ISSN 0253-2778

CN 34-1054/N

2019 Vol. 49, No. 10

Display Method:
Original Paper
Time-optimal control of open quantum ensembles based on sampling and learning
QI Peng, KUANG Sen
2019, 49(10): 775-780. doi: 10.3969/j.issn.0253-2778.2019.10.001
For open quantum ensembles with Hamiltonian fluctuations composed of a large number of single quantum systems, a two-stage approximate time-optimal control algorithm is proposed in the framework of coherence vectors of density matrices and achieves a high-fidelity state transition of all member systems to a common target state within an approximate minimum control time. According to the parameter distribution rule that characterizes Hamiltonian fluctuations, this algorithm first samples the whole ensemble to obtain a sample system set. Then, based on the obtained sample system set and via the basic gradient method, the fidelity and the control time are optimized in the two stages respectively, and the resulting optimal control law is obtained. Numerical simulation experiments on a two-level open quantum ensemble verify the effectiveness of the proposed algorithm.
Research on high-order residual convolution neural network for crop disease recognition application
ZENG Weihui, LI Miao, ZHANG Jian, HUANG Xiaoping, WANG Jingxian, YUAN Yuan
2019, 49(10): 781-790. doi: 10.3969/j.issn.0253-2778.2019.10.002
Current research works focusing on the image recognition of crop disease in simple background have achieved great success. However, when handling the problem of crop disease recognition with various noise and complex backgrounds, it is difficult to meet the requirement of recognition accuracy. To address these issues, a new high-order residual convolution neural network for crop disease recognition is proposed, which can realize crop disease recognition that is both accurate and anti-interference. Extensive experimental results demonstrate that the proposed method has high accuracy, strong robustness as well as good anti-interference ability, and can better meet the practical application requirements for crop disease recognition.
Medical image segmentation algorithm based on dictionary learning and sparse clustering
ZHANG Binkai, WANG Xiang, ZHENG Jinjin
2019, 49(10): 791-796. doi: 10.3969/j.issn.0253-2778.2019.10.003
To improve the segmentation performance of medical images, dictionary learning was combined with clustering algorithm, and a medical image segmentation algorithm was proposed taking dictionaries as clustering centers and using sparse representation to cluster for segmentation. For a single medical image, unsupervised adaptive segmentation can be achieved by alternately iterating the sparse coding and updating the dictionary to convergence. For the medical image sequence, the sample images can be picked to obtain the trained dictionaries to complete the segmentation of the image sequence. According to the segmentation results of the synthetic images and the magnetic resonance images of the human brain from SBD database, it can be perceived that the proposed algorithm could not only improve segmentation accuracy, but also maintain the accuracy and consistency of sequential medical image segmentation.
Active user detection and channel estimation based on expectation propagation
DAI Weijia, LI Letian, ZHOU Wuyang
2019, 49(10): 797-804. doi: 10.3969/j.issn.0253-2778.2019.10.004
In the 5th-generation (5G) wireless communication network, massive machine type communication (mMTC) is an emerging research topic. For mMTC, non-orthogonal multiple access (NOMA) has been proposed to support its large-scale connectivity. Due to the sparsity of mMTC, compressed sensing based algorithms can be used to identify the active users and recover the sparse channel state information (CSI) vector. A Bayesian message passing algorithm based on expectation propagation (EP) is proposed for joint active user detection (AUD) and channel estimation (CE) in NOMA. The proposed method approximates the complicated target distribution with a Gaussian distribution to achieve linear complexity. By introducing a damping factor, the convergence performance of the algorithm can be effectively ensured. Simulations demonstrate that the EP-based algorithm can achieve better performance in joint AUD and CE than the exiting algorithms, especially in the low SNR regime.
Human posture recognition method based on indoor positioning technology
HUANG Xiaoping, ZHANG Jian, HU Zelin, LI Miao, ZENG Weihui, LI Hualong
2019, 49(10): 805-811. doi: 10.3969/j.issn.0253-2778.2019.10.005
Solitary elderly person posture recognition, especially when falling down, is a problem of concern today. The traditional method based on machine vision is flawed with too much privacy invasion, high cost and complex factors such as the implementation process, while the method based on acceleration sensor has a lower recognition rate in the stillness of the gesture. This paper introduces a new kind of body posture recognition scheme that employs indoor positioning technologies. The main job is to build an indoor positioning system, and paste tags to the key parts of the clothes and hat. The tags can receive ultra-wideband (UWB) signal from the positioning system. The UWB signal is used to get the distance which is important for the positioning. Finally, body posture can be easily recognized. In gesture recognition algorithm, this paper USES the least squares and the improved extend Kalman filter to suppress the noise of the distances measurement, so as to improve the accuracy of location. The simulation algorithm shows that the improved extend Kalman filter is effective.
Robot control policy transfer based on progressive neural network
SUI Hongjian, SHANG Weiwei, LI Xiang, CONG Shuang
2019, 49(10): 812-819. doi: 10.3969/j.issn.0253-2778.2019.10.006
In the field of robotic control, it is appealing to solve complicated control tasks through deep learning techniques. However, collecting enough robot operating data to train deep learning models is difficult. Thus, in this paper a transfer approach based on progressive neural network (PNN) and deep deterministic policy gradient (DDPG) is proposed. By linking the current task model and pretrained task models in the model pool with a novel structure, the control strategy in the pretrained task models is transferred to the current task model. Simulation experiments validate that, the proposed approach can successfully transfer control policies learned from the source task to the current task. And compared with other baselines, the proposed approach takes remarkably less time to achieve the same performance in all the experiments.
Cost optimization of request dispatching and container deployment in cloudlets
ZHENG Xiaojie, LI Jing
2019, 49(10): 820-827. doi: 10.3969/j.issn.0253-2778.2019.10.007
With the development of the Internet of Things (IoT), cloudlet is serving more low-latency, high-bandwidth applications. The application request is dynamic in time and space. If a cloudlet only processes the surrounding requests, some cloudlets will be overloaded while others are underloaded. In addition, IoT applications vary in importance. Some unimportant services with large requests may preempt the cloudlet resources, resulting in critical services unable to be executed. Cloudlet load imbalance and critical service starvation will increase the cost of the cloud infrastructure provider. Cost optimization of request dispatching and container deployment in container-based cloudlets are investigated and a cost optimization greedy algorithm named CO-Greedy is proposed which optimizes cost by dispatching the request to the surrounding cloudlet. The experimental results show that the algorithm has better performance in all scenarios.
Feature fusion-based face verification on second generation identity card
WANG Zhong, CHEN Enhong, LIU Guiquan
2019, 49(10): 828-834. doi: 10.3969/j.issn.0253-2778.2019.10.008
The second-generation ID card face verification refers to judging whether the photo on the second-generation ID card matches its user Due to its low resolution, the second-generation ID card photo differs greatly from the photo taken on the spot in terms of clarity, facial changes, and the external environment, resulting in the low recognition rate of the conventional face recognition method. To solve this problem, t a second-generation ID card facial verification system based on feature fusion is proposed. The system consists of five parts: image acquisition, preprocessing, feature extraction, feature comparison, and result judgment. First, the second-generation ID card image and camera photo are collected and image preprocessing is performed. The global and local features of the second-generation card photo and camera photo are then extracted separately. Global features are extracted by PCA and LDA methods, and local features are extracted by the histogram directional binary code (HDBC) method. Then, the global and local features are calculated in the common feature space, and the similarity between the global features local features is obtained. Finally, the user of the second-generation ID card is tested based on the given threshold. Experiments have been performed on a large number of real second-generation ID card datasets. Compared with the traditional single feature extraction algorithms, the recognition rate of the proposed method is significantly improved.
Collaborative filtering recommendation algorithm based on semantic similarity
WANG Gensheng, PAN Fangzheng
2019, 49(10): 835-841. doi: 10.3969/j.issn.0253-2778.2019.10.009
To solve the problem that collaborative filtering recommendation algorithm does not consider the semantic relationship between recommendation objects,an improved collaborative filtering recommendation algorithm based on semantic similarity of recommendation objects is proposed. First,the semantic information of the recommended object is embedded into a low dimensional semantic space by using the knowledge map representation learning algorithm;then the semantic similarity between the recommended objects is calculated and integrated into the similarity calculation of collaborative filtering recommendation algorithm, thus compensating for the shortcoming that the collaborative filtering recommendation algorithm does not consider the semantic knowledge of the recommendation object. The experimental results show that the improved algorithm has higher accuracy, recall and coverage than the traditional collaborative filtering recommendation algorithm.
Parallel ISOMAP algorithm based on Spark
SHI Lukui, GUO Linlin, FANG Zizhe, ZHANG Jun
2019, 49(10): 842-850. doi: 10.3969/j.issn.0253-2778.2019.10.010
To reduce the dimension of the nonlinear high-dimensional data in the big data environment, a parallel ISOMAP algorithm based on Spark is proposed, where a Spark-based parallel block Davidson method is designed and implemented to quickly solve eigenvalues and eigenvectors of the large scale matrices. Simultaneously, a row-block matrix multiplication strategy based on RDD partition is proposed for the difficulty of computation and transmission of the large scale matrices, which converts the matrix rows in each partition into block matrices. The row-block matrices are not restricted by the map operator to RDD calculation one by one, and can treat operations at the matrix level by using linear algebraic Library in Spark. The experimental results show that the row-block matrix multiplication strategy effectively improves the efficiency of matrix operations; the parallel block Davidson method can quickly solve the eigenvalues and eigenvectors of the large scale matrices and effectively improve the performance of parallel ISOMAP algorithm; and the parallel ISOMAP algorithm can adapt to dimensionality reduction in the big data environment.
Multi-strategy ant colony algorithm for solving the maximum clique problem based on Spark
GU Junhua, WANG Shoubin, WU Junyan, ZHANG suqi
2019, 49(10): 851-860. doi: 10.3969/j.issn.0253-2778.2019.10.011
Social network analysis has become one of the hotspots in data mining research. Aggregating subgroups are important indicators for measuring the structure of social networks. The maximum clique structure is the most compact condensed subgroup in social networks. The study of the maximum clique problem has also become an important angle of social network analysis. With the development of big data, the massiveness of the nodes and the complexity of the edge structure in the graph put forward a higher requirement for solving the maximum clique problem. Therefore, a multi-strategy ant colony algorithm is proposed for solving the maximum clique problem first, the algorithm uses a multi-conditional selection strategy to expand the search space, increases the diversity of feasible solutions, and avoids falling into the local optimal solution. At the same time, a local search strategy is adopted to improve the accuracy and convergence speed of the algorithm; Then, the algorithm is implemented in parallel on the Spark distributed platform, which verifies the parallelism of the algorithm and improves the efficiency of the algorithm in handling large-scale community networks.