Table of Content

31 May 2019, Volume 37 Issue 3

Previous Issue    Next Issue

Signal and Information Processing

Small Harbor Detection Based on PLSA and BoW in High Resolution Remotely Sensed Imagery

BI Qi, TONG Xin, ZHANG Jiyong, XU Kai, ZHANG Han, QIN Kun

2019, 37(3):  301-312.  doi:10.3969/j.issn.0255-8297.2019.03.001

Asbtract ( 956 )   PDF (9372KB) ( 225 )  

References | Related Articles | Metrics

Remotely sensed high resolution imaging is an effective way to monitor small harbors along coastlines. Harbors vary a lot in features and are usually difficult to describe. This paper studies a harbor detection method based on probabilistic latent semantic analysis (PLSA) and bag of words (BoW). Firstly, coastlines are extracted to shrink searching areas. Then grey level histogram, normalized differential water index and fractal dimension features are introduced to PLSA model to generate feature description sets. Eigenvectors of speeded up robust features are introduced to BoW model to generate visual dictionary. Finally, after collecting samples of small harbors, support vector machine (SVM) classifiers are trained based on the above features. Based on the trained classifier, small harbor detection is implemented based on 22 images. Experiments show that the proposed method reaches the best accuracy with relatively low time cost than single-feature models and single PLSA or BoW model.

Fine-Grained Fingerprint-Based Indoor Localization System

TIAN Xiyan, DU Liufeng

2019, 37(3):  313-326.  doi:10.3969/j.issn.0255-8297.2019.03.002

Asbtract ( 843 )   PDF (5976KB) ( 87 )  

References | Related Articles | Metrics

To improve the accuracy and stability of the indoor localization system, this paper proposes a fingerprinting positioning technique based on Wi-Fi channel state information (CSI). The CSI characterizes fine-grained channel information through subcarriers, which can better eliminate interferences such as multipath effect. Based on the CSI, an efficient and accurate construction scheme for fingerprints database is achieved with the matrix completion theory, and a stable and real-time fingerprint matching method is implemented with the Bayesian rule. The field evaluation results in two scenarios show that the proposal achieves the desired goals in terms of both fingerprints database construction accuracy and localization performance.

Protein Small Molecule Affinity Prediction Based on Natural Language Processing

OUYANG Zhiyou, CHEN Chen, WANG Yuqian, CHEN Jingang, YIN Zhao, ZHOU Qingsong

2019, 37(3):  327-335.  doi:10.3969/j.issn.0255-8297.2019.03.003

Asbtract ( 2414 )   PDF (1839KB) ( 177 )  

References | Related Articles | Metrics

The interaction between proteins and small molecules plays a very important role in drug research and development. However, the existing methods for predicting the affinity of small molecules have some problems, such as high cost and low accuracy. In this paper, a new protein small molecule affinity prediction method is proposed based on natural language processing (NLP) technology, which using NLP to analysis the protein structure data and small molecule fingerprint data, as well as using gradient boosting decision tree (GBDT) model to predict the affinity. Experiments show that the proposed method has performance over the exiting methods in terms of accuracy.

Three-Dimensional Photoacoustic Image Reconstruction Using Weighted Alternating Direction Method

QI Mengyu, ZHAO Lili, LIU Xin, YAN Zhuangzhi

2019, 37(3):  336-348.  doi:10.3969/j.issn.0255-8297.2019.03.004

Asbtract ( 805 )   PDF (7859KB) ( 86 )  

References | Related Articles | Metrics

Based on conventional alternating direction method (ADM), an advanced method called weighted ADM is proposed for three-dimensional photoacoustic reconstructions, to obtain better images with fewer measurements. Take advantage of structural information of targets as priori information, the iteration process of ADM is improved and optimized, and the reconstructed images were compared with the sparse equations and least squares methods (LSQR) and conventional ADM method. Simulation analysis showed that the proposed method is able to provide photoacoustic images with better accuracy and better spatial resolution in the circumstance of under-sampling, compared with the two other methods.

Diameter Control Technology of Ultra-fine Sutures Based on DMA and Improved GPC Algorithm

WANG Deming, SUI Xiuwu, ZHANG Yang, WAN Kaixin, SHI Feng

2019, 37(3):  349-358.  doi:10.3969/j.issn.0255-8297.2019.03.005

Asbtract ( 886 )   PDF (3521KB) ( 103 )  

References | Related Articles | Metrics

In order to solve the problem of suture diameter inhomogeneity caused by slow and imprecise wire diameter detection and uncontrollable forming process, a control system is designed by a rapid measurement and accurate control method. The system, based on the embedded system STM32, consists of a linearly coupled charge-coupled device (CCD) image sensor, 14-bit CCD signal processor AD9822, direct memory access (DMA), and improved adaptive generalized predictive control (GPC) algorithm. The suture samples satisfying the requirements of the 32th United States Pharmacopoeia Suture standard have been produced on the JK1601 spinning machine by the control system, and the uniformity of the wire diameter has increased from 74.6% to 95%.

Automatic Extraction of Regions of Peripheral Nerve Internal Fascicular Groups from MicroCT Images

LI Fang, ZHONG Yingchun, QI Jian, LUO Peng

2019, 37(3):  359-368.  doi:10.3969/j.issn.0255-8297.2019.03.006

Asbtract ( 661 )   PDF (8310KB) ( 105 )  

References | Related Articles | Metrics

A novel algorithm to extract the region of peripheral nerve internal fascicular groups automatically from MicroCT images is proposed. 1) the contour and non-contour of fascicular groups in the first MicroCT image were marked manually. 2) the sparse auto encoder (SAE) was employed to extract the features unsupervisedly and the corresponding neural network was trained in order to recognize the contour in the second image. 3) the contour and non-contour of fascicular groups in the second MicroCT image were used as the labeled samples to train a new neural network. 4) above process was repeated until all images were processed completely. The experimental results show that the accuracy of the algorithm in this paper reaches 84.7% of manual's and it can process 522 images only in 0.3~0.4 h. Additionally, not only the single groups but also the splitting and emerging groups can be extracted rapidly and accurately.

A Method of Low Frequency Noise Measurement for Operational Amplifier

CHEN Xiaojuan, ZHANG Xinchao, JANG Shan, KANG Aimin

2019, 37(3):  369-377.  doi:10.3969/j.issn.0255-8297.2019.03.007

Asbtract ( 1719 )   PDF (4372KB) ( 294 )  

References | Related Articles | Metrics

This paper presents a new type of integrated operational amplifier low-frequency noise measurement system built by discrete components. First, the noise model of the operational amplifier is established. Then sub-circuit modules of the test system are designed and simulated separately. Finally, the low-frequency noise measurement and data acquisition of the integrated operational amplifier are performed. The experimental results show that this system can realize the measurement of low frequency noise and power spectral density of integrated operational amplifiers in the frequency range of 10⁰~10⁵ kHz. The measured input equivalent noise value is basically consistent with that presented in the product data-sheet, with the measurement error is 0.039.

Chinese FastText Short Text Classification Method Integrating TF-IDF and LDA

FENG Yong, QU Bohao, XU Hongyan, WANG Rongbing, ZHANG Yonggang

2019, 37(3):  378-388.  doi:10.3969/j.issn.0255-8297.2019.03.008

Asbtract ( 2003 )   PDF (2762KB) ( 428 )  

References | Related Articles | Metrics

FastText text classification model has the advantages of high speed and high efficiency, but its application in Chinese short text classification has the problem of low precision. To solve this problem, a Chinese FastText short text classification method integrating TF-IDF and LDA is proposed. In the input phase of FastText text classification model, the dictionaries generated after n-gram processing are filtered by TF-IDF, and corpus thematic analysis is conducted by LDA model, then the feature dictionary is supplemented according to the obtained results. Thus, the highly differentiated entries are biased in the process of computing the mean value of input word sequence vectors, making them more suitable for Chinese short text classification environment. The experimental results show that the proposed method has higher precision in Chinese short text classification.

CFMoment: Closed Frequent Itemsets Mining Based on Data Stream

WANG Jingwei, WU Shaohua, QU Zhiguo

2019, 37(3):  389-397.  doi:10.3969/j.issn.0255-8297.2019.03.009

Asbtract ( 977 )   PDF (1435KB) ( 98 )  

References | Related Articles | Metrics

Mining closed frequent itemsets over stream data is an important research issue of mining association rules in data mining. In this paper, we propose an efficient closed frequent itemsets mining algorithm in stream data, CFMoment, to maintain the set of closed frequent itemsets in data streams with a sliding window. The new algorithm can be applied to many stream data processing applications with high real-time requirements. It proposes to reduce the time and memory requirements in sliding windows by using the effective bit-sequence representation of items, which further improves the efficiency of closed frequent itemsets in stream data mining and effectively reduces the memory requirements in running process. Experiments show that the proposed algorithm not only attains highly accurate mining results, but also runs significantly faster and consumes less memory than the existing algorithm Moment.

A Method for Color Image Quality Assessment

CAO Xin, LI Zhanming, HU Wenjin

2019, 37(3):  398-406.  doi:10.3969/j.issn.0255-8297.2019.03.010

Asbtract ( 902 )   PDF (2949KB) ( 185 )  

References | Related Articles | Metrics

Human visual system is not only highly sensitive to the structural information of the image, but also closely related to the color information. Image quality assessment methods based on structural similarity are mostly implemented without considering color. Aiming at this problem, a new image quality assessment method is proposed. The proposed method first extracts the value, hue and saturation of the color image according to the characteristics of the human visual system, and convolves the value component with the Scharr operator to extract the image value channel edge feature to obtain the edge feature of the intense-changing part of brightness, and simultaneously the hue and saturation are treated as the color feature. Secondly, the method extracts the edge feature of the grayedout image to obtain the edge feature of the slow-changing part of the brightness, and finally fuses the above features to obtain more complete image features so as to establish a color image quality assessment model. A large number of comparative experiments were performed on the LIVE database. The results show that the assessment results of the algorithm are generally more consistent with the subjective assessment results, compared with other widely used image quality assessment algorithms.

User Transportation Mode Recommendation Based on Contextual Factor and User Preference

WANG Feng, QU Junfeng, ZHAO Yongbiao, GU Qiong

2019, 37(3):  407-418.  doi:10.3969/j.issn.0255-8297.2019.03.011

Asbtract ( 731 )   PDF (2654KB) ( 123 )  

References | Related Articles | Metrics

In order to recommend satisfactory transportation mode, context of user transportation is firstly built, and then context factors are extracted from data of user behavior which is generated by social application based on Mobile Internet. According to the travel purposiveness and preference of users, context factors and user preference are taken into comprehensive consideration in this paper. In the view of context factors, there is a great deal of difference among users' selections of transportation mode due to different travel conditions and their preference. Experiment results show that the proposed scheme can present the selection percentage of single transportation mode and composite transportation mode under different circumstances of context factors. It is also proven that the recommended travel plans are closely aligned with users' needs in terms of user satisfaction. The proposed method shows superior performance on the T-test of different datasets, such as NDCG@5, NDCG@10, and MAP, to the collaborative filtering method based on Pearson correlation coefficients and the weighted slot-one algorithm.

Copy-Move Forgery Detection Based on Super Pixel Segmentation

LIU Jiarui, LU Wei, LIU Ke, HUANG Xinchao, LIN Cong, LIU Xianjin

2019, 37(3):  419-426.  doi:10.3969/j.issn.0255-8297.2019.03.012

Asbtract ( 846 )   PDF (3216KB) ( 110 )  

References | Related Articles | Metrics

A clustering method to detect the copy-move area based on the results of superpixel segmentation is proposed. Different from the traditional K-means clustering method, the proposed clustering method is to place random seeds in the image and segment the region by using super-pixel segmentation method, and then obtain the regions containing pre-matched feature points. In this algorithm, the number of matched feature points in each region is used as a criterion to determine whether the matched feature points in the region are effective feature points. When the number of matching points reaches a certain threshold, the points in the sub-regions are clustered into one group, so that the clustering results are closer to the image content and in accordance with the actual situation. Experiments show that the proposed method based on super-pixel segmentation is more effective than the traditional SIFT method.

Feature Selection of Human Activity Recognition Based on Tri-axial Accelerometer

FAN Shurui, JIA Yating, LIU Jinghua

2019, 37(3):  427-436.  doi:10.3969/j.issn.0255-8297.2019.03.013

Asbtract ( 1543 )   PDF (2048KB) ( 298 )  

References | Related Articles | Metrics

In order to solve the problem of activity classification in human motion pattern recognition, the feature selection of human activity recognition is studied in detail. By taking signal statistics on the x-axis, y-axis and z-axis, 117 features such as standard deviation, skewness and kurtosis are obtained. The three algorithms of Fisher score, Relief-F and Chi square are combined with the hierarchical classification method to select the feature subset, and the action classification is conducted by using the support vector machine (SVM). Experiments showed that the feature set selected by the three feature selection algorithms helps to identify three static movements of standing, sitting and lying and three dynamic movements of walking, going upstairs and downstairs with high precision, which is conducive to the subsequent research on low complexity online identification method.