algorithm， HM-EECBS algorithm has the solution time reduced by about 88% and 73% respectively； when there is 5%，10% area congestion in warehouse， the success rate of HM-EECBS algorithm is increased by about 49% and 20% respectively， which illustrates that the proposed algorithm is suitable for solving MAPF problems in large-scale and congested warehousing and logistics environments.

Broad quantum state tomography model based on adaptive feature extraction

Wenjie YAN, Dongyue DANG

2024, 44(12):  3861-3866.  DOI: 10.11772/j.issn.1001-9081.2023121725

Asbtract ( )   HTML ( )   PDF (1573KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Aiming at the problem of exponential growth of data dimension faced by Quantum State Tomography （QST）， a Broad QST model based on Adaptive Feature Extraction （AFE_BQST） was proposed. Firstly， an adaptive feature extraction strategy was introduced to avoid the uncertainty of mapping feature nodes caused by random generation of weights. Secondly， Broad Learning System （BLS） was used to map the input data to a more appropriate feature space in a non-iterative way for feature extraction of large-capacity data. Finally， experiments were executed in the cases of low- and high-dimensional quantum state data to compare AFE_BQST with Broad QST （BQST）， Deep neural network QST （D_QST）， Convolutional neural network QST （C_QST） and U-shaped network QST （U_QST） models by using two indicators of average fidelity and running time. Experimental results show that in the case of small samples with low-dimensional quantum state， compared with the sub-optimal baseline model BQST， AFE_BQST improves the fidelity by 0.045 percentage points with the similar running time； in the case of large samples with high-dimensional quantum state， compared with the sub-optimal baseline model D_QST， AFE_BQST improves the fidelity by 0.175 percentage points with the running time reduced by 99%. The above results prove that AFE_BQST is able to extract quantum state data features adaptively and reconstruct quantum state data accurately and efficiently.

Adaptive computing optimization of sparse matrix-vector multiplication based on heterogeneous platforms

Bo LI, Jianqiang HUANG, Dongqiang HUANG, Xiaoying WANG

2024, 44(12):  3867-3875.  DOI: 10.11772/j.issn.1001-9081.2023111707

Asbtract ( )   HTML ( )   PDF (3526KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Sparse Matrix-Vector multiplication （SpMV） is an important numerical linear algebraic operation. The existing optimizations for SpMV suffer from issues such as incomplete consideration of preprocessing and communication time， lack of universality in storage structures. To address these issues， an adaptive optimization scheme for SpMV on heterogeneous platforms was proposed. In the proposed scheme， the Pearson correlation coefficients were utilized to determine highly correlated feature parameters， and two Gradient Boosting Decision Tree （GBDT） based algorithms eXtreme Gradient Boosting （XGBoost） and Light Gradient Boosting Machine （LightGBM） were employed to train prediction models to determine the optimal storage format for a certain sparse matrix. The use of grid searches to identify better model hyperparameters for model training resulted in both of those algorithms achieving more than 85% accuracy in selecting a more suitable storage structure. Furthermore， for sparse matrices with the HYBrid （HYB） storage format， the ELLPACK （ELL） and COOrdinate （COO） storage format parts in these metrices were computed on the GPU and CPU separately， establishing a CPU+GPU parallel hybrid computing mode. At the same time， hardware platforms were also selected for sparse matrices with small data sizes to improve computational speed. Experimental results demonstrate that the adaptive computing optimization achieves an average speedup of 1.4 compared to the Compressed Sparse Row （CSR） storage format in cuSPARSE library， and average speedup of 2.1 and 2.6 compared to the HYB and ELL storage formats， respectively.

Relay control model for concurrent data flow in edge computing

Ming ZHANG, Le FU, Haifeng WANG

2024, 44(12):  3876-3883.  DOI: 10.11772/j.issn.1001-9081.2023121812

Asbtract ( )   HTML ( )   PDF (2940KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In order to improve the data transmission efficiency in edge computing application scenarios and manage the concurrent data traffic effectively， a relay control model for concurrent data flow in edge computing was designed. Firstly， based on the features of Data Plane Development Kit （DPDK） such as bypassing kernel， multi-core processing， and sending and receiving packets from multiple network ports， the concurrent receiving and forwarding processing of data flow was realized. Secondly， by establishing a system model with Model Predictive Control （MPC） as the core， state prediction was used to optimize control inputs and provide timely feedback and adjustment， so as to achieve the control of data traffic. Finally， a Weighted Round-Robin （WRR） algorithm was proposed to allocate weights according to buffer size and recent usage time in order to achieve load balancing of data flow. Experimental results show that the proposed model is able to control the real-time data flow rate effectively in edge network environment， and has the control error between -1% and 2%. The proposed model improves the data flow sending bit rate of edge nodes in real application scenarios compared with traditional Linux kernel forwarding， and the transmission quality and packet delay are also improved accordingly. It can be seen that the proposed model can meet the demands for low latency and high bandwidth in edge clusters and internet of things data centers， and can optimize critical computing resources while reducing peak loads.

Network and communications

Throughput optimization algorithm of full-duplex two-way relay network with non-linear energy harvesting and residual hardware impairments

Feng QIAO, Runhe QIU

2024, 44(12):  3884-3892.  DOI: 10.11772/j.issn.1001-9081.2023121746

Asbtract ( )   HTML ( )   PDF (3047KB) ( )  

Figures and Tables | References | Related Articles | Metrics

To comply to the non-linear output characteristic and non-ideal hardware of Energy Harvesting （EH） circuits， a constant-linear-constant EH model was applied in Simultaneous Wireless Information and Power Transfer （SWIPT）-enabled full-duplex two-way relay network， and Residual Hardware Impairments （RHIs） were taken into account. To reduce the impact of RHIs and Residual Self-Interference （RSI） on the outage performance of relay network， an optimization algorithm based on golden section method was proposed for improving throughput. Firstly， considering the piecewise Signal-to-Noise-plus-Distortion Ratio （SNDR）， the end-to-end outage probability was expanded by the law of total probability， and the differentiation and variable substitution were used for classification discussion， so as to convert the complex expression of the outage probability into a single integral， thus obtaining a closed-form expression of the outage probability. Secondly， the obtained end-to-end outage probability was expressed in piecewise way by the SNDR threshold sizes， and the Overall System Ceiling （OSC） caused by RHI was verified. Thirdly， the system throughput derived from the outage probability was transformed into a function related to the Time Switching （TS） ratio. Finally， after analyzing the monotonicity of the function， the proposed algorithm was used to maximize the system throughput. Simulation results verify the piecewise outage probability expression and better system throughput brought by optimization algorithm. Besides， compared to linear EH models， the application of nonlinear EH models in full-duplex mode results in system throughput decrease instead of increase at high SNDRs.

Hybrid intelligent reflecting surface and relay assisted secure transmission scheme based on cooperative interference

Yan SHI, Yue WU, Dongqing ZHAO

2024, 44(12):  3893-3898.  DOI: 10.11772/j.issn.1001-9081.2023121761

Asbtract ( )   HTML ( )   PDF (2050KB) ( )  

Figures and Tables | References | Related Articles | Metrics

To solve the problems of large channel fading damage， low resource utilization and security loss in high-spectrum short packet communications， a hybrid Intelligent Reflecting Surface （IRS） and relay assisted secure transmission scheme based on cooperative interference was proposed， which used artificial noise to interfere with the channel quality of eavesdroppers in the Multi-Input Single-Output （MISO） system to improve physical layer security. Firstly， the closed-form solution of the base station beamforming vector at the transmitter was derived to optimize the multi-antenna beamforming problem of the base station. Then， the Successive Convex Approximation （SCA） method was used to obtain the optimal allocation ratio of noise power， the gradient descent based Riemannian Manifold （RM） optimization method was used to obtain the optimal phase shift matrix， and the local optimal solutions of three sub-problems were solved respectively. Finally， an alternating optimization algorithm was used to iteratively obtain the global optimal solution. Simulation results show that the proposed algorithm has good convergence performance. When the number of IRS elements is 128， the secrecy rate of the proposed scheme is double of that of the IRS-only scheme and about triple as much as that of the relay-only scheme. In addition， when the location of IRS in the system is not fixed， the optimal power allocation scheme of the hybrid network has higher security performance.

Multimedia computing and computer simulation

Speaker verification method based on speech quality adaptation and triplet-like idea

Chao WANG, Shanshan YAO

2024, 44(12):  3899-3906.  DOI: 10.11772/j.issn.1001-9081.2023121857

Asbtract ( )   HTML ( )   PDF (1102KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Aiming at the problem that current Speaker Verification （SV） methods suffer from serious performance degradation in complex test scenarios or when the speech quality degradation is large， a speaker verification Method based on speech Quality Adaptation and Triplet-like idea （QATM） was proposed. Firstly， the feature norms of the speaker's voice were utilized to correlate the speech quality， Then， through judging the quality of the speech samples， the importance of speech samples of different qualities was adjusted by different loss functions， so as to pay attention to the hard samples with high speech quality and ignore the hard samples with low speech quality. Finally， the triplet-like idea was utilized to simultaneously improve AM-Softmax （Additive Margin Softmax） loss and AAM-Softmax （Additive Angular Margin Softmax） loss， aiming to pay more attention to hard speaker samples to cope with the damage of hard samples with too poor speech quality to the model. Experimental results show that when the training set is VoxCeleb2 development set， the proposed method reduces the Equal Error Rate （EER） compared to the AAM-Softmax loss-based method on VoxCeleb1-O test set in network architecture Half-ResNet34， ResNet34， and ECAPA-TDNN （Emphasized Channel Attention， Propagation and Aggregation in Time Delay Neural Network） by 6.41%， 3.89%， and 7.27%， respectively. When the training set is Cn-Celeb.Train， the proposed method reduces the EER by 5.25% on evaluation set Cn-Celeb.Eval compared to the AAM-Softmax loss-based method in network architecture Half-ResNet34. It can be seen that the accuracy of the proposed method is improved in both ordinary and complex scenarios.

Self-supervised monocular depth estimation using multi-frame sequence images

Wei XIONG, Yibo CHEN, Lizhen ZHANG, Qian YANG, Qin ZOU

2024, 44(12):  3907-3914.  DOI: 10.11772/j.issn.1001-9081.2023111713

Asbtract ( )   HTML ( )   PDF (2652KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Multi-frame self-supervised monocular depth estimation constructs a Cost Volume （CV） based on the relationship between current frame and the previous frame， serving as an additional input source for the monocular depth estimation network. This approach provides a more accurate description of the temporal and spatial structure of scene videos. However， the cost volume becomes unreliable in the presence of dynamic objects or untextured regions in the scene. Overreliance on the unreliable information within the cost volume leads to a decrease in depth estimation accuracy. To tackle the issue of unreliable information in the cost volume， a multi-frame fusion module was designed to reduce the weights of unreliable information sources dynamically and mitigate the impact of unreliable information sources on the network. Besides， to handle the negative impact of unreliable information sources in cost volume on network training， a network was designed to guide the training of the depth estimation network， preventing the depth estimation network from overly depending on unreliable information. The proposed method achieves excellent performance on KITTI dataset， with absolute relative error， squared relative error， and Root Mean Square Error （RMSE） decreased by 0.015， 0.094， and 0.200， respectively， compared to the benchmark method Lite-Mono. In comparison to similar methods， the proposed method not only has higher precision， but also requires fewer computational resources. The proposed network structure makes full use of the advantages of multi-frame training， while avoiding the defects of multi-frame training （i.e.， the influence of cost volume uncertainty on the network）， and improves the model precision effectively.

Parallel medical image registration model based on convolutional neural network and Transformer

Xin ZHAO, Xinjie LI, Jian XU, Buyun LIU, Xiang BI

2024, 44(12):  3915-3921.  DOI: 10.11772/j.issn.1001-9081.2023121828

Asbtract ( )   HTML ( )   PDF (2554KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Medical image registration models aim to establish the correspondence of anatomical positions between images. The traditional image registration method obtains the deformation field through continuous iteration， which is time-consuming and has low accuracy. The deep neural networks not only achieve end-to-end generation of deformation fields， thereby speeding up the generation of deformation fields， but also further improve the accuracy of image registration. However， all of the current deep learning registration models use single Convolutional Neural Network （CNN） or Transformer architecture， and have the problems such as the inability to fully utilize the advantages of the combination of CNN and Transformer， resulting in insufficient registration accuracy， and the inability to maintain the original topology effectively after image registration. To solve these problems， a parallel medical image registration model based on CNN and Transformer — PPCTNet （Parallel Processing of CNN and Transformer Network） was proposed. Firstly， the model was constructed using Swin Transformer， which currently has the excellent registration accuracy， and LOCV-Net （Lightweight attentiOn-based ConVolutional Network）， a very lightweight CNN. Then， the feature information extracted by Swin Transformer and LOCV-Net were fully integrated by designing a fusion strategy， so that the model not only had the local feature extraction capability of CNN and the long-distance dependency capability of Transformer， but also had the advantage of being lightweight. Finally， based on the brain Magnetic Resonance Imaging （MRI） dataset， PPCTNet was compared with 10 classical image alignment models. The results show that compared to the currently excellent registration model TransMorph （hybrid Transformer-ConvNet network for image registration）， PPCTNet has the highest registration accuracy 0.5 percentage points higher， and the folding rate of deformation field 1.56 percentage points reduced， maintaining the topological structures of the registered images. Besides， compared with TransMorph， PPCTNet has the parameters reduced by 10.39×10⁶， and the computational cost reduced by 278×10⁹， which reflects the lightweight advantage of PPCTNet.

Small target detection model in overlooking scenes on tower cranes based on improved real-time detection Transformer

Yudong PANG, Zhixing LI, Weijie LIU, Tianhao LI, Ningning WANG

2024, 44(12):  3922-3929.  DOI: 10.11772/j.issn.1001-9081.2023121796

Asbtract ( )   HTML ( )   PDF (3128KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In view of a series of problems of security guarantee of construction site personnel such as casualties led by falling objects and tower crane collapse caused by mutual collision of tower hooks， a small target detection model in overlooking scenes on tower cranes based on improved Real-Time DEtection TRansformer （RT-DETR） was proposed. Firstly， the multiple training and single inference structures designed by applying the idea of model reparameterization were added to the original model to improve the detection speed. Secondly， the convolution module in FasterNet Block was redesigned to replace BasicBlock in the original BackBone to improve performance of the detection model. Thirdly， the new loss function Inner-SIoU （Inner-Structured Intersection over Union） was utilized to further improve precision and convergence speed of the model. Finally， the ablation and comparison experiments were conducted to verify the model performance. The results show that， in detection of the small target images in overlooking scenes on tower cranes， the proposed model achieves the precision of 94.7%， which is higher than that of the original RT-DETR model by 6.1 percentage points. At the same time， the Frames Per Second （FPS） of the proposed model reaches 59.7， and the detection speed is improved by 21% compared with the original model. The Average Precision （AP） of the proposed model on the public dataset COCO 2017 is 2.4， 1.5， and 1.3 percentage points higher than those of YOLOv5， YOLOv7， and YOLOv8， respectively. It can be seen that the proposed model meets the precision and speed requirements for small target detection in overlooking scenes on tower cranes.

Frontier and comprehensive applications

Survey of application of deep learning in meteorological data correction

Hongru JIANG, Wei FANG

2024, 44(12):  3930-3940.  DOI: 10.11772/j.issn.1001-9081.2023121756

Asbtract ( )   HTML ( )   PDF (1587KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Data correction is one of the core processes in data assimilation， which aims to improve the assimilation effect of data by correcting and calibrating the data. Aiming at the issue of multiple errors in meteorological observations leading to biases in meteorological data， the application of deep learning in meteorological data correction was reviewed， and the application scenarios include meteorological model correction， weather forecast， and climate prediction. Firstly， the importance of meteorological data correction was introduced， and traditional meteorological data correction methods such as statistics and traditional machine learning were looked back with advantages and limitations of the methods analyzed. Secondly， the application of deep learning in data correction in three scenarios was detailed， the deep learning methods include Convolutional Neural Network （CNN）， Recurrent Neural Network （RNN）， and Transformer. At the same time， by summarizing the current research progress， the strengths and weaknesses of deep learning methods and traditional methods in data correction were discussed. Finally， the limitations of deep learning in data correction were summed up， and the optimization approaches and future development directions of deep learning in meteorological data correction were pointed out.

Dynamic monitoring method of flight chain operation status

Jianli DING, Hui HUANG, Weidong CAO

2024, 44(12):  3941-3948.  DOI: 10.11772/j.issn.1001-9081.2023121758

Asbtract ( )   HTML ( )   PDF (2366KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In order to grasp the overall status of flight operation more accurately， a dynamic monitoring method of flight chain operation status was proposed. Firstly， from the perspective of the whole flight chain， a flight chain data processing method was designed on the basis of the flight chain operation business process and data characteristics， and the operation status characteristics of relevant flights and airports in all life cycle of the flight chain were integrated. Then， a dynamic monitoring function model of flight chain operation status including flight chain delay prediction module， error compensation module based on historical data， and flight chain status monitoring module was constructed. Finally， a dynamic update strategy of the model was designed on the basis of incremental learning in order to improve the model robustness. Through simulation experiments in laboratory environment， it can be seen that the proposed method achieves excellent results in terms of computational efficiency and accuracy， and the accuracy reaches 92.07%. Therefore， the proposed method can monitor the operation status of flight chain effectively， help to achieve precise control of the flight operation situation， and improve operational control efficiency.

fMRI brain age prediction model with lightweight multi-scale convolutional network

Yanran SHEN, Xin WEN, Jinhao ZHANG, Shuai ZHANG, Rui CAO, Baolu GAO

2024, 44(12):  3949-3957.  DOI: 10.11772/j.issn.1001-9081.2023121764

Asbtract ( )   HTML ( )   PDF (2350KB) ( )  

Figures and Tables | References | Related Articles | Metrics

In view of the low accuracy of functional Magnetic Resonance Imaging （fMRI） brain age prediction and the lack of research on the combination of this problem and deep learning， an fMRI brain age prediction model with Lightweight Multi-scale Convolutional Network （LMCN） was proposed. Firstly， the Pearson correlation coefficient （R） of the Region Of Interest （ROI） in fMRI was calculated to obtain the Functional Connectivity （FC） matrix of the ROI as input. Secondly， the number of FC channels was increased to ensure the number of features and the size of the feature map was reduced simultaneously. At the same time， the multi-scale dilated convolution module RFB （Receptive Field Block） with the characteristics of human visual attention was used to extract age features. Finally， the predicted brain age was output by the fully connected layer， and the ablation prediction results of each brain region were calculated to explore the key brain regions influencing the brain age prediction results. Evaluation was carried out on two public datasets， E-NKI and Cam-CAN. It can be seen that the memory required for LMCN parameters is 2.30 MB， which is 60.3% and 52.0% less than those of MobileNetV3 and ShuffleNetV2 respectively. In terms of prediction results， on E-NKI dataset， LMCN has the Mean Absolute Error （MAE） of 5.16， the R of 0.947， and the Root Mean Square Error （RMSE） of 6.40. Compared to the model that combines network-based feature selection with the least angle regression， LMCN has the MAE decreased by 1.34 and the R increased by 0.037； on Cam-CAN dataset， LMCN has the MAE of 5.97， the R of 0.904， and the RMSE of 7.93. Compared to the connectome-based machine learning model， LMCN has the R increased by 0.019， and the RMSE decreased by 0.64. The results show that while LMCN has small number of parameters and is easy to deploy， it can improve the accuracy of fMRI brain age prediction effectively and provide clues for assessing the brain status of healthy adults.

Theoretical tandem mass spectrometry prediction method for peptide sequences based on Transformer and gated recurrent unit

Changjiu HE, Jinghan YANG, Piyu ZHOU, Xinye BIAN, Mingming LYU, Di DONG, Yan FU, Haipeng WANG

2024, 44(12):  3958-3964.  DOI: 10.11772/j.issn.1001-9081.2023121846

Asbtract ( )   HTML ( )   PDF (1699KB) ( )  

Figures and Tables | References | Related Articles | Metrics

Aiming at the issues in the existing prediction methods， such as only predicting b and y backbone fragment ions， as well as single model's difficulty in capturing the complex relationships within peptide sequences， a theoretical tandem mass spectrometry prediction method for peptide sequences based on Transformer and Gated Recurrent Unit （GRU）， named DeepCollider， was proposed. Firstly， through self-attention mechanism and long-distance dependencies， the deep learning architecture combining Transformer and GRU was used to enhance the modeling ability of relationship between peptide sequences and fragment ion intensities. Secondly， unlike the existing methods encoding peptide sequences to predict all b and y backbone ions， fragmentation flags were utilized to mark fragmentation sites within peptide sequences， thereby enabling the encoding of fragment ions at specific fragmentation sites and prediction of the corresponding fragment ions. Finally， Pearson Correlation Coefficient （PCC） and Mean Absolute Error （MAE） were employed as evaluation metrics to measure the similarity between predicted spectrometry and experimental spectrometry. Experimental results demonstrate that DeepCollider shows advantages in both PCC and MAE metrics compared to the existing methods limited to predicting b and y backbone fragment ions， such as pDeep and Prosit methods， with an increase of 0.15 in PCC value and a decrease of 0.005 in MAE value. It can be seen that DeepCollider not only predicts b， y backbone ions and their corresponding dehydrated and deaminated neutral loss ions， but also further improves the peak coverage and similarity of theoretical spectrometry prediction.