Journal of Shaanxi Normal University(Natural Science Edition)

Latest Issues

  • SABM:an enhanced SAM model for segmenting any butterflies in the ecological images

    XIE Juanying;LAN Xiang;XU Shengquan;School of Artificial Intelligence and Computer Science,Shaanxi Normal University;School of Life Sciences,Shaanxi Normal University;

    Segmenting the butterflies from ecological images will provide accurate butterfly masks,guaranteeing the accuracy of the automatic butterfly species identification using the ecological images of butterflies.Therefore,the segmentation study of butterfly ecological images is of great significance.However,existing butterfly ecological image dataset cannot train an excellent butterfly segmentation model with strong generalization due to the small number of samples in the dataset and the mimicries and wing folds of butterflies in the butterfly ecological images.To address these issues,a new enhanced SAM(segment anything model)with good and robust segmentation capability is proposed.This enhanced SAM is named as SABM(segment any butterfly model)for segmenting the butterfly ecological images.This SABM introduces two-way convolution module,butterfly token,and a 3-layer MLP(multi-layer perceptron)to enhance SAM to adapt to the ecological butterfly image segmentation task.The 2-fold cross validation experimental results on the available butterfly ecological image dataset containing 707 ecological butterfly images demonstrate that this proposed SABM obtains an excellent segmentation performance for the ecological butterfly images.It is superior to SAM and its variants,particularly the SOTA model of SAM variants.Additionally,the segmentation experiments on the entirely new 7 645 butterfly ecological images show that this SABM has strong generalization capability,and it can segment all these 7 645 ecological butterfly images efficiently.This segmentation results provide a 10 times larger dataset than the available one for future butterfly segmentation task utilizing the ecological images while providing a much better dataset for the automatic butterfly species identification task through ecological images of butterflies,and a very challenging dataset for testing the performance of a clustering algorithm.Furthermore,the robust of the proposed SABM is tested on medical image datasets.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1622K]

  • Brain network detection and its application based on deep learning and functional magnetic resonance imaging

    SUN Chenjing;FENG Ruping;YANG Yuanyuan;MA Siyuan;CHANG Yuxin;GAO Jie;GE Bao;ZHAO Shijie;LI Jin;QIANG Ning;School of Physics &Information Technology,Shaanxi Normal University;School of Mathematics and Information Technology,Yuncheng University;School of Automation,Northwestern Polytechnical University;

    Detection of brain networks based on functional magnetic resonance imaging data is crucial for understanding cognitive and functional aspects of the brain,as well as exploring brain disorders.With the development of deep learning techniques,an increasing number of researchers have applied them in the field of brain network detection.The main research achievements and advancements in this field are summarized.Firstly,it introduces the basic principles of brain network detection based on fMRI.Then,it discusses the deep learning models and their applications in brain network detection,analyzing their strengths and limitations.Finally,it summarizes the challenges and future research directions in applying deep learning methods to brain network detection.An important reference for further promoting research and applications in brain network detection using deep learning and fMRI imaging is provided.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1516K]

  • Integrating the improved A-star algorithm and the enhanced DWA algorithm for the path planning

    NI Jianyun;ZHANG Fengjie;SHANG Hongzhi;GU Haiqing;CAO Wenjun;School of Electrical Engineering and Automation,Tianjin University of Technology;

    Aiming at the performance requirements of mobile robot path planning for shortest path length and smoothness,an improved A-star algorithm is proposed.This approach is combined with an enhanced dynamic window approach(DWA)algorithm to solve the problem of dynamic path planning for mobile robots.A new heuristic function of A-star algorithm is designed,along with dynamic weight allocation and the introduction of a new evaluation function,which realizes the shortest global planning path and reduces the inflection point and redundant nodes of the planning path.The distance function of static obstacle and dynamic obstacle is designed for DWA algorithm,along with the addition of a path deviation distance function.By incorporating guidance from the global path,the DWA algorithm generates a trajectory that closely adheres to the global path,enabling timely avoidance of unknown and dynamic obstacles while ensuring alignment with the globally optimal path.The simulation results show that the improved A-star algorithm reduces the path length by 34.4%and the inflection point by 53.5%compared with the traditional algorithm in complex environment.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1650K]

  • A sequential recommendation model for generating positive sample pairs based on noise perturbation

    CAO Fasheng;CAO Wei;SU Yanqing;Department of Cognitive Science and Technology/School of Data Science and Information Engineering,Guizhou Minzu University;

    Although the results of the SASRec(self-attention sequential recommendation)model on both sparse and dense datasets are superior to various sequence recommendation models,in sequence recommendation,it also suffers from representation degradation,that is,frequently occurring Items are often concentrated in a small region of the representation space,degrading recommendation performance.In order to solve this problem,a comparative learning loss function is introduced.Gaussian noise was added in the embedding space for data enhancement,and the original item sequence and the item sequence after data enhancement were used to construct positive sample pairs to promote similar instances in the mapping the closer the distance in the space,and the distribution of different instances in the mapping space showed uniformity.As far as possible,the instance can retain its own personalized information after being mapped to embedding.A comprehensive experimental study on two benchmark datasets shows that,although it appears to be very simple,the proposed method can smoothly adjust the popularity bias of the learned representations.The contrastive learning is based on the graph recommendation model SGL(self-supervised graph learning for recommendation),which suffers from negative sampling bias in representation learning.The model proposed in this paper can effectively improve the recommendation performance.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1467K]

  • Rock art image classification of Helan Mountain based on dual-branch attention fusion network

    FAN Ye;WANG Yang;Library,Northwest University;Institute for Advanced Study in History of Science,Northwest University;Northwest University Shaanxi Provincial Key Laboratory of Digital Humanities for Cultural Heritage;

    As a distinctive historical and cultural heritage,rock art faces significant challenges in image classification due to its vast quantity,wide distribution and complex overlapping elements,which hinder efficient and accurate identification.Recent advances in deep learning offer new opportunities for rock art conservation research.A dual-branch attention fusion network(DBAFN)that integrates the local feature extraction capability of ResNet50 with the global semantic modeling strength of ViT(vision Transformer)is proposed.Using Helan Mountain rock art as a case study,our method employs a gated attention mechanism to dynamically weight features,significantly enhancing classification accuracy.Evaluated on a dataset containing 1 200 Helan Mountain rock art images(human faces,animals,hunting scenes),DBAFN achieves a weighted average classification accuracy of 85.62%,outperforming standalone ResNet50(81.46%)and ViT(80.02%)models.Notably,the F_1-score for hunting scenes reaches 82.35%.Experiments demonstrate that DBAFN effectively resolves misclassification caused by interleaved elements,providing an innovative technical pathway for semantic analysis of cultural heritage and interdisciplinary research while advancing the application of artificial intelligence in digital conservation of cultural relics.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1650K]

  • Automatic classification of mural painting styles using 3D hyperspectral imaging and Transformer networks

    RONG Yan;YANG Jinglong;ZHANG Pengchang;ZENG Zimu;School of History and Culture,Xianyang Normal University;Xianyang Cultural Heritage Conservation Center;Spectral Imaging Technology Research Office of Xi'an Institute of Optics and Fine Mechanics,Chinese Academy of Sciences;

    Ancient murals carry profound historical,cultural,and artistic value.A network based on 3 Dhyperspectral Transformer for the classification of style features of Chinese Central Plains style and Western Regions style murals is proposed.By collecting hyperspectral images of murals and constructing corresponding datasets.On this basis,the proposed network is trained using transfer learning methods.The experimental results show that compared with other traditional methods,the accuracy of this study has increased by 0.92%,the precision has increased by0.12%,the recall has increased by 1.4%,and the F_1 score has increased by 0.75%compared to the optimal algorithm.In addition,compared with the deep learning classification method based on color images,the accuracy improved by 4.5%,the precision improved by 1.9%,the recall improved by 5.4%,and the F_1 score improved by 3.7% compared to the method with the best results.The study verified the unique advantages of hyperspectral information in improving classification accuracy.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1570K]

  • Triadic concept reduction based on representative triadic concept matrix

    YAN Li;ZHAO Siyu;WEI Ling;School of Mathematics,Northwest University;School of Mathematics and Information Sciences,Baoji University of Arts and Sciences;Institute of Concepts,Cognition and Intelligence,Northwest University;

    Triadic concept analysis is a theoretical approach to knowledge discovery of a triadic context by constructing triadic concepts.A triadic concept reduct is a minimal triadic concept set that preserves triadic relation of a triadic context unchanged.The problems of triadic concept reduction based on the representative triadic concept matrix are studied. Firstly,the representative triadic concept matrix is defined.On the basis of the matrix,judging theorems of a triadic concept consistent set and a triadic concept reduct are given.Secondly,the minimum representative triadic concept matrix is defined and the algorithm to obtain triadic concept reducts through the matrix is given.Finally,the characteristics of core,relatively necessary and absolutely unnecessary triadic concepts are given respectively from the perspective of the minimum representative triadic concept matrix.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1460K]

  • Uncertainty measure method of intuitionistic fuzzy information systems and its attribute reduction

    YANG Yixuan;YU Jianhang;SU Zuqiang;YU Hong;LIU Mingyang;Chongqing Key Laboratory of Computational Intelligence,Chongqing University of Posts and Telecommunications;

    With the rapid development and widespread application of information technology,the volume of data has experienced explosive growth,accompanied by an increase in data types and uncertainty.Traditional precise mathematical models are inadequate for accurate description in the face of such complexity.However,intuitionistic fuzzy information systems offer an effective means of representing fuzzy uncertainties,facilitating knowledge discovery from complex data.The evolution of data forms has imposed significant limitations on traditional granulation mechanisms based on equivalence or similarity.Equivalence relations are overly strict,while similarity relation may violate the monotonicity of granularity.To address these challenges,a granulation method based on the dominance relation for intuitionistic fuzzy information systems is proposed,then defines the information entropy and rough entropy to measure uncertainty in the system.Furthermore,to reduce the dimensionality of the data and eliminate redundant attributes,thus enhancing the efficiency of data processing and classification accuracy,a measure of relative and absolute significance of attributes is constructed based on rough entropy and then a heuristic attribute reduction algorithm for intuitionistic fuzzy information systems is designed.Finally,the effectiveness of the algorithm is verified and analyzed with the UCI datasets,which can effectively improve the efficiency of obtaining reductions compared to traditional algorithms.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1442K]

  • Matrix-based dynamic updating of multi-granulation rough fuzzy sets

    ZHANG Zeting;LI Leijun;MI Jusheng;LI Meizheng;School of Mathematical Sciences,Hebei Normal University;Hebei Key Laboratory of Computational Mathematics and Applications;School of Computer and Cyberspace Security,Hebei Normal University;

    The grain structure in information system often develops dynamically with the passage of time,and dynamic acquisition of potentially useful knowledge is of great significance to decision-making.Incremental learning is an effective way to combine current information with previous knowledge to study such problems.Although incremental learning has been successful in many aspects,less research has been conducted in multi-granularity fuzzy environments.In order to solve this problem,a matrix operator is used to represent the upper and lower approximations of optimistic and pessimistic multi-granularity rough fuzzy sets.Then,when the attributes increase or decrease,the update mechanism based on the matrix is introduced,and the corresponding dynamic update algorithm is designed.Finally,the time complexity of the algorithm is analyzed,and the validity of the proposed dynamic update algorithm is verified in the UCI data set.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1748K]

  • Delineate knowledge structure based on variable precision competence model

    YANG Jingjing;HUANG Chuanyi;LI Jinjin;School of Mathematics and Computer Science,Quanzhou Normal University;Key Laboratory of New Technologies and Knowledge Engineering for Big Data Management in Fujian Province(Quanzhou Normal University);School of Mathematics and Statistics,Minnan Normal University;

    Competence-based knowledge space theory is a theoretical framework used to delineate an individual's level of competence in a particular subject area.The competency model is a key framework for structuring knowledge within this theoretical framework.In the competence model,it is assumed that there may be more than one way of solving aproblem,and that the individual's ability to solve the problem requires full mastery of one of the ways of solving the problem.However,it is possible for an individual to solve a problem by acquiring some of the skills of a particular method for solving the problem.Thus,it can be seen that the conditions of the competence model are too strict,and there is an urgent need to expand the competence model in practical applications.Therefore,three variable-precision competence models for constructing knowledge structures based on the competence inclusion degree are proposed;investigates the nature of constructing knowledge structures,and finds that the three variable-precision competence models have the same effect as the original variable-precisionα-competence model in constructing families of knowledge structures.Furthermore,the maximum competency inclusion matrix of the skill function is defined,employs the matrix method to derive the knowledge structure constructed by variable precision competency models,and develops the corresponding algorithms.The issue of simplifying skill subsets and selecting paths,which preserve the knowledge structure,is also addressed.Finally,six datasets were selected for experimental verification of the proposed matrix method's validity and superiority in constructing the structure.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1718K]

  • On character sums with multiplicative coefficients over flat numbers

    SHI Hui;XU Zhefeng;School of Mathematics,Northwest University;

    Let q≥3 be a sufficiently large prime,x be a real number such that■，χ be any non-principal character mod q,f(n)be a multiplicative function satisfying|f(n）|≤1.It is obtained that several new bounds for sums of the form■where ■ is defined by ■~ and (a)_q means that the minimal positive residue of that amod q.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1430K]

  • On the Kloosterman sums and its hybrid power mean

    MENG Yuanyuan;CHEN Li;School of Mathematics,Xi'an University of Finance and Economics;

    The properties of the trigonometric sums,character sums and the classical Gauss sums are used to study the computational problem of two kinds of hybrid power mean involving the Kloosterman sums and character sums.Some exact computational formulae for them are given.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1387K]

  • On the hybrid mean value of the Dedekind sums and one kind special Kloosterman sums

    ZHANG Jin;LIU Xiaoge;School of Information Engineering,Xi'an University;Research Center for Number Theory and Its Applications,Northwest University;

    Using the elementary methods and the properties of Kloosterman sums,Dedekind sums and Gauss sums to study the calculating problem of the hybrid mean value involving the Dedekind sums and one kind special Kloosterman sums,and two exact calculating formulae and a sharp asymptotic formula for the hybrid power mean are given.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1444K]

  • Generalized k-FL quaternion and its related algebraic properties

    DENG Yong;School of Mathematics and Statistics,Kashi University;Research Center of Modern Mathematics and Its Application,Kashi University;

    In order to further reveal the core properties of quaternion sequences under the theoretical framework of 2-parameter generalized quaternion algebra, 2-parameter k-FL generalized quaternion is introduced based on k-Fibonacci generalized quaternion and k-Lucas generalized quaternion.Firstly,the definitions and basic properties of several famous sequences are briefly reviewed,as well as their generalizations on 2-parameter generalized quaternion algebras.Secondly,the k-FL generalized quaternion is defined by using the linear combination of k-Fibonacci and k-Lucas generalized sequences which on the 2-parameter generalized quaternion algebra,and some of it's famous identities connected with k-Fibonacci and k-Lucas sequence are studied.Finally,the new perspectives and application prospects of further research on k-FL generalized quaternions are summarized.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1518K]

  • Several mean value formulae of Dirichlet L-functions

    ZHANG Miao;HU Jiayuan;School of Digital and Intellegent Industry,Inner Mongolia University of Science and Technology;School of Science,Inner Mongolia University of Science and Technology;

    Use the relations between the Dirichlet L-functions and the Dedekind sums to study the calculating problems of some special mean square values of Dirichlet L-functions.Several new and significative mean value formulae for them are given.

    2025 06 v.53;No.222 [Abstract][OnlineView][Download 1484K]