ISSN 0253-2778

CN 34-1054/N

2020 Vol. 50, No. 1

Display Method:
Research Article
Two-stage grasping detection for robots based on RGB images
XIONG Junlin, ZHAO Duo
2020, 50(1): 1-10. doi: 10.3969/j.issn.0253-2778.2020.01.001
Recently, robots have played big roles in more and more cases. An accurate grasp detection is a key component of a robot working process. An end-to-end method for robotic grasp detection in an RGB image containing objects is proposed in such a case, which takes the whole picture as input and gives the prediction result directly without using traditional sliding windows or region extraction. Obviously, different grasp points lead to different grasp orientations. The grasp detection method takes two steps. First, a convolutional neural network is trained to predict the positions of grasp points. Next, a square area with the preceding grasp point as the center is taken from the image, where the edges are extracted using the Canny edge detection and the lines are detected using Hough Transform. A principal-directiondetection algorithm is proposed to analyze these lines and detect grasp orientations and the distance between two parallel fingers. The method gives a better grasp detection and has an influence on computer vision using both deep learning and traditional algorithms.
Research characteristics of building interior lighting from 1998 to 2018 in English literatures —— Visualization analysis based on CITESPACE
WANG Lixiong, MA Xiufeng
2020, 50(1): 11-17. doi: 10.3969/j.issn.0253-2778.2020.01.002
With the increases in our requirements for the quality of the indoor lighting environment, the research in the field of indoor lighting in global architecture is progressing and developing. By using CITESPACE, analyses 733 articles of SCI journals on indoor lighting (from 1998 to 2018), including volume analysis, national or regional evolution analysis, distribution analysis of research institutions, citation journals analysis, keyword co-occurrence analysis, etc. The results show that indoor lighting research has experienced three stages: the starting stage in 1998, the fast and stable development stage and the regression development stage; there are many research institutions in Asia, Europe and North America, and their achievements have great influence; journals such as Energy and Buildings, Build and Environment and Solar Energy have been cited more frequently and have more academic authority; there are strong correlations among research topics, and the research framework is relatively complete; The theory of energy conservation has been held all through the researches, emphasizing the concept of sustainable development; functional lighting gradually turns to the field of behavioral and health lighting; genetic algorithms combined with new lighting research methods has become a research hotspot at the current stage.
A novel mapping between machine learning and phase transition
LONG Fei, HUANG Kun, LI Feng
2020, 50(1): 18-28. doi: 10.3969/j.issn.0253-2778.2020.01.003
Phase transition is an important physical concept which represents the transformation from one thermodynamic state to another. It has a wide application in physics such as condensed matter physics, particle physics and astrophysics. Machine learning is the study of computer algorithms that improve automatically through experience. It is also a very active research field with the success of deep learning in recent years. This paper visually gives the evidence that the training process of supervised learning is quite similar to phase transition. Accordingly, it is possible to identify the central concepts of phase transition theory, such as symmetry breaking or critical point in machine learning. The Ising model and an autoencoder of the MNIST dataset are used as two examples to explain the “phase transition” in the training process of machine learning. This novel mapping between machine learning and phase transition brings a new method to understand machine learning and makes it possible to solve machine learning problems from a physical perspective.
Service identification of WeChat traffic based on fuzziness and semi-supervised self-paced co-training
LIU Weikang, QIN Xiaowei, WEI Guo
2020, 50(1): 29-38. doi: 10.3969/j.issn.0253-2778.2020.01.004
Accurate service identification of network data streams is a prerequisite for providing differentiated services. The commonly used supervised learning is difficult to implement when constructing training data sets due to the need for a large number of human annotations. Semi-supervised learning based on a small amount of annotated data has become one of the research hotspots. Semi-supervised framework of Self-paced Co-training adopts the method of collaboration that processes the easier pieces first using multiple perspectives when dealing with unlabeled data. However, this method only uses confidence as the criterion to select pseudo labels for samples, which can easily lead to the gradual decline of multi-perspective differences in the training process, resulting in the decline of synergy gain and the limitation of model performance. Therefore, for the recognition of WeChat data streams, a self-paced co-training model based on fuzziness (FBSpaCo) is proposed. When labeling pseudo labels, the fuzziness evaluation mechanism is introduced. Experiments show that the model can effectively avoid the decline of the difference between two perspectives in the training process. Compared with the existing methods, the recognition accuracy is greatly improved.
A trip fault identification method of distribution network based on image processing and deep learning
DU Zhaoxin, XIE Haining, SONG Jie, ZHOU Desheng, ZOU Xiaofeng, CHEN Ran, ZENG Ping
2020, 50(1): 39-48. doi: 10.3969/j.issn.0253-2778.2020.01.005
The currently trip fault identification method for distribution networks is mainly based on manual discrimination, which causes the problem of larger workload and lower accuracy. By combining image processing and deep learning technology, automatic identification of distribution network fault types can be realized. First, the telemetry current waveform is converted to a current value group with time stamps by image processing technology. Then feature vectors are constructed with telecommunicating signal, telecontrol signal and normalized current value. Based on the fault type criterion, the deep neural network model is built and trained to realize fault type identification. Finally, the model is optimized by adjusting the number of hidden layers and neurons in the deep neural network. The experimental results show that the types of trip fault can be identified quickly with the proposed method, that and the accuracy is better than existing methods, which means the new method is practical and effective.
The effect of haps unstable movement on cellular handover performance
WANG Xiaopeng, ZHOU Jiaxi, LI Letian, ZHOU Wuyang
2020, 50(1): 49-56. doi: 10.3969/j.issn.0253-2778.2020.01.06
The high-altitude platform station (HAPS) communication system has its unique advantages such as flexible deployment and large coverage. It can be widely used in a variety of communication scenarios in the future and has attracted the attention of researchers from various countries. Under the influence of stratospheric wind, HAPS will inevitably move within a certain range. In this paper, two modes of movement affected by path loss are discussed: vertical movement and swing movement. First, a more realistic ground coverage model is established and its calculation formulas are derived. Then, based on this coverage model, the handover probability of these two movement modes is calculated and analyzed, and is compared with the handover probability estimated using the existing coverage models. The results verified the validity of the proposed model.
Link prediction in complex networks based on mutual information
QI Fangpeng, WANG Tong, FU Zhongqian
2020, 50(1): 57-63. doi: 10.3969/j.issn.0253-2778.2020.01.007
A new perspective of dealing with link prediction problem was derived due to the application of mutual information in complex networks. Traditional mutual information algorithm (MI) not only considers the neighbor information of nodes, but also the structural information of common neighbors. Although MI has better performance compared with traditional methods which are based on common neighbors, it doesn’t effectively differentiate between different common neighbors. A new algorithm (MMI) was proposed by considering the influence of different common neighbors, which performs better than MI in precision.
Three quantum ranging and positioning schemes to reduce atmospheric interferences
CONG Shuang, WU Wenshen, SHANG Weiwei, CHEN Ding
2020, 50(1): 64-71. doi: 10.3969/j.issn.0253-2778.2020.01.008
Three quantum ranging positioning schemes against atmospheric interference are proposed based on the analysis of the principle and process of quantum ranging positioning schemes, considering the effects of distance error caused by entangled light passing through the atmospheric ionosphere and troposphere on the accuracy of the system, the relationship among the propagation distance error of the entangled light in the ionosphere, the free electron density in the ionosphere and the frequency of the entangled light, as well as the relationship between the propagation distance error of the entangled light in the troposphere and factors such as pressure, temperature and other factors. Expressions of the three quantum ranging positioning schemes to reduce the ranging error caused by atmospheric interference are derived by performing theoretical analyses.Numerical simulation examples are given to demonstrate that the dual-frequency correction scheme based on three satellites plus one ground station minimizes errors in ranging the atmosphere.
Sybil attack detection scheme based on AOA in heterogeneous wireless sensor networks
ZHANG Shuguang, WANG Qian, WANG Hao, ZHONG Juan
2020, 50(1): 72-78. doi: 10.3969/j.issn.0253-2778.2020.01.009
In the wireless sensor network, sybil attack nodes launch attacks by forging multiple identities. If different sybil identities send messages with different transmitting powers, the sybil attack behavior will be difficult to detect. To solve this problem, this paper proposes a sybil attack detection scheme based on the angle of arrival (AOA) in a heterogeneous network environment. In this scheme, heterogeneous nodes detect the angle of arrival from surrounding nodes, and use the angle information to establish a list of suspicious sybil nodes. Through the information interaction between neighboring heterogeneous nodes, the sybil attack node can be located cooperatively. For the special case of heterogeneous nodes, a single heterogeneous node enhanced detection mechanism is proposed to detect sybil nodes. Through theoretical analysis and simulation experiments, the scheme can quickly and accurately identify the malicious nodes, reduce the energy consumption of the nodes, and extend network life.
A self-learning framework for pedestrian detection
WANG Zhong, SHI Peibei, LIU Guiquan
2020, 50(1): 79-85. doi: 10.3969/j.issn.0253-2778.2020.01.010
The performance of offline trained pedestrian detectors significantly drops when they are applied to the specific scene. Although manual labeling can improve detection performance, it requires a lot of human effort. In this paper, a self-learning framework is proposed for pedestrian detection, which can adapt any offline trained detector to a specific scene and obtain a better performance. Firstly, Cascade classifier is used as an offline classifier, while a Gaussian Mixture Model (GMM) is trained using a set of public pedestrian photos. Next, a low threshold offline classifier is used to perform pedestrian detection on a specific scene and the confidence score of candidate detections is obtained. Then, samples with high confidence scores are selected as positive samples, while those with low confidence scores are taken as a negative samples, and GMM is used to represent the candidate detection again. Finally, a discriminative pedestrian classifier is trained online using the SVM classifier to re-estimate candidate objects. Experimental results on public and self-made datasets show that the proposed method can improve the accuracy of the generic pedestrian detector and significantly outperforms the traditional methods.