Loading...
Welcome to Journal of University of Chinese Academy of Sciences,Today is

Current Issue

2022, Vol.39, No.3 Previous Issue    Next Issue
Research Articles
Robust ordinal mislabel logistic regression based on γ-divergence
GUO Meijun, REN Mingyang, LI Shiming, ZHANG Sanguo
2022, 39 (3): 289-301.  DOI: 10.7523/j.ucas.2020.0056
Abstract ( 563 ) PDF (0KB) ( 16 )
Ordinal multi-classification methods have been studied widely. Traditional ordinal multi-classification methods assume that the sample label is not mislabeled. Due to the complexity of the real data and the limited artificial experience, it is unrealistic to obtain completely accurate labels, in which conventional methods perform poorly. In this article, we propose an ordinal mislabel logistic regression method based on γ-divergence, which possessing strong robustness when dealing with ordinal mislabeled response data. That is to say, when mislabeled, the weight of the sample in parameter estimation equation diminish compared to the case that the sample is properly labeled. Our method not only possesses the robustness but also can ignore the mislabel probabilities in the model. We construct the model by minimizing γ-divergence estimation and solve the model by gradient descent algorithm. Both simulation studies and real data analysis demonstrate that the method, namely robust ordinal mislabel logistic regression, is efficient to analyze ordinal mislabeled response data.
Research of confidence intervals for precision matrix in high dimensional network data
ZHENG Zemin, ZHOU Huiting
2022, 39 (3): 302-308.  DOI: 10.7523/j.ucas.2020.0038
Abstract ( 317 ) PDF (0KB) ( 0 )
With the development of the Internet, science, and technology, the surge of bigdata on an unprecedented scale has brought complex network data between different individuals. It is practical significance to uncover the network connection by studying the confidence intervals of the precision (inverse covariance) matrix in graphical models. One natural and important question is how to efficiently obtain confidence intervals of the precision matrix. This paper proposes the De-ISEE (De-innovated scalable efficient estimation) statistic, whose confidence intervals enjoy efficient computation while maintaining a desirable coverage rate. Both average coverage and computational advantages of the methods have been demonstrated by our numerical studies in network data. Moreover, this paper applies the De-ISEE method to riboflavin data and gene expression data, and finds that De-ISEE method could be an important tool for studying gene association.
Reshape of the drainage landform by exogenous inflow: river incision and drainage capture
LI Yifei, SHI Yaolin, ZHANG Huai
2022, 39 (3): 309-320.  DOI: 10.7523/j.ucas.2021.0026
Abstract ( 383 ) PDF (0KB) ( 0 )
Since the late Neogene and early Quaternary, many endorheic basins have been transformed into exorheic basins. The transformation of the endorheic lake brings a large amount of water to its downstream drainage. The inflow of exogenous water reshapes the drainage landscape and influences the evolution of rivers within the basin. Based on the finite volume method, we present a numerical experimental analysis of the coupling effect between the exogenous inflow and the uplifting landscape evolution. The result indicates that the exogenous inflow will deeply incise the main downstream channel. The tributary incision is enhanced with the erosion basis's elevation decreasing, and so the elevation of the overall drainage gradually decreases. When the exogenous inflow discharge is large enough, the downstream drainage captures its adjacent drainage. The incision coefficient controls the time that the drainage capture begins. Two factors affect the drainage capture velocity: the incision coefficient and the exogenous inflow discharge. The average elevation of the drainage depends on the uplift rate, the incision coefficient, and the exogenous inflow discharge. The numerical results provide useful insights into the evolution of the Jinshan Yellow River and Fen River drainage.
A study on the plate driving forces based on torque balance
YU Xiao, WANG Shimin
2022, 39 (3): 321-331.  DOI: 10.7523/j.ucas.2020.0030
Abstract ( 330 ) PDF (0KB) ( 0 )
Plate dynamics is an important topic in geodynamics research. In this paper, the driving forces for current lithospheric plates are studied based on the principle of torque balance for stable moving rigid bodies. A method for identifying plate driving forces by calculating the angle between the torque and the plate angular momentum is proposed, and the stability of rigid plates is quantitatively analyzed under the constraint of the viscous resistant shear forces acting on the bottom of the plates. Furthermore, the moments of inertia as well as the angular momentum are computed for each plate in terms of the latest lithospheric structure model and absolute plate motion model, and the boundary and basal torques are obtained for the plates based on simple physical models. The results show that the residual torque that balances the ridge and the basal torques may be explained by those originated from the subduction slabs, collision zones, and rift zones, and that the ridge and slab forces are the main driving forces for plate motions, while the inter-plate collision are important driving forces for the Eurasian plate.
Ultra-short-term prediction and analysis of wind speed and direction of freestyle skiing aerial skill track
DENG Ziwei, SHAO Yun, WANG Guojun, HUANG Fuxiang, YANG Jiaqi
2022, 39 (3): 332-342.  DOI: 10.7523/j.ucas.2020.0043
Abstract ( 312 ) PDF (0KB) ( 0 )
Freestyle skiing aerial skills are the dominant snow sports in China, and the wind has a particularly significant impact on this sport. This article aims to realize ultra-short-term prediction of wind speed and direction on the track, provide practical and effective forecast information for this sport, and provide auxiliary support for athlete stability control and technical training. In view of the non-stationary and violent fluctuations of the track wind, the discrete wavelet transform is used to extract the characteristic components of the wind speed and direction sequence, the NAR neural network model is established for the low-frequency approximate component, and the ARIMA model is established for the high-frequency detail component, and then the results of each component are combined the final prediction result. The error analysis shows that the combined model can effectively improve the prediction lag of the single model, improve the prediction accuracy and have the ability to predict sudden changes in wind speed and direction. The prediction results is further analyzed, and converted into indicators that characterize track wind stability to provide more intuitive forecast information. Finally, the analysis of model calculation time shows that this method can meet the needs of practical applications.
Land surface deformation analysis based on time series InSAR technologies in the Gaizi valley section of China-Pakistan Corridor
WANG Yuqing, TANG Lingli, WANG Xinhong, ZHOU Zengguang, LI Ziyang, LI Chuanrong
2022, 39 (3): 343-351.  DOI: 10.7523/j.ucas.2020.0046
Abstract ( 409 ) PDF (0KB) ( 0 )
Geological disasters occur frequently along the China-Pakistan Corridor, and the monitoring of small surface deformations can help discriminate and analyze the geological disasters in the study area, which is of great significance to the construction of the corridor and the protection of basic engineering facilities along the corridor. Based on a total of 30 Sentinel-1A radar images in the interference wide-range mode acquired from April 5, 2015 to December 9, 2018, the time-series surface deformation extraction of the Gaizi valley section of China-Pakistan Corridor was carried out using PS-InSAR and SBAS-InSAR technologies. The results show that the deformation information obtained by the two methods is rather consistent in the spatial variation trend of deformation rate, and the deformation rate map obtained by the SBAS-InSAR technology is relatively more continuous in spatial distribution. According to the SBAS-InSAR deformation results, during the analysis period, obvious surface deformation occurred in many places in the study area, and the relatively large line-of-sight deformation mainly distributed on the slopes on both sides of the China-Pakistan Highway and on the glacial front area northern to the Bulunkou. Using the extracted time series deformation information and with the help of optical images, the deformation characteristics of typical unstable slopes and glacial movements were analyzed. On unstable slope, the cumulative deformation of PS points on the upper part of the slope can reach -30mm during the monitoring period. In terms of glacier movement, many PS points with large positive and negative deformation values are concentrated on the front edge of the glacier, and the cumulative positive and negative deformations can reach +60 mm and -80 mm. The paper can accumulate preliminary experience for the detection of geological hazard related deformation in the alpine mountain area and benefit for its future applications to the China-Pakistan Corridor.
Point cloud registration method with layer-by-layer filtering of matching points
ZHANG Tao, XIAO Jun, WANG Ying
2022, 39 (3): 352-359.  DOI: 10.7523/j.ucas.2020.0020
Abstract ( 252 ) PDF (0KB) ( 0 )
Point cloud registration of rock mass is the basis of rock mass engineering. Although the classic point cloud registration method can be well applied to ordinary point clouds, it can not be used for rock point cloud registration. Due to the complex surface structure of the point cloud of the rock mass, most of the area is flat, based on the characteristics of point cloud of rock mass, this paper proposes a rock mass point cloud registration algorithm that filters matching points layer by layer through geometric features, by introducing the eigenvalues and eigenvector matrices of the covariance matrix of matching point pairs, and geometric features such as curvature and principal direction, the matching point pairs can be accurately found. The experimental test and analysis results on different rock mass point clouds show that the algorithm in this paper has obvious advantages in accuracy.
Name disambiguation based on encoding attributes and graph topology
MA Yingying, WU Youlong, TANG Hua
2022, 39 (3): 360-368.  DOI: 10.7523/j.ucas.2020.0019
Abstract ( 274 ) PDF (0KB) ( 0 )
Aiming at solving the problem of author name ambiguity, we propose a novel name disambiguation method based on encoding attributes and graph topology. A word2vec model is used to construct document representation vectors by encoding the attributes of documents. The relationship of documents is then encoded into the document embedding vectors by a graph auto-encoder and similar documents are aggregated. To further improve the accuracy of the clustering results, a graph embedding model is proposed to introduce the document-document network and author-author network topology into the document vectors afterword, thus related papers are moved closer. This method utilizes the information of document attributes and relationship networks at the same time, finds document representation vectors using an unsupervised model and improves the performance of name disambiguation. Experimental results on the real author dataset AMiner show that our method is superior to several state-of-the-art graph-based solutions.
Super-resolution reconstruction algorithm by combining L1 and L0 prior models
LI Li, YIN Zengshan, SHI Shen
2022, 39 (3): 369-376.  DOI: 10.7523/j.ucas.2020.0013
Abstract ( 287 ) PDF (0KB) ( 0 )
Super-resolution (SR) reconstruction can reconstruct a high-resolution image from low-resolution image sequences and improve image quality. Reconstructing a high-resolution image with edge preserving and low noise is still a challenge in SR. Therefore, the L0 norm of the image gradient is added as prior knowledge in the L1 prior model, and a SR reconstruction algorithm by combining the L1 and L0 prior model is proposed in this paper, which not only retains the advantage of L1 prior model preserving edges, but also retains the advantage of L0 prior model suppressing noise. Compared with bicubic interpolation, total variation (TV) prior model, and L1 prior model, the validity of the algorithm is verified through the analysis of simulation experimental data and real experimental data.
A new band registration method for coated dual-lens multispectral camera
ZHENG Jiayi, FANG Junyong, WANG Xiao, ZHANG Xiaohong, LIU Xue
2022, 39 (3): 377-385.  DOI: 10.7523/j.ucas.2020.0021
Abstract ( 302 ) PDF (0KB) ( 0 )
In order to solve the problem of band registration of coated dual lens multispectral camera, a fast registration method based on improved SURF (speeded up robust features) algorithm is proposed. Firstly, the distortion of each band image is corrected. Then, the fast version of SURF algorithm (F-SURF) is used to register the undistorted band images. Compared with the original SURF algorithm, the F-SURF algorithm optimizes the number of octave, discards the rotation invariance and introduces the PROSAC algorithm. The experimental results indicate that the registration accuracy of the F-SURF algorithm is better than that of the classic SURF algorithm, and the time efficiency is significantly improved.
Spatial resolution improvement of spectrum sensing data of LEO satellite based on image super-resolution
WEI Rui, XIE Zhuochen, LIU Jie, LIU Huijie
2022, 39 (3): 386-392.  DOI: 10.7523/j.ucas.2020.0033
Abstract ( 383 ) PDF (0KB) ( 0 )
In the low orbit satellite Internet of things system, the spatial resolution of spectrum data perceived by satellites is low, and the details of the spatial electromagnetic environment are difficult to analyze. To solve this problem, this paper proposes to process the spatial distribution of the spectrum in the form of two-dimensional images, and to adopt the appropriate image super-resolution reconstruction algorithm according to the characteristics of spatial spectrum sensing data, so as to improve the spatial resolution of the spectrum, enhance the details in the spectrum situation. Simulation results show that the image signal of existence can be directly observed from the grey value. The bicubic interpolation method chosen according to spectral data characteristics, the Bayesian method based on L1 norm prior, and the learning method based on image blocks matching can effectively improves the spatial resolution of the spectrum data. When evaluating with PSNR, the reconstruction algorithm based on L1 norm prior is better. But, the learning method based on image blocks matching enhances the ripples in the spectrum sensing data. From a visual point of view, the effect of improving details is slightly better.
An artificial-potential-field method for real-time UAV navigation in unknown environments
SONG Xiaocheng, LIU Xiaopei, LU Jie
2022, 39 (3): 393-402.  DOI: 10.7523/j.ucas.2020.0022
Abstract ( 321 ) PDF (0KB) ( 6 )
This paper proposes a local planning method for real-time obstacle avoidance in unknown environments. This method constructs a Dirichlet boundary value problem once the obstacle points are obtained from sensors. This problem is solved by FDM (finite difference method), and hence it generates a Laplacian potential field based on the local map. The potential field is replaced by a new one when the sensing data got updated. This construction can efficiently deal with complex environments, and there is no local minimum in the field. The reference velocities are generated by the directions of the negative gradient in the field, which are tracked by the PID controller, in order to achieve autonomous UAV (unmanned aerial vehicle)navigation. Finally, MATLAB experiments are taken under different scenes, and the result shows this method is valid for real-time obstacle avoidance in different unknown environments.
A low cost multi-armed bandit algorithm for dense wireless network
ZHAO Yao, LUO Xiliang
2022, 39 (3): 403-409.  DOI: 10.7523/j.ucas.2020.0011
Abstract ( 155 ) PDF (0KB) ( 0 )
In recent years, people's demand for mobile wireless services has been increasing. In order to meet this challenge, ultra-dense wireless networks are considered to be the infrastructure and important components of the next-generation wireless communication network. Massive deployment of small base stations can reduce the number of network users in each cell, which can in turn provide the users with high-speed and low-latency wireless service. However, the inevitable problem brought with it at the same time is that users will cause frequent network handover when choosing access to ensure that they can access the network with the best service provider. User association problem is often modeled as the online learning model. This paper aims to find an efficient online user association scheme to deal with the additional network performance loss caused by frequent handover. Based on the analysis of the multi-armed bandit (MAB) model, this paper proposes an improved algorithm based on the arm elimination strategy, and demonstrates the effectiveness of the algorithm through rigorous theoretical analysis and numerical simulation experiments.
Energy efficient opportunistic routing for wireless multihop networks: a deep reinforcement learning approach
JIN Xiaohan, YAN Yan, ZHANG Baoxian
2022, 39 (3): 410-420.  DOI: 10.7523/j.ucas.2020.0035
Abstract ( 497 ) PDF (0KB) ( 0 )
Opportunistic routing has been an efficient approach for improving the performance of wireless multihop networks due to its salient features to take advantage of the broadcast and lossy nature of wireless channels. In this paper, we propose a deep reinforcement learning based energy efficient opportunistic routing algorithm for wireless multihop networks, which enables a learning agent to train and learn optimized routing policy to reduce the transmission time while balancing the energy consumption to extend the life of the network in an opportunistic way. Furthermore, the proposed algorithm can significantly alleviate the cold start problem and achieve better initial performance. Simulation results demonstrate that the proposed algorithm yield better performance as compared with existing algorithms.
Brief Report
Image adversarial attack algorithm based on high-dimensional feature
LIN Daquan, FAN Rui, ZHANG Liangfeng
2022, 39 (3): 421-431.  DOI: 10.7523/j.ucas.2020.0034
Abstract ( 513 ) PDF (0KB) ( 0 )
In order to attack state-of-the-art adversarial defense methods, an image adversarial attack algorithm based on high-dimensional features called FB-PGD(feature based projected gradient descent) is proposed. It increases the similarity between clean image features and target image features by adding perturbation to clean image iteratively, then adversarial examples will be generated. In the experimental section, by comparing with existing adversarial attack algorithms on different defense models, the result shows that this attack algorithm not only has strong attack performance in the previous defense methods but also increases attack success rate more than 20[WTB4]%[WTBZ] compared to common adversarial attack algorithms in two state-of-the-art defense methods on a variety of datasets. So, the adversarial attack algorithm can be used as a new benchmark to test defense.