Abstract:Biomedical signal processing plays an important role in life science research, p revention and treatment of diseases and medical instrument industry. Since biome dical signal is detected from human beings, it can be diverse and complicated du e to physiological and psychological reasons. This paper summarizes and classifi es the commonly used biomedical signals, features and the corresponding processi ng approaches. The applications of biomedical signal processing on electrocardio gram (ECG) and electroencephalogram (EEG) signal are introduced. New advances in biomedical signal processing in recent years are also deliberated. Finally, s ome thoughts are provided with respect to the future researches on biomedical si gnal processing.

Abstract:As a new biometric technique, finger vein recognition has attracted lots of attention from research groups and industry at home and abroad, and shows great market potentials. This paper introduces main current researches on finger vein recognition, including finger vein imaging principle and image enhancement techniques, feature extraction methods， finger vein related multi-feature and multi-model fusion methods. Feature extraction methods are described comprehensively, and classified into four categories, i.e., vein pattern feature, texture feature, minutiae feature, and learned feature by machine learning methods. Based on these analyses, we further summary some challenges in finger vein recognition and its applications, including lowering the price of imaging device, improving the quality of image, and decreasing the effect from low quality image, the finger displacement, large scale populations and outdoor image acquisition. And we hope that the challenges can inspire some new ideas in the future.

Abstract:To investigate the discrepancies in brain function of first-episode depressed p atients, five first-episode depressed patients and one healthy volunteer of the same a ge undergo resting state functional magnetic resonance imaging (fMRI) scans by Si emens 3.0T and collecting data are anylized by the amplitude of low frequency f luctuation (ALFF). Then, the ALFF results are possessed by two sample t te s t. The result shows the ALFF value of t he depressed group are decreased in the particular areas, including the left cer ebellar lobe, the left superior temporal gyrus, the bilateral caudate nucleus, t he right lingual gyrus, the left superior frontal gyrus, the right rolandic oper c ulu m, the right anterior cingulate gyri, the right middle frontal gyrus, the right in ferior frontal gyrus, the right supplementary motor area, the left superior parietal gyrus, and the right postcentral gyrus. The research result suggests that the first episode depressed patients have functional discrepancies in the frontal lobe, th e temporal lobe,the cingulate gyrus and the caudate nucleus. These areas are dir ectly correlated with emotion, cognition and memory. Abnormalities in these area s are closely bound up with depression.

Abstract:Heart sound is an importa nt physiological signal, and it contains a large number of physiological and pat hological information. According to the characteristics of heart sound, the symb ol entropy based on probability distribution is proposed. The algorithm makes a breakthrough at linear constraints. On the one hand, it distributes more symbols for the region where the amplitude distribution of the first heart is dense and distributes relatively less symbols for the sparse region, so as to achieve the reduction of redundancy of data; On the other hand, it uses an adaptive method t o determine the size of the symbol set. Then the symbol entropy becomes more sen sitive to the changes of the heart sound signal and can rapidly capture the no nlinear abnormal state of heart signal. Thus the algorithm can make little or no impact of the non-stationary mutation interference and the sequence probabilit y distribution on the entropy. Simulation results show that the algorithm not onl y has significant feasibility and effectiveness but also provides a new way for the rapid diagnosis of heart failure.

Abstract:Automated segmentation of brain MRI image is an important computer based technology with wide applicability i n medicine field, and of great significance in the study of hu man brain diseases. There exists a variety of segmentation methods, such as the the thresholding method, the region growing method, and the clustering method, w hich are broadly applied to natural images. However, those methods are not as p owerful or practical as atlas-based method when applied to clinical medical ima g e. The development of the atlas-based method for brai n image segmentation is reviewed, and representative algorithms are introduc ed. The basic princi ples of these parcellation algorithms are described as well as the components of a state of the art segmentation system. On this basis, the segmentation pro cedures are introduced, and its various applications in clinical medicine are di scussed . Finally, the current status and the futu re potential of the automated segmentation′s applications in clinical medicine are summarized.

Abstract:Automatic detection of markers for fluoroscopic images is the key for fluoroscop y-based navigation system. Due to the small radius of markers and background no ises, the robust and the precision of traditional circle detect ion algorithms are not satisfied. In order to solve this problem, a geom etric-parameter based algorithm, which can extract markers automatically fr om X ray images, is proposed. The method can detect the circles of different ra dius in fluoro scopic images and non standard circles by adjusting the parameter. The e x perimental results show that the proposed algorithm holds high efficiency and robustness.

Abstract:In view of that the traditional genetic selection method selects a large number of redundant genes, which leads to a lower sample fore cast accu racy, a feature gene selection method is put forward based on the signal noise r ation and the neighborhood rough set(SNRS). Firstly, the signal to noise ratio (SNR) i ndex is used to obtain the primary feature subset which have a greater impact on classification. Secondly, the rough neighborhood intensive algorithm is used to o ptimize the primary feature subset. Finally, feature gene subset is classified b y different classifier. Experiment results show that the proposed method can get a higher classification accuracy using less feature gene than the traditiona l ones, which verifies the feasibility and validity of the method.

Abstract:To improve the efficiency and accuracy of medica l image segmentation and provide more fully effective basis for clinical diagnos is and adjunctive therapy, a medical image segmentation method based on line int ercept histogram reciprocal cross entropy is proposed. Firstly, the line interce pt histogram is defined. Then, the line intercept histogram of the medical image is built considering its two-dimensional information. Finally, the reciprocal cross entropy criterion for threshold selection based on the line intercept histo gram is derived, according to which, the medical image is segmented. A large num ber of experimental results show that, compared with other methods, including two-dimensional reciprocal entropy method based on niche chaos particle swarm optimization (NCPSO), two-dim ensional exponent gray entropy method based on decomposition, symmetric cross en tropy method based on two dimensional histogram oblique segmentation, two dimens ional Tsallis cross entropy method based on particle swarm optimization (PSO) an d so on, the proposed method has superior image segmentation performance. In its segmentation result, object region is complete and accurate, and the edge details a re clear and richer. Moreover, the running time is greatly reduced. It is a fast and effective new segmentation method which can be used in medical image research.

Abstract:Breast cancer is one of the most serious diseases greatly threatening human′s health. A patient will not miss the best time for treatment only with early detection and early diagnosis. Mass is the most important and common lesion of breast, so breast mass feature extraction is helpful to improve the efficiency and accuracy of diagnosis. The past algorithms do not consider spatial information of〖JP+2〗 mass images, resulting in low classification accuracy. Aiming at this problem, a new breast mass feature extraction algorithm is proposed based on the edge of neighborhood. It combines the Chan Vese active contour model with the bag of words. The adaptive parameters regulation methods are designed to control edges of mass images. The final representation can be obtained by combining or weighting those features in the neighborhood. Experimental results show that the proposed methods can achieve a better classification accuracy.

Abstract:Accurate classification and recognition of pulmonary nodules are key process of lung cancer computer aided diagnosis (CAD) system. Meanwhile，there are still some scientific and technical challenges, including the difficulty of the feature representation and samples labeled, and the lack of accurate and effective recognition and classification algorithms. A multi classification algorithm is presented combining weakly supervised ECOC algorithm with pulmonary nodules features expression of shape. In order to improve the classification accuracy, we select a series of accurate shape feature description vectors by deliberating the shape features of pulmonary nodules. During the training phase, the coded matrix is constructed by a series of binary classifiers, which are generated by a small amount of labeled pulmonary nodules from experts. Finally, the Humming distance between the code of testing sample and each row of the coded matrix are calculated to determine the category of the testing sample. Experimental results show that the proposed method can obtain more accurate classification results.

Abstract:Continuous loop averaging deconvolution (CLAD) is a recently developed method to restore t he auditory evoked potential (AEP) under high stimulus rate condition. This meth od solves the deconvolution problem in frequency domain for computational effici ency, but suffers from stringent limitation in selecting a stimulus sequence wit h required spectral property. Hereby we propose a new method to solve the decon volution problem in time domain by constructing a linear transform matrix to mod el the convolution process. To understand the AEP distortion caused by the ill posed matrix generated from a bed stimulus sequence, we assess the matrix prope rty using singular value decomposition (SVD) technique and introduce Tikhonov r egularization method to deal with the ill posedness. In the stimulation experim e nt, we compare some typical sequences with different ill posedness conditions a nd restore the transient AEPs under various noise levels. These results justify the proposed approach to the AEP deconvolution with less restriction on the s equence selection.

Abstract:Western scales are a significant basis for assessment of Parkinson′s disease(PD), while these scales contain a large number of cross duplicates scales, which hampers rapid assessment of PD. Therefore, optimizing these wetern scales is significant for rapid diagnosis of PD. And the method of the optimization of Parkinsons scale based on principal component analysis(PCA) is raised. The weighted projective vector is extracted based on principal component analysis, and scale problems are divided on the basis of the projected vector using local recursive segmentation algorithm based on Ostu threshold, Finally, based on contribution factors(CF), a new scale is designed. Experiment results confirm that the new combinations of scale which accounts for 21% of the original western scales is highly comparable to original western scales for identifying PD support vector machine(SVM).

Abstract:RNA Seq(RNA sequencing), based on high throughput sequencing, is a new technique for transcriptome research.Considering the difficulties in the analysis of transcript expression using RNA Seq data, an improved method, improvement of latent dirichlet allocation for sequencing data(LDASeqⅡ) is proposed to calculate the transcript expression.To deal with multi-mappings between reads and isoforms and non-uniform distribution of reads along reference, LDASeqⅡ utilizes the known gene-isoform annotation to constrain the hyperparameters and normalizes the read counts by exon length for each individual exon.By introducing ″pseudo-exon″ and ″pseudo-transcript″, the conjunction reads and noise reads gain proper treatments.LDASeqⅡ is validated using two real datasets on gene and transcript expression calculation and compared with latent dirichlet allocation for sequencing data(LDASeq) and other two popular methods Cufflinks and RNA Seq by expectation maximization(RSEM). The results show that LDASeqⅡ obtains more accurate transcript and gene expression measurements than other approaches.

Abstract:Aiming at the radio frequency(RF) fingerprint identification of the wire less transmitter in low SNR condition, a novel RF fingerprint transform method b ased on nonlinear parameters in power amplifier for wireless transmitters is proposed. Based on nonlinear power amplifier models and wireless channel, Kalman filter is applied to estimate the nonlinear coefficients of the model of amplifier with the prior knowledge of communication frames,and the coefficient vec tors are used as RF fingerprints for the identification of the a ccording transmitters hardware. Theoretical analysis and numerical simulations d emonstrat e that the novel RF fingerprint transform method has the advantage in high accur acy at low signal to noise ratio(SNR). The proposed RF fingerprint transform m ethod c an be used in the physical layer fusion identification of wireless or wire comm unication individuals.

Abstract:In the three dimensional vision system, recognizing and tracking human motion g esture is the crucial step to identify human motion in machine vision field. Due to the complexity of human motion, the existing methods, based on the low quality depth images, canno t provide a high accuracy and a good robust for 3D gesture tracking and recogn ition. Adressing the low quality depth images of the human gesture tracking and recognition, a method is presented based on three step search algorithm. Firstl y, the obtained depth images are analyzed to achieve the human body contour. The n, the settled special skeleton points are tracked based on the depth images, an d the three-step search algorithm is utilized to access the motion estimation a nd get the track motion of the human gesture. Finally, the motion recognition is achieved by using the obtained skeleton point coordinates. Experimental results show that the proposed method is robust to overcome the impact of the illuminat ion, and it also provides improved accurate results of human motion tracking and gesture recognition.

Abstract:The segmentation of foliage organs from 3D point clouds is an elemental work of forestry informatization measurement. However, the foliage point cloud data has a similar color, and the point construction is complex which ca n not be expressed easily. Therefore, a novel feature called local tangent p lane distribution is proposed, and fused with original data, scatter spatial distribution and n ormal distribution to construct a multi-dimension feature, which can character i ze different foliage organs more effectively. Then three kinds of classi fiers, including standard SVM, PSVM, GEPSVM, are used as a compa rison. And then the graph cut is also utilized for a re classification at subsequent process i ng to improve the classification performance. A variety of comparat ive experimental results show that the proposed mutli dimension feature seg mentation method can effectively classify the foliage organs from point cloud d ata. The recognition rate can reach 98%.

Abstract:Since the characteristics that optical microscopic images of gynecological secre tions are very complex, an effective algorithm to detect and collect the number of lactobacillus is proposed. Firstly, the windowed Laws energy is used to ident ify the texture features of the background and other components, thus keeping th e background region with lactobacillus. Secondly, the Laws energy is utilized to pre-segment the lactobacillus, and the average gray of the pre segmentation re gions is calculated. Thirdly, with the combination of the background grey, the a ccurate segment is carried out. On the basis of the imaging characteristics of t he lactobacillus, the values of area, length width ratio and duty ratio of tar get area are extracted to remove impurities. Finally, the statistical analysis i s carried out on the number of lactobacillus. Experimental results demonstrate t hat the proposed method can effectively detect lactobacillus in complex gynecolo gical medical micrographs and offer well-analyzed results.

Abstract:Principal curves are a feature extraction met hod based on the nonlinear transformation. Meanwhile, they are smooth self-consistent curves th at pass through the ″middle″ of the distribution and satisfy the ″self coincidence″. Thus, structural features of t he data can be extracted. Based on the soft K-segments algorithm for principal c urves, the skeletonization extraction of the fingerprint image is not smo oth enough, which often appears small circle and short branches. To solve this proplem, th e soft K -segments algorithm for principal curves and the specialties of fingerprint are analyzed. A new evaluation function is also proposed. And an improved soft K segmen ts algorithm for principal curves is put forward. Compared with those of the original alg o rithms, the smoothness and the accuracy of the proposed algorithm can be illustrated by experiments.

Abstract:The accurate and fast calculation of transcriptome expression level plays an important role in transcriptome research. Based on the previously devised Gamma model for exon array data (GME), a parallel computing method is proposed to improve the computational efficiency of GME on large scale Affymetrix exon chip datasets by taking full advantage of multi-core or cluster computation environment. The princi ples of the GME model and the parallel computing strategy are introduced. The proposed method i s verified using real datasets with various scales. The experimental results show that the propos ed parallel computing approach greatly improves the efficiency of GME model. Thus the GME model is applicable for the analysis on large scale exon array datasets

Abstract:A personalized human body modeling method with high precision bas ed on single Kinect is proposed. Firstly, the high-precision point cloud of a human head usi ng single Kinect is obtained. Then, the point cloud is preprocessed based on maintaining the accuracy of the head. Finally, by employing the hierar chical compactly supported radial basis function (CS RBFs), the sampled point cloud is fitted with the existing human body to get 3D human body model. Simulation results show that the proposed method can enhance accuracy and speed of hu man body modeling.

Abstract:To effectively remove salt and pepper noise in digital images and improve image quality, a new algorithm for removing sa lt an d pepper noise is given based on the analysis of some typical removing noise me t hods. Firstly, according to the characteristics of salt and pepper noise，a nois e detection algorithm, which is based on dynamic window and the neighborhood pix els statistical information, is designed. The noise and the non-noise are effectively distinguished. And then, the noise is r emov ed by using improved adaptive median filter algorithm, in which the adaptive window size and the filtered val ue optimization strategy are introduced. Experimental results show that this me thod can not only remove salt and pepper noise in images, but also effective l y protect the details of image features. The algorithm is better than other methods for the image with high density of noise.

Abstract:Three gases are tested to investigate the effects of data preprocessing algorithm and optimization of sensor array on electronic noses. Preprocessing algorithms are chosen via principal component analysis (PCA), and the relative difference algorithm is determined for preprocessing data of the el ectronic nose for its good classification effect. To optimize the initial array, we first remove sensors abnormally responsing by observing the sensors′ res ponse trend and coefficient of variation. Then we analyze PCA factor loading and conduct multi-collinearity test to determine possible optimal array s using the correlation coefficient and variance inflation factor analysis. Final ly, we apply back propagation(BP) neural network to verify the possible optimal arrays through g as recognition. We determine the final array as well as select other array for controlled study. The results of the check computation certify th at the optimization method of sensor array can not only e liminate anomalies and redundant sensors, but also works well on the classificat ion of test samples.

Abstract:Aiming at the problem of human faces wi th varying expression and illumination, as well as occlusion and disguise, a face recognition algorithm is proposed based on local structural sparse representati on. This algorithm combines low rank matrix recovery with structural incoherenc e and discrete cosine transform (DCT) method to remove occlusion, disguise and il lumination variations in face image. Meanwhile, the par tial inf ormation is fully utilized by using sparse codes of local image patches with spatial layout. In th e classification stage, the algorithm effectively improves the recognition rate based on a novel alignment pooling method. Extensive experime nts are conducted on p ublicly available face databases. Compared with the related state of the art met hods, the experimental results demonstrate the accuracy and efficiency of the pr oposed method.

Abstract:In high frequency (HF) communication, M-ary frequency shift keying (MFSK) is a common modulation mode. The symbol rate estimation of MFSK signal is meaningful for the non-coo peration receiving. According to the analysis of channel distu rbance effect on signals, an algorithm is proposed to estimate the symbol rat e of MFSK signal. By means of extracting and filtering wavelet bridge, clust ering the distance of zero crossing, the effect of multi path and Doppler phenomenon in the symbol rate estimation can be overcome. Simulation experime nts show that the algorithm can reach a good accuracy degree when the channel is in low SNR and contains the multi path and Doppler effect. And it can be used for practical engineering.