ISSN 
2020 Vol. 49, No. 4
column
2020, 49(4): 499-510.
doi: 10.12178/1001-0548.2020274
Abstract:
Scale-free networks and related concepts including power-law distribution and preferential attachment mechanism are core concepts in network science. Researches on these concepts have a long history, but there are still some debates by now. The paper aims at reveal the links among some important historic researches on these concepts from a network perspective, including reveal the fundamental works of Price that had been neglected for a long time. This paper introduces and discusses some controversial issues on these concepts, including quarrels between Simon and Mandelbrot around mechanisms for power law in the middle of 20th century, and recently debates on whether scale-free networks are rich or rare. Finally, the paper discusses future researches on scale-free networks.
Scale-free networks and related concepts including power-law distribution and preferential attachment mechanism are core concepts in network science. Researches on these concepts have a long history, but there are still some debates by now. The paper aims at reveal the links among some important historic researches on these concepts from a network perspective, including reveal the fundamental works of Price that had been neglected for a long time. This paper introduces and discusses some controversial issues on these concepts, including quarrels between Simon and Mandelbrot around mechanisms for power law in the middle of 20th century, and recently debates on whether scale-free networks are rich or rare. Finally, the paper discusses future researches on scale-free networks.
2020, 49(4): 481-481.
Abstract:
2020, 49(4): 482-489.
doi: 10.12178/1001-0548.2020180
Abstract:
Human intelligence can make adaptive decisions according to current environment when performing tasks, and is capable of thinking and emotional awareness. Artificial intelligence can make high performance computing instead of humans. However, increasing demand makes environments and tasks become more complicated, thus human intelligence or artificial intelligence alone is unable to perform complex tasks. Human-robot collaboration intelligence system is a novel technology based on hybrid intelligence, which can perform complex tasks in high-dynamic environment. It is the extension and expansion of human behavior and intelligence to realize the complementary advantages. In this paper, the main human - robot collaborative intelligence system and its theory are summarized. And then, the clinical applications of typical human - robot collaborative intelligence system is analyzed. Finally, this paper prospects the research trends of human-machine collaborative intelligence systems.
Human intelligence can make adaptive decisions according to current environment when performing tasks, and is capable of thinking and emotional awareness. Artificial intelligence can make high performance computing instead of humans. However, increasing demand makes environments and tasks become more complicated, thus human intelligence or artificial intelligence alone is unable to perform complex tasks. Human-robot collaboration intelligence system is a novel technology based on hybrid intelligence, which can perform complex tasks in high-dynamic environment. It is the extension and expansion of human behavior and intelligence to realize the complementary advantages. In this paper, the main human - robot collaborative intelligence system and its theory are summarized. And then, the clinical applications of typical human - robot collaborative intelligence system is analyzed. Finally, this paper prospects the research trends of human-machine collaborative intelligence systems.
2020, 49(4): 490-498, 510.
doi: 10.12178/1001-0548.2020179
Abstract:
With the development of artificial intelligence (AI) technology, intelligent systems, such as intelligent driving cars and intelligent robots, gradually replace or assist human beings to do simple or complex work in various scenes. Starting from the intelligent algorithm in intelligent vehicle, this paper summarizes the research progress of artificial intelligence perception algorithm and decision algorithm in intelligent vehicle. Secondly, the uncertainty of intelligent algorithm is discussed. Finally, from the point of view of the security problems brought by the uncertainty of the intelligent algorithm, this paper discusses the significance and development of the expected functional security, and discusses the necessity of human-computer co driving to solve the expected functional security of the current intelligent driving vehicle.
With the development of artificial intelligence (AI) technology, intelligent systems, such as intelligent driving cars and intelligent robots, gradually replace or assist human beings to do simple or complex work in various scenes. Starting from the intelligent algorithm in intelligent vehicle, this paper summarizes the research progress of artificial intelligence perception algorithm and decision algorithm in intelligent vehicle. Secondly, the uncertainty of intelligent algorithm is discussed. Finally, from the point of view of the security problems brought by the uncertainty of the intelligent algorithm, this paper discusses the significance and development of the expected functional security, and discusses the necessity of human-computer co driving to solve the expected functional security of the current intelligent driving vehicle.
2020, 49(4): 511-518, 554.
doi: 10.12178/1001-0548.2019196
Abstract:
In this paper, a temporal control strategy is proposed by analyzing a rumor spreading model based on microscopic Markov chain in a two-layer network. The obtained results show that both sub-networks can show synchronized outbreak, where the maximum fractions of the infected population simultaneously emerge when a vigorous control measure is taken. In addition, the control effect depends on the selection of the network topological degree, the target layer, and the control time. The later (earlier) control or larger (smaller) network topological degree needs more (less) resource. The control effect is better when the target layer is chosen at a larger network topological degree. These results imply that the temporal control strategy is effective in preventing rumor spreading.
In this paper, a temporal control strategy is proposed by analyzing a rumor spreading model based on microscopic Markov chain in a two-layer network. The obtained results show that both sub-networks can show synchronized outbreak, where the maximum fractions of the infected population simultaneously emerge when a vigorous control measure is taken. In addition, the control effect depends on the selection of the network topological degree, the target layer, and the control time. The later (earlier) control or larger (smaller) network topological degree needs more (less) resource. The control effect is better when the target layer is chosen at a larger network topological degree. These results imply that the temporal control strategy is effective in preventing rumor spreading.
2020, 49(4): 519-523.
doi: 10.12178/1001-0548.2019087
Abstract:
At present, the research on the importance of temporal network nodes is mainly carried out from the aspects of timing path, connectivity, network efficiency and so on. In this paper, we consider the inter-layer connection and inter-layer coupling, introduce the ranking aggregation theory based on scoring matrix, and propose a method based on ranking aggregation to identify the importance of nodes in temporal networks. Manufacturing and Enrons datasets show that the average increase of Spearman correlation coefficient based on ranking aggregation is 2.41% and 18.63%, which shows the applicability and effectiveness of this method in the measurement of node importance in temporal network.
At present, the research on the importance of temporal network nodes is mainly carried out from the aspects of timing path, connectivity, network efficiency and so on. In this paper, we consider the inter-layer connection and inter-layer coupling, introduce the ranking aggregation theory based on scoring matrix, and propose a method based on ranking aggregation to identify the importance of nodes in temporal networks. Manufacturing and Enrons datasets show that the average increase of Spearman correlation coefficient based on ranking aggregation is 2.41% and 18.63%, which shows the applicability and effectiveness of this method in the measurement of node importance in temporal network.
2020, 49(4): 524-529.
doi: 10.12178/1001-0548.2019075
Abstract:
In this paper, we analyze the cultural pattern of Chinese online music society using the dataset collected from NetEase Cloud Music. Our analysis focuses on the region diversity of user’s favorite music, and average preference diversities defined on the region diversity. We calculate the region diversity and the average preference diversity of users at each region, and analyze the correlation between these diversities and the development level of economy, and find that users in the developed regions have higher probability in following niche music. And also, we detect two major musical cultural circles from the differences on the distribution of user’s region diversity of each area, and analyze the correlation between region diversity and the diversity on the source of residents to investigate the cultural impact of migration. These findings provides series of global feature on Chinese cultural structure from the perspective of music.
In this paper, we analyze the cultural pattern of Chinese online music society using the dataset collected from NetEase Cloud Music. Our analysis focuses on the region diversity of user’s favorite music, and average preference diversities defined on the region diversity. We calculate the region diversity and the average preference diversity of users at each region, and analyze the correlation between these diversities and the development level of economy, and find that users in the developed regions have higher probability in following niche music. And also, we detect two major musical cultural circles from the differences on the distribution of user’s region diversity of each area, and analyze the correlation between region diversity and the diversity on the source of residents to investigate the cultural impact of migration. These findings provides series of global feature on Chinese cultural structure from the perspective of music.
2020, 49(4): 530-536.
doi: 10.12178/1001-0548.2019057
Abstract:
The evolution of a collaborative innovation network depends on the interrelationships among the innovation subjects. Every single small change affects the network topology, which leads to different evolution results. A logical relationship exists between network evolution and innovative behaviors. An accurate understanding of the characteristics of the network structure can help the innovative subjects to adopt appropriate innovative behaviors. Based on the resource-based theory, complex network theory and exploratory analysis of two typical cases, this paper analyzes the behavior rules of collaborative innovation network, and establishes the connection rule based on resource priority. Fully combining the stage-cooperation characteristic of collaborative innovation, this paper builds a network evolution model. Simulation experiments are carried out on network characteristics and evolution rules. The results show that resource priority connection conforms to the real situation, and is conducive to maintaining the continuous innovation network vitality, and thus has certain management practical significance for behavioral decisions of enterprise, government and so on.
The evolution of a collaborative innovation network depends on the interrelationships among the innovation subjects. Every single small change affects the network topology, which leads to different evolution results. A logical relationship exists between network evolution and innovative behaviors. An accurate understanding of the characteristics of the network structure can help the innovative subjects to adopt appropriate innovative behaviors. Based on the resource-based theory, complex network theory and exploratory analysis of two typical cases, this paper analyzes the behavior rules of collaborative innovation network, and establishes the connection rule based on resource priority. Fully combining the stage-cooperation characteristic of collaborative innovation, this paper builds a network evolution model. Simulation experiments are carried out on network characteristics and evolution rules. The results show that resource priority connection conforms to the real situation, and is conducive to maintaining the continuous innovation network vitality, and thus has certain management practical significance for behavioral decisions of enterprise, government and so on.
2020, 49(4): 537-541.
doi: 10.12178/1001-0548.2019190
Abstract:
With the development of network function virtualization (NFV), it is more flexible and important for service function chain (SFC) constrained traffic flows to find the shortest path to get through the network. Existing works only consider conditions within a single-domain network. However, when it comes to multiple domains or geo-distributed networks, due to the network’s concern of revealing sensitive information, member networks conceal their own network details to each other. This paper proposed a new algorithm, privacy-preserved multi-domain shortest path problem under service function chaining constraints, to solve the multi-domain transfer problem of traffic flows, preserving the privacy information of each member network as well as choosing the shortest path for the flows.
With the development of network function virtualization (NFV), it is more flexible and important for service function chain (SFC) constrained traffic flows to find the shortest path to get through the network. Existing works only consider conditions within a single-domain network. However, when it comes to multiple domains or geo-distributed networks, due to the network’s concern of revealing sensitive information, member networks conceal their own network details to each other. This paper proposed a new algorithm, privacy-preserved multi-domain shortest path problem under service function chaining constraints, to solve the multi-domain transfer problem of traffic flows, preserving the privacy information of each member network as well as choosing the shortest path for the flows.
2020, 49(4): 542-547.
doi: 10.12178/1001-0548.2019235
Abstract:
The black hole attack is the main kind of attacks against AODV protocol in an Ad hoc network, black hole attack is a black hole node spoofing a legitimate node by changing the serial number or hop count, thereby drop the received packets, a much more threat by a joint attack of multiple black holes. This paper proposes an authentication chain mechanism based on the secure recursive function to overcome the black hole attack, by which the successive nodes of a route are unified only on the deterministic state transition relation of the recursive function to implement a unique association for security so that the whole route form an authentication chain. With the mechanism of this paper, even if the pseudo-random linear sequence as a kind of simple recursive function is adopted, as long as its linear complexity is greater than the number of nodes in the authenticating chain so that the attacker cannot obtain continuous state values of more than two times the linear complexity of the sequence, and then the secure authentication of the route can be guaranteed. Therefore, the mechanism and method presented in this paper is a novel and effective method to defend against the black hole attack.
The black hole attack is the main kind of attacks against AODV protocol in an Ad hoc network, black hole attack is a black hole node spoofing a legitimate node by changing the serial number or hop count, thereby drop the received packets, a much more threat by a joint attack of multiple black holes. This paper proposes an authentication chain mechanism based on the secure recursive function to overcome the black hole attack, by which the successive nodes of a route are unified only on the deterministic state transition relation of the recursive function to implement a unique association for security so that the whole route form an authentication chain. With the mechanism of this paper, even if the pseudo-random linear sequence as a kind of simple recursive function is adopted, as long as its linear complexity is greater than the number of nodes in the authenticating chain so that the attacker cannot obtain continuous state values of more than two times the linear complexity of the sequence, and then the secure authentication of the route can be guaranteed. Therefore, the mechanism and method presented in this paper is a novel and effective method to defend against the black hole attack.
2020, 49(4): 548-554.
doi: 10.12178/1001-0548.2019167
Abstract:
In the millimeter-wave massive multiple-input multiple-output (Massive-MIMO) system, the analog precoding algorithm with low-resolution phase shifter has a high algorithm complexity. To cope with the computational complexity problem, an improved iterative phase hybrid precoding algorithm is proposed. It decomposes the analog precoding optimization matrix into column vectors one by one, and uses the second-order rank matrix to obtain the suboptimal optimization function approximately. Then, the optimization function is transformed into a summation form and the optimal phase is obtained by iterative convergence. Meanwhile, in order to maximize hardware efficiency, an analog beamforming scheme with 1 bit resolution phaser is designed. Theoretical analysis and simulation results show that the proposed algorithm has lower computational complexity and improving spectral efficiency.
In the millimeter-wave massive multiple-input multiple-output (Massive-MIMO) system, the analog precoding algorithm with low-resolution phase shifter has a high algorithm complexity. To cope with the computational complexity problem, an improved iterative phase hybrid precoding algorithm is proposed. It decomposes the analog precoding optimization matrix into column vectors one by one, and uses the second-order rank matrix to obtain the suboptimal optimization function approximately. Then, the optimization function is transformed into a summation form and the optimal phase is obtained by iterative convergence. Meanwhile, in order to maximize hardware efficiency, an analog beamforming scheme with 1 bit resolution phaser is designed. Theoretical analysis and simulation results show that the proposed algorithm has lower computational complexity and improving spectral efficiency.
2020, 49(4): 555-562.
doi: 10.12178/1001-0548.2019169
Abstract:
The network architecture and the heterogeneity of hybrid energy supply will lead to extreme imbalance of load distribution in 5G small cell base stations (SBS), which causes the waste of resources. A new challenge is faced about how to utilize renewable energy and radio resource efficiently. In this paper, an approach of user association mechanism and resource allocation strategy is proposed for a given communication rate. In order to minimize the total energy of system, the bias factor of the SBS ’ favorite is used to describe the extent of user association, and an estimation algorithm of energy hungry probability is presented for uncertainty reduction in energy by utilizing the large deviation theory. Moreover, the resource allocation strategy is proposed to allocate band resource reasonably through the Lagrange dual algorithm. Numerical simulation results show that the energy consumption of the proposed algorithm can reduce 82.47% than that of the maximum received power algorithm. Also, the utilization rate of green energy of the proposed algorithm will increase 48% than that of the maximum channel gain algorithm.
The network architecture and the heterogeneity of hybrid energy supply will lead to extreme imbalance of load distribution in 5G small cell base stations (SBS), which causes the waste of resources. A new challenge is faced about how to utilize renewable energy and radio resource efficiently. In this paper, an approach of user association mechanism and resource allocation strategy is proposed for a given communication rate. In order to minimize the total energy of system, the bias factor of the SBS ’ favorite is used to describe the extent of user association, and an estimation algorithm of energy hungry probability is presented for uncertainty reduction in energy by utilizing the large deviation theory. Moreover, the resource allocation strategy is proposed to allocate band resource reasonably through the Lagrange dual algorithm. Numerical simulation results show that the energy consumption of the proposed algorithm can reduce 82.47% than that of the maximum received power algorithm. Also, the utilization rate of green energy of the proposed algorithm will increase 48% than that of the maximum channel gain algorithm.
2020, 49(4): 563-568.
doi: 10.12178/1001-0548.2019214
Abstract:
Image noise is the main factor affecting the accuracy of monocular vision positioning. Based onbidimensional empirical mode decomposition (BEMD) and threshold denoising, an interval threshold imagenoise filtering method based on bidimensional ensemble empirical mode decomposition (BEEMD) isproposed. The image is decomposed into multiple bidimensional intrinsic mode function (IMF) componentsand one residual component of different scales by BEEMD. The pure noise IMF components are eliminatedaccording to the 2-norm criterion and the probability density function method of image and IMFcomponents, the reasonable regulatory factor α is selected, and the image denoising is realized by theimproved interval threshold denoising method. The proposed algorithm is applied to monocular visionand compared with the BEMD algorithm. The results show that the method could not only effectivelysuppress the modal aliasing problem in BEMD, but also effectively reduce the influence of image noise, so theaccuracy and reliability of monocular vision are improved.
Image noise is the main factor affecting the accuracy of monocular vision positioning. Based onbidimensional empirical mode decomposition (BEMD) and threshold denoising, an interval threshold imagenoise filtering method based on bidimensional ensemble empirical mode decomposition (BEEMD) isproposed. The image is decomposed into multiple bidimensional intrinsic mode function (IMF) componentsand one residual component of different scales by BEEMD. The pure noise IMF components are eliminatedaccording to the 2-norm criterion and the probability density function method of image and IMFcomponents, the reasonable regulatory factor α is selected, and the image denoising is realized by theimproved interval threshold denoising method. The proposed algorithm is applied to monocular visionand compared with the BEMD algorithm. The results show that the method could not only effectivelysuppress the modal aliasing problem in BEMD, but also effectively reduce the influence of image noise, so theaccuracy and reliability of monocular vision are improved.
2020, 49(4): 569-574.
doi: 10.12178/1001-0548.2019145
Abstract:
Based on the traditional Pan-sharpening technology, we propose a high spectral and multispectral integration framework. In the framework, hyperspectral and multispectral image fusion (HS-MS) problem is simplified to a number of multiband and single band (MB-IB) image fusion problems. On this basis, an image fusion algorithm based on collaborative representation using local adaptive dictionary pair (LACRF) algorithm based on local adaptive (LA) dictionary and collaborative representation (CR) is proposed for image fusion of each multi-band and single-band, to obtain multi-band (HRMB) images with high spatial resolution, and finally to obtain hyperspectral images (HHS) with high spatial resolution. According to the experimental results, LACRF algorithm has a better fusion effect.
Based on the traditional Pan-sharpening technology, we propose a high spectral and multispectral integration framework. In the framework, hyperspectral and multispectral image fusion (HS-MS) problem is simplified to a number of multiband and single band (MB-IB) image fusion problems. On this basis, an image fusion algorithm based on collaborative representation using local adaptive dictionary pair (LACRF) algorithm based on local adaptive (LA) dictionary and collaborative representation (CR) is proposed for image fusion of each multi-band and single-band, to obtain multi-band (HRMB) images with high spatial resolution, and finally to obtain hyperspectral images (HHS) with high spatial resolution. According to the experimental results, LACRF algorithm has a better fusion effect.
2020, 49(4): 575-583.
doi: 10.12178/1001-0548.2018168
Abstract:
Using jamming efficient carrier-to-noise ratio and code tracking error as the jamming effect evaluation index, the influence degree of current GPS and Galileo navigation signals suffered from single frequency, band-limited Gaussian noise and matching spectrum jamming is researched respectively by theoretical derivation and simulation analysis. The inner relationship among signal modulation mode, receiver parameters set and enhanced anti-jamming performance is obtained. Comparing the signal systems of Beidou Navigation Satellite System (BDS) with GPS and Galileo, we achieve the following suggestions for enhancing anti-jamming performance: B1-C of BDS can be optimized to the modulation combination of CBOC(6,1,1/11); B2 can be designed TD-AltBOC(15,10) modulation with the same pseudo random modulation sequence of pilot channel; and B1-A and B3-A could employ the signal modulated by TDDM-BOCc(15,2.5) and BOCc(15,2.5).
Using jamming efficient carrier-to-noise ratio and code tracking error as the jamming effect evaluation index, the influence degree of current GPS and Galileo navigation signals suffered from single frequency, band-limited Gaussian noise and matching spectrum jamming is researched respectively by theoretical derivation and simulation analysis. The inner relationship among signal modulation mode, receiver parameters set and enhanced anti-jamming performance is obtained. Comparing the signal systems of Beidou Navigation Satellite System (BDS) with GPS and Galileo, we achieve the following suggestions for enhancing anti-jamming performance: B1-C of BDS can be optimized to the modulation combination of CBOC(6,1,1/11); B2 can be designed TD-AltBOC(15,10) modulation with the same pseudo random modulation sequence of pilot channel; and B1-A and B3-A could employ the signal modulated by TDDM-BOCc(15,2.5) and BOCc(15,2.5).
2020, 49(4): 584-589.
doi: 10.12178/1001-0548.2019178
Abstract:
Main-lobe jamming is an important jamming pattern faced by radar in modern electronic warfare. In multiple-radar system, the received jamming signals in different receiving stations would be highly coherent due to the wide beam of jammer. However, the received target echoes are independent because of the random fluctuation of target radar cross section (RCS) with the variation of view angle. According to this fact, this paper proposes a main-lobe jamming suppression method based on the amplitude ratio feature. After the time alignment of each receiving station is performed, the time domain signal is converted into the amplitude ratio feature domain by making the amplitude ratio of the echo signals at each time point. In this feature domain, jamming signals would be "smoothed" in the background due to its correlation, and targets would be "highlighted" due to the independence of the target echoes. Finally, in the amplitude ratio feature domain, the target is subjected to constant false alarm detection with a fixed threshold. Simulation experiments show that the proposed method can effectively suppress the main-lobe jamming and improve the target detection performance under the main-lobe jamming condition.
Main-lobe jamming is an important jamming pattern faced by radar in modern electronic warfare. In multiple-radar system, the received jamming signals in different receiving stations would be highly coherent due to the wide beam of jammer. However, the received target echoes are independent because of the random fluctuation of target radar cross section (RCS) with the variation of view angle. According to this fact, this paper proposes a main-lobe jamming suppression method based on the amplitude ratio feature. After the time alignment of each receiving station is performed, the time domain signal is converted into the amplitude ratio feature domain by making the amplitude ratio of the echo signals at each time point. In this feature domain, jamming signals would be "smoothed" in the background due to its correlation, and targets would be "highlighted" due to the independence of the target echoes. Finally, in the amplitude ratio feature domain, the target is subjected to constant false alarm detection with a fixed threshold. Simulation experiments show that the proposed method can effectively suppress the main-lobe jamming and improve the target detection performance under the main-lobe jamming condition.
2020, 49(4): 590-596.
doi: 10.12178/1001-0548.2019285
Abstract:
The accurate extraction of brain tumor from magnetic resonance imaging (MRI) images is the key of clinical diagnostics and treatment planning. A two-stage coarse-to-fine brain tumor segmentation framework is proposed for brain tumor segmentation in multi-modal MRI images. There are two parts in our framework. One is coarse segmentation part, the other one is fine segmentation part. A five-classification task is performed in coarse segmentation part with a deep convolutional neural network. Four coarse probability maps are generated according the five-classification results. The fine segmentation part takes these coarse maps as the mask to guide the network to pay more attention on regions of high probability. Besides, to alleviate the data imbalance problem, there is a two-branch output structure in fine segmentation part. One branch outputs five-classification results with a mask soft-max cross entropy loss function. The other branch outputs a binary result which labels the whole tumor with a mean-square loss function. Our proposal was validated in the BRATS 2015 dataset. It can be proven that our approach achieves a competitive result.
The accurate extraction of brain tumor from magnetic resonance imaging (MRI) images is the key of clinical diagnostics and treatment planning. A two-stage coarse-to-fine brain tumor segmentation framework is proposed for brain tumor segmentation in multi-modal MRI images. There are two parts in our framework. One is coarse segmentation part, the other one is fine segmentation part. A five-classification task is performed in coarse segmentation part with a deep convolutional neural network. Four coarse probability maps are generated according the five-classification results. The fine segmentation part takes these coarse maps as the mask to guide the network to pay more attention on regions of high probability. Besides, to alleviate the data imbalance problem, there is a two-branch output structure in fine segmentation part. One branch outputs five-classification results with a mask soft-max cross entropy loss function. The other branch outputs a binary result which labels the whole tumor with a mean-square loss function. Our proposal was validated in the BRATS 2015 dataset. It can be proven that our approach achieves a competitive result.
2020, 49(4): 597-602.
doi: 10.12178/1001-0548.2020001
Abstract:
In order to solve the problem of label noise in breast ultrasound image classification, an efficient method called cooperative label correction network (COLC-Net) is proposed. In this method, based on the noise distribution characteristics of the breast ultrasound BI-RADS (breast imaging-reporting and data system) rating, soft labels are proposed for breast ultrasound images, and two networks are proposed for collaborative training. Excellent knowledge is distilled from the two networks to modify the soft labels. With the increase of the accuracy of soft labels, the negative effects of noise labels can be reduced and the learning of clean labels can be enhanced. In order to verify the effectiveness of the method, extensive comparisons are conducted with existing state-of-the-art methods on the dataset. The results demonstrate the effectiveness of the proposed method.
In order to solve the problem of label noise in breast ultrasound image classification, an efficient method called cooperative label correction network (COLC-Net) is proposed. In this method, based on the noise distribution characteristics of the breast ultrasound BI-RADS (breast imaging-reporting and data system) rating, soft labels are proposed for breast ultrasound images, and two networks are proposed for collaborative training. Excellent knowledge is distilled from the two networks to modify the soft labels. With the increase of the accuracy of soft labels, the negative effects of noise labels can be reduced and the learning of clean labels can be enhanced. In order to verify the effectiveness of the method, extensive comparisons are conducted with existing state-of-the-art methods on the dataset. The results demonstrate the effectiveness of the proposed method.
2020, 49(4): 603-608, 621.
doi: 10.12178/1001-0548.2019039
Abstract:
Stereoscopic motion in depth is a critical factor affecting visual perception. This paper utilized electroencephalogram (EEG) technique to investigate the effects of motion in depth. Two types of velocities (comfort and discomfort) are selected for the EEG experiment based on subjective results. The squared biserial correlation coefficient r2 is computed. The classification accuracies of different brain areas are obtained by SVM after CSP. The results prove that the EEG sub-band α and β in partial area are more active in stimuli period, and the most discriminative waveband is concentrated on α band is the parietal area for two types of motion-in-depth. The higher accuracy on Dorsal pathways shows that it is feasible to identify two labels caused by motion in depth automatically using EEG.
Stereoscopic motion in depth is a critical factor affecting visual perception. This paper utilized electroencephalogram (EEG) technique to investigate the effects of motion in depth. Two types of velocities (comfort and discomfort) are selected for the EEG experiment based on subjective results. The squared biserial correlation coefficient r2 is computed. The classification accuracies of different brain areas are obtained by SVM after CSP. The results prove that the EEG sub-band α and β in partial area are more active in stimuli period, and the most discriminative waveband is concentrated on α band is the parietal area for two types of motion-in-depth. The higher accuracy on Dorsal pathways shows that it is feasible to identify two labels caused by motion in depth automatically using EEG.
2020, 49(4): 609-614.
doi: 10.12178/1001-0548.2019162
Abstract:
Infrared target classification has important values in target recognition. At present, convolutional neural network has achieved excellent performance in visible image classification. However, for infrared images, the available networks can't achieve satisfying results due to the small number of annotated samples and large imaging differences. In this paper, visible images are used as source domain, infrared images as target domain. Transfer learning is used to address the challenges in the deep learning framework. In the transfer learning, if the target domain network can represent the distribution of its domain well, the performance and generalization of the target domain network should be more effective. Therefore, the convolutional autoencoder is trained with a large number of unannotated infrared samples, which greatly enhances the feature representation in the infrared image domain. By reducing the feature distribution distance between the two domains, the feature distributions become similar. The classification performance in the source domain is transferred to the target domain. With the changes above, the accuracy rate is improved by 11.27% compared with the method based on the visible images fine-tuning.
Infrared target classification has important values in target recognition. At present, convolutional neural network has achieved excellent performance in visible image classification. However, for infrared images, the available networks can't achieve satisfying results due to the small number of annotated samples and large imaging differences. In this paper, visible images are used as source domain, infrared images as target domain. Transfer learning is used to address the challenges in the deep learning framework. In the transfer learning, if the target domain network can represent the distribution of its domain well, the performance and generalization of the target domain network should be more effective. Therefore, the convolutional autoencoder is trained with a large number of unannotated infrared samples, which greatly enhances the feature representation in the infrared image domain. By reducing the feature distribution distance between the two domains, the feature distributions become similar. The classification performance in the source domain is transferred to the target domain. With the changes above, the accuracy rate is improved by 11.27% compared with the method based on the visible images fine-tuning.
2020, 49(4): 615-621.
doi: 10.12178/1001-0548.2019151
Abstract:
Road traffic accidents are a concrete manifestation of road traffic safety levels. In the current traffic accident prediction work, there is an insufficient mining of the time series relationship in the data, the predicted time period is too macroscopic, and the influencing factors related to traffic accidents are missing. Aiming at the above problems, a gradient boosted regression tree (GBRT) traffic accident model based on time series relationship is proposed. The model predicts the number of daily traffic accidents, deaths, and the number of vehicles involved in Leicester, England, from 2005 to 2015. Experimental results show that adding the time series relationship helps to improve the prediction accuracy of the model. The prediction results serve as a reference for the decision-making of the traffic management department. The modeling method brings positive reference significance to the modeling work of the same type of prediction problems.
Road traffic accidents are a concrete manifestation of road traffic safety levels. In the current traffic accident prediction work, there is an insufficient mining of the time series relationship in the data, the predicted time period is too macroscopic, and the influencing factors related to traffic accidents are missing. Aiming at the above problems, a gradient boosted regression tree (GBRT) traffic accident model based on time series relationship is proposed. The model predicts the number of daily traffic accidents, deaths, and the number of vehicles involved in Leicester, England, from 2005 to 2015. Experimental results show that adding the time series relationship helps to improve the prediction accuracy of the model. The prediction results serve as a reference for the decision-making of the traffic management department. The modeling method brings positive reference significance to the modeling work of the same type of prediction problems.
2020, 49(4): 622-630.
doi: 10.12178/1001-0548.2019116
Abstract:
A discrete whale optimization algorithm (DWOA) is proposed to overcome the defects of low convergence and the lack of ability to solve discrete optimization problems when solving high-dimensional complex problems. In DWOA, the convergence factor is introduced to adjust the distance of the individual from the optimal whale position, the adaptive inertia weight is designed to balance the global exploration and local exploitation ability, the whale optimization algorithm (WOA) is discretized by the improved Sigmoid function. The optimization experiments are conducted on the 9 benchmark functions and oilfield measures planning. Simulation results show that the proposed DWOA has a great improvement in convergence speed and convergence precision.
A discrete whale optimization algorithm (DWOA) is proposed to overcome the defects of low convergence and the lack of ability to solve discrete optimization problems when solving high-dimensional complex problems. In DWOA, the convergence factor is introduced to adjust the distance of the individual from the optimal whale position, the adaptive inertia weight is designed to balance the global exploration and local exploitation ability, the whale optimization algorithm (WOA) is discretized by the improved Sigmoid function. The optimization experiments are conducted on the 9 benchmark functions and oilfield measures planning. Simulation results show that the proposed DWOA has a great improvement in convergence speed and convergence precision.
2020, 49(4): 631-635.
doi: 10.12178/1001-0548.2020076
Abstract:
MXenes, a new and intriguing family of two-dimensional (2D) materials, have recently attracted considerable attention owing to their excellent properties such as high electrical conductivity and mobility, tunable structure, and termination groups. In this study, Ti3C2Tx (a typical MXene material) was incorporated as an interface modification layer between the electronic transport layer and the perovskite absorber layer of perovskite solar cells, with the aim of efficiency enhancement. Results show that the average grain size of the perovskite increased from 0.46 to 1.16 μm after the introduction of Ti3C2Tx compared with the pristine sample, and the power conversion efficiency was improved from 15.78% to 19.39%. This work brings opportunities for the research of MXene as potential materials in high-performance perovskite solar cells
MXenes, a new and intriguing family of two-dimensional (2D) materials, have recently attracted considerable attention owing to their excellent properties such as high electrical conductivity and mobility, tunable structure, and termination groups. In this study, Ti3C2Tx (a typical MXene material) was incorporated as an interface modification layer between the electronic transport layer and the perovskite absorber layer of perovskite solar cells, with the aim of efficiency enhancement. Results show that the average grain size of the perovskite increased from 0.46 to 1.16 μm after the introduction of Ti3C2Tx compared with the pristine sample, and the power conversion efficiency was improved from 15.78% to 19.39%. This work brings opportunities for the research of MXene as potential materials in high-performance perovskite solar cells
2020, 49(4): 636-640.
doi: 10.12178/1001-0548.2019222
Abstract:
Robotic touch sensors need to satisfy both accurately sense on the force and precisely position of the force. In this paper, PVDF-TrFE piezoelectric film was fabricated on a flexible circuit board through an in-situ polarization method. The crystallinity of PVDF is improved by a small amount of hydroxyapatite doping, which is beneficial to the conversion of β crystal form, and thus increases the piezoelectric coefficients of PVDF. With the increased piezoelectric coefficient, the electrical response precision of PVDF to force is improved, and makes the PVDF-based pressure sensing device more sensible. The flexible PVDF piezoelectric sensor is fabricated by using a flexible PCB, which can not only realize three-dimensional sensing of the force level, but also locate the position of the force, so that it can be applied to the tactile sensing of the robot.
Robotic touch sensors need to satisfy both accurately sense on the force and precisely position of the force. In this paper, PVDF-TrFE piezoelectric film was fabricated on a flexible circuit board through an in-situ polarization method. The crystallinity of PVDF is improved by a small amount of hydroxyapatite doping, which is beneficial to the conversion of β crystal form, and thus increases the piezoelectric coefficients of PVDF. With the increased piezoelectric coefficient, the electrical response precision of PVDF to force is improved, and makes the PVDF-based pressure sensing device more sensible. The flexible PVDF piezoelectric sensor is fabricated by using a flexible PCB, which can not only realize three-dimensional sensing of the force level, but also locate the position of the force, so that it can be applied to the tactile sensing of the robot.
