Train Support Vector Machine Classifiers Research Articles

Heartbeats classification using QRS and T waves autoregressive features and RR interval features

AbstractThis paper describes a method for the classification of heartbeats based on electrocardiogram QRS complex and T wave autoregressive (AR) features and interbeat distance (RR interval) features. The QRS complex (length n samples) is modelled using AR model of order P. A second AR model of order P1 is applied to the electrocardiogram segment including ST segment and T wave (ST/T segment) with a length of n1 samples. In addition, RR interval features such as the distance between one heartbeat and its preceding (pre‐RR) and following (post‐RR) heartbeats are extracted. The use of AR modelling is motivated by the fact that QRS complex and ST/T segment corresponding to normal beats present different patterns (in shapes and amplitudes) than those corresponding to ventricular ectopic beats. Lengths and AR model orders that minimize the number of misclassifications are determined. Obtained features are used to train and test support vector machine classifier. The proposed method is tested on MIT/BIH Arrhythmia Database and it is compared to state of the art methods. Overall classification accuracy reached 97.02%, whereas it reached 98.86% in the case of subject‐specific scheme. Results are better or comparable to those obtained by state of the art methods.

Building block level urban land-use information retrieval based on Google Street View images

Land-use maps are important references for urban planning and urban studies. Given the heterogeneity of urban land-use types, it is difficult to differentiate different land-use types based on overhead remotely sensed data. Google Street View (GSV) images, which capture the façades of building blocks along streets, could be better used to judge the land-use types of different building blocks based on their façade appearances. Recently developed scene classification algorithms in computer vision community make it possible to categorize different photos semantically based on various image feature descriptors and machine-learning algorithms. Therefore, in this study, we proposed a method to derive detailed land-use information at building block level based on scene classification algorithms and GSV images. Three image feature descriptors (i.e., scale-invariant feature transform-Fisher, histogram of oriented gradients, GIST) were used to represent GSV images of different buildings. Existing land-use maps were used to create training datasets to train support vector machine (SVM) classifiers for categorizing GSV images. The trained SVM classifiers were then applied to case study areas in New York City, Boston, and Houston, to predict the land-use information at building block level. Accuracy assessment results show that the proposed method is suitable for differentiating residential buildings and nonresidential buildings with an accuracy of 85% or so. Since the GSV images are publicly accessible, this proposed method would provide a new way for building block level land-use mapping in future.

Computer aided diagnosis of pulmonary hamartoma from CT scan images using ant colony optimization based feature selection

BackgroundComputer-aided diagnosis (CAD) systems for the detection of lung disorders play an important role in clinical decision making. CAD systems provide a second opinion to the physician in interpreting computed tomography (CT) images. In this work, a CAD system to diagnose pulmonary hamartoma nodules from chest CT images is proposed. MethodsSegmentation of lung parenchyma from CT images is carried out using Otsu’s thresholding method. Nodules are considered to be the region of interests (ROIs) in this work. Texture, shape and run length based features are extracted from the ROIs. Cosine similarity measure (CSM) and rough dependency measure (RDM) are used independently as filter evaluation functions with ant colony optimization (ACO) to select two subsets of features. The selected subsets are used to train two classifiers namely support vector machine (SVM) and Naive Bayes (NB) classifiers using 10-fold cross validation. All the four trained classifiers are tested and the performance measures are estimated. ResultsCT slices of patients affected with pulmonary cancer and hamartoma are used for experimentation. From the lung parenchymal tissues of 300 CT slices, 390 nodules are extracted. The feature selection algorithms, ACO-CSM and ACO-RDM are run for different feature subset sizes. The selected features are used to train SVM and NB classifiers. From the results obtained, it is inferred that SVM classifier with the feature subsets chosen by ACO-RDM feature selection approach yielded a maximum classification accuracy of 94.36% with 38 features. ConclusionFrom the results, it can be clearly inferred that selecting relevant features to train the classifier has a definite impact on the performance of the classifier.

Automatic epileptic seizure detection in EEGs using MF-DFA, SVM based on cloud computing.

Epilepsy is a chronic disease with transient brain dysfunction that results from the sudden abnormal discharge of neurons in the brain. Since electroencephalogram (EEG) is a harmless and noninvasive detection method, it plays an important role in the detection of neurological diseases. However, the process of analyzing EEG to detect neurological diseases is often difficult because the brain electrical signals are random, non-stationary and nonlinear. In order to overcome such difficulty, this study aims to develop a new computer-aided scheme for automatic epileptic seizure detection in EEGs based on multi-fractal detrended fluctuation analysis (MF-DFA) and support vector machine (SVM). New scheme first extracts features from EEG by MF-DFA during the first stage. Then, the scheme applies a genetic algorithm (GA) to calculate parameters used in SVM and classify the training data according to the selected features using SVM. Finally, the trained SVM classifier is exploited to detect neurological diseases. The algorithm utilizes MLlib from library of SPARK and runs on cloud platform. Applying to a public dataset for experiment, the study results show that the new feature extraction method and scheme can detect signals with less features and the accuracy of the classification reached up to 99%. MF-DFA is a promising approach to extract features for analyzing EEG, because of its simple algorithm procedure and less parameters. The features obtained by MF-DFA can represent samples as well as traditional wavelet transform and Lyapunov exponents. GA can always find useful parameters for SVM with enough execution time. The results illustrate that the classification model can achieve comparable accuracy, which means that it is effective in epileptic seizure detection.

An Approach of Modulation Recognition Using 4th Order Cyclic Cumulants Vector of Digital Signal for Cognitive Radio

Modulation recognition is an important issue in cognitive radio research area, however, high recognition precision is usually achieved by relative large number of training data and more various features of digital signal, which call for much more resource. In this paper, a novel modulation recognition approach is proposed, 4th order cyclic cumulants vectors of digital signal is applied for modulation recognition, which are constructed as features to train support vector machine classifiers for further recognition. The experimental result shows the proposed approach can get comparative high precision for ASK, BPSK and PSK signal recognition in additive white Gaussian noise channel while using relative small number of training samples and small in dimension, which reduce cost remarkably.

Epileptic seizure classifications of single-channel scalp EEG data using wavelet-based features and SVM.

In this study, wavelet-based features of single-channel scalp EEGs recorded from subjects with intractable seizure are examined for epileptic seizure classification. The wavelet-based features extracted from scalp EEGs are simply based on detail and approximation coefficients obtained from the discrete wavelet transform. Support vector machine (SVM), one of the most commonly used classifiers, is applied to classify vectors of wavelet-based features of scalp EEGs into either seizure or non-seizure class. In patient-based epileptic seizure classification, a training data set used to train SVM classifiers is composed of wavelet-based features of scalp EEGs corresponding to the first epileptic seizure event. Overall, the excellent performance on patient-dependent epileptic seizure classification is obtained with the average accuracy, sensitivity, and specificity of, respectively, 0.9687, 0.7299, and 0.9813. The vector composed of two wavelet-based features of scalp EEGs provide the best performance on patient-dependent epileptic seizure classification in most cases, i.e., 19 cases out of 24. The wavelet-based features corresponding to the 32-64, 8-16, and 4-8 Hz subbands of scalp EEGs are the mostly used features providing the best performance on patient-dependent classification. Furthermore, the performance on both patient-dependent and patient-independent epileptic seizure classifications are also validated using tenfold cross-validation. From the patient-independent epileptic seizure classification validated using tenfold cross-validation, it is shown that the best classification performance is achieved using the wavelet-based features corresponding to the 64-128 and 4-8 Hz subbands of scalp EEGs.

Enhancing the Biological Relevance of Machine Learning Classifiers for Reverse Vaccinology.

Reverse vaccinology (RV) is a bioinformatics approach that can predict antigens with protective potential from the protein coding genomes of bacterial pathogens for subunit vaccine design. RV has become firmly established following the development of the BEXSERO® vaccine against Neisseria meningitidis serogroup B. RV studies have begun to incorporate machine learning (ML) techniques to distinguish bacterial protective antigens (BPAs) from non-BPAs. This research contributes significantly to the RV field by using permutation analysis to demonstrate that a signal for protective antigens can be curated from published data. Furthermore, the effects of the following on an ML approach to RV were also assessed: nested cross-validation, balancing selection of non-BPAs for subcellular localization, increasing the training data, and incorporating greater numbers of protein annotation tools for feature generation. These enhancements yielded a support vector machine (SVM) classifier that could discriminate BPAs (n = 200) from non-BPAs (n = 200) with an area under the curve (AUC) of 0.787. In addition, hierarchical clustering of BPAs revealed that intracellular BPAs clustered separately from extracellular BPAs. However, no immediate benefit was derived when training SVM classifiers on data sets exclusively containing intra- or extracellular BPAs. In conclusion, this work demonstrates that ML classifiers have great utility in RV approaches and will lead to new subunit vaccines in the future.

Learning to Diagnose Cirrhosis with Liver Capsule Guided Ultrasound Image Classification.

This paper proposes a computer-aided cirrhosis diagnosis system to diagnose cirrhosis based on ultrasound images. We first propose a method to extract a liver capsule on an ultrasound image, then, based on the extracted liver capsule, we fine-tune a deep convolutional neural network (CNN) model to extract features from the image patches cropped around the liver capsules. Finally, a trained support vector machine (SVM) classifier is applied to classify the sample into normal or abnormal cases. Experimental results show that the proposed method can effectively extract the liver capsules and accurately classify the ultrasound images.

Fault Diagnosis in Medium Voltage Drive Based on Combination of Wavelet transform and Support Vector Machine

Nowadays, Medium Voltage Drive (MVD) has been widely applied in the field of high-powered motor speed-regulation. These types of converter use a lot of insulated gate bipolar translators (IGBTs). So it is very important to find an effective way to diagnose IGBT open-circuit faults. This study describes a method of diagnosis for IGBT open-circuit faults in MVD whose topology is cell series of multi-level. This method combines wavelet transform (WT) and support vector machine (SVM). The wavelet transform is used to extract fault features and SVM is used to classify the fault states of a single power unit. Then, the trained SVM classifier is used to scan all power units of MVD sequentially. Results of simulation on the platform of MATLAB/Simulink show that this method has a good diagnosis capability. It can diagnose the IGBT open-circuit faults of the whole inverter system, and diagnosis accuracy is up to 96%. So, this method has a good application prospect.

Fingerprint liveness detection using gradient-based texture features

Fingerprint-based recognition systems have been increasingly deployed in various applications nowadays. However, the recognition systems can be spoofed by using an accurate imitation of a live fingerprint such as an artificially made fingerprint. In this paper, we propose a novel software-based fingerprint liveness detection method which achieves good detection accuracy. We regard the fingerprint liveness detection as a two-class classification problem and construct co-occurrence array from image gradients to extract features. In doing so, the quantization operation is firstly conducted on the images. Then, the horizontal and vertical gradients at each pixel are calculated, and the gradients of large absolute values are truncated into a reduced range. Finally, the second-order and the third-order co-occurrence arrays are constructed from the truncated gradients, and the elements of the co-occurrence arrays are directly used as features. The second-order and the third-order co-occurrence array features are separately utilized to train support vector machine classifiers on two publicly available databases used in Fingerprint Liveness Detection Competition 2009 and 2011. The experimental results have demonstrated that the features extracted with the third-order co-occurrence array achieve better detection accuracy than that with the second-order co-occurrence array and outperform the state-of-the-art methods.

Steganalysis of LSB matching using characteristic function moment of pixel differences

Nowadays, many steganographic tools have been developed, and secret messages can be imperceptibly transmitted through public networks. This paper concentrates on steganalysis against spatial least significant bit (LSB) matching, which is the prototype of many advanced information hiding methods. Many existing algorithms deal with steganalysis problems by using the dependencies between adjacent pixels. From another aspect, this paper calculates the differences among pixel pairs and proves that the histogram of difference values will be smoothed by stego noises. We calculate the difference histogram characteristic function (DHCF) and deduce that the moment of DHCFs (DHCFM) will be diminished after stego bits are hidden in the image. Accordingly, we compute the DHCFMs as the discriminative features. We calibrate the features by decreasing the influence of image content on them and train support vector machine classifiers based on the calibrated features. Experimental results demonstrate that the DHCFMs calculated with nonadjacent pixels are helpful to detect stego messages hidden by LSB matching.

MO-DE-207B-06: Computer-Aided Diagnosis of Breast Ultrasound Images Using Transfer Learning From Deep Convolutional Neural Networks

Purpose: To assess the performance of using transferred features from pre-trained deep convolutional networks (CNNs) in the task of classifying cancer in breast ultrasound images, and to compare this method of transfer learning with previous methods involving human-designed features. Methods: A breast ultrasound dataset consisting of 1125 cases and 2393 regions of interest (ROIs) was used. Each ROI was labeled as cystic, benign, or malignant. Features were extracted from each ROI using pre-trained CNNs and used to train support vector machine (SVM) classifiers in the tasks of distinguishing non-malignant (benign+cystic) vs malignant lesions and benign vs malignant lesions. For a baseline comparison, classifiers were also trained on prior analytically-extracted tumor features. Five-fold cross-validation (by case) was conducted with the area under the receiver operating characteristic curve (AUC) as the performance metric. Results: Classifiers trained on CNN-extracted features were comparable to classifiers trained on human-designed features. In the non-malignant vs malignant task, both the SVM trained on CNN-extracted features and the SVM trained on human-designed features obtained an AUC of 0.90. In the task of determining benign vs malignant, the SVM trained on CNN-extracted features obtained an AUC of 0.88, compared to the AUC of 0.85 obtained by the SVM trained on human-designed features. Conclusion: We obtained strong results using transfer learning to characterize ultrasound breast cancer images. This method allows us to directly classify a small dataset of lesions in a computationally inexpensive fashion without any manual input. Modern deep learning methods in computer vision are contingent on large datasets and vast computational resources, which are often inaccessible for clinical applications. Consequently, we believe transfer learning methods will be important for computer-aided diagnosis schemes in order to utilize advancements in deep learning and computer vision without the associated costs. This work was partially funded by NIH grant U01 CA195564 and the University of Chicago Metcalf program. M.L.G. is a stockholder in R2/Hologic, co-founder and equity holder in Quantitative Insights, and receives royalties from Hologic, GE Medical Systems, MEDIAN Technologies, Riverain Medical, Mitsubishi, and Toshiba. K.D. received royalties from Hologic.

A Classification Approach for Model-Based Fault Diagnosis in Power Generation Systems Based on Solid Oxide Fuel Cells

Solid oxide fuel cells (SOFCs) are a promising option for power generation plants, but the design of fault diagnosis methods remains a key challenge. We propose the use of a quantitative model for such a plant (validated by real experiments) with a support vector machine (SVM) to detect and classify possible faults. The adoption of a classification approach as an identification strategy in a model-based fault diagnosis process represents a major innovation in the field of SOFC plants. Constant–voltage and constant–current control strategies are investigated. In both cases, an adequately trained SVM classifier is used to provide a high probability of correct classification when the plant functions at different steady-state operating conditions for random sizes of the considered faults and for realistic magnitudes of the errors affecting the model predictions. In addition, the relative importance of the easy-to-measure residuals, which are used as features in the SVM classification process, are discussed based on an advanced feature selection technique.

Classification of Normal and Apoptotic Cells from Fluorescence Microscopy Images Using Generalized Polynomial Chaos and Level Set Function.

Accurate automated quantitative analysis of living cells based on fluorescence microscopy images can be very useful for fast evaluation of experimental outcomes and cell culture protocols. In this work, an algorithm is developed for fast differentiation of normal and apoptotic viable Chinese hamster ovary (CHO) cells. For effective segmentation of cell images, a stochastic segmentation algorithm is developed by combining a generalized polynomial chaos expansion with a level set function-based segmentation algorithm. This approach provides a probabilistic description of the segmented cellular regions along the boundary, from which it is possible to calculate morphological changes related to apoptosis, i.e., the curvature and length of a cell's boundary. These features are then used as inputs to a support vector machine (SVM) classifier that is trained to distinguish between normal and apoptotic viable states of CHO cell images. The use of morphological features obtained from the stochastic level set segmentation of cell images in combination with the trained SVM classifier is more efficient in terms of differentiation accuracy as compared with the original deterministic level set method.

A Method for Automatic and Objective Scoring of Bradykinesia Using Orientation Sensors and Classification Algorithms.

Correct assessment of bradykinesia is a key element in the diagnosis and monitoring of Parkinson's disease. Its evaluation is based on a careful assessment of symptoms and it is quantified using rating scales, where the Movement Disorders Society-Sponsored Revision of the Unified Parkinson's Disease Rating Scale (MDS-UPDRS) is the gold standard. Regardless of their importance, the bradykinesia-related items show low agreement between different evaluators. In this study, we design an applicable tool that provides an objective quantification of bradykinesia and that evaluates all characteristics described in the MDS-UPDRS. Twenty-five patients with Parkinson's disease performed three of the five bradykinesia-related items of the MDS-UPDRS. Their movements were assessed by four evaluators and were recorded with a nine degrees-of-freedom sensor. Sensor fusion was employed to obtain a 3-D representation of movements. Based on the resulting signals, a set of features related to the characteristics described in the MDS-UPDRS was defined. Feature selection methods were employed to determine the most important features to quantify bradykinesia. The features selected were used to train support vector machine classifiers to obtain an automatic score of the movements of each patient. The best results were obtained when seven features were included in the classifiers. The classification errors for finger tapping, diadochokinesis and toe tapping were 15-16.5%, 9.3-9.8%, and 18.2-20.2% smaller than the average interrater scoring error, respectively. The introduction of objective scoring in the assessment of bradykinesia might eliminate inconsistencies within evaluators and interrater assessment disagreements and might improve the monitoring of movement disorders.

A FAST AUTOMATED VISION SYSTEM FOR CONTAINER CORNER CASTING RECOGNITION

With the development of container port automation, the automated vision systems for containers have been widely used in automated ports. This paper presents a rapid automated vision system for container corner casting recognition. The histograms of oriented gradients (HOG) descriptors are used to preprocess the image of the container and the vectors of HOG are then built. A trained support vector machine (SVM) classifier is applied to recognize the right corner casting of the container. At last, through symmetry, a flipping mirror algorithm is used for quick left corner casting recognition. The experimental results show this algorithm scans and detect the two corner castings of the container almost twice as fast as the traditional algorithms.

An efficient hybrid eye detection method

Eye detection is the most important and critical task of diverse applications such as face detection and recognition. However, most eye detection methods do not fully consider detection robustness to people with glasses, illumination variation, head pose change, and eye occlusions. This paper proposes an efficient hybrid eye detection method based on a gray intensity variance filter (VF) and support vector machines (SVMs). Firstly, the VF is used for eliminating most of noneye region images to keep less candidate eye regions. Then accurate two eye regions are determined easily through the trained SVM classifier. Moreover, this paper provides an assessment of the sensitivity of obtained parameters in the SVM classifier on eye detection accuracy. The proposed method was evaluated on different face databases. The experimental results show that the method can improve the performance of eye detection and achieve a higher detection accuracy compared with state-of-the-art methods.

基于SVM和决策树的自然图像低能见度天气现象识别

一个地区的能见度不但反映了该地区大气环境的质量，并且与人们的生活有着密切相关的联系。通常，低能见度天气也严重地影响了人们的经济生产，因此其观测具有重要意义。大气环境能见度较低的原因与气象条件有着密切的关联，低能见度的天气现象主要有雨、雪、雾霾、沙尘等。本文提出了一种基于室外单幅自然图像的低能见度天气现象识别算法，该算法通过低能见度天气现象对图像光学信息的影响，提取图像的对比度、饱和度、亮度等特征参数信息进行训练和分类，在训练过程中根据各类别特征之间的距离建立分类决策树，并为决策树构建支持向量机(SVM)分类器，对低能见度天气进行自动分类识别。通过对互联网上的大量低能见度天气光学图像的训练和测试，算法对低能见度的天气现象的平均识别率可达70%。该算法可以为分布式识别提供技术支持，然后采用分布式识别投票，最终可以把识别正确率提高到95%以上。 The visibility in a region not only reflects the quality of the atmospheric environment, but also has close relationship with people’s life. In general, low visibility weather affects people’s economic development, so the real-time observation of low visibility is of much signification. The reason of low visibility is closely associated with meteorological conditions. The low visibility weather phenomena mainly contain rain, snow, fog, etc. This paper proposes a recognition method which is based on low visibility weather phenomenon by means of the influence of low visibility weather phenomenon on the image information such as the image contrast, saturation and brightness that can be employed for training and classification. We establish a classification decision tree according to the distance between the different categories in the process of training and building support vector machine (SVM) classifier for the decision tree. It can classify the low visibility weather image automatically and intelligently. Through testing a huge amount of images downloaded from the internet, the experimental results show that weather image mean recognition rate is over 70%. After adopting the voting scheme via distributed recognition, the final low visibility weather recognition rate is more than 95%.

SVM and ANN Based Classification of Plant Diseases Using Feature Reduction Technique

Computers have been used for mechanization and automation in different applications of agriculture/horticulture. The critical decision on the agricultural yield and plant protection is done with the development of expert system (decision support system) using computer vision techniques. One of the areas considered in the present work is the processing of images of plant diseases affecting agriculture/horticulture crops. The first symptoms of plant disease have to be correctly detected, identified, and quantified in the initial stages. The color and texture features have been used in order to work with the sample images of plant diseases. Algorithms for extraction of color and texture features have been developed, which are in turn used to train support vector machine (SVM) and artificial neural network (ANN) classifiers. The study has presented a reduced feature set based approach for recognition and classification of images of plant diseases. The results reveal that SVM classifier is more suitable for identification and classification of plant diseases affecting agriculture/horticulture crops.

A novel analysis of color component for top dyed melange yarn with support vector machine

AbstractIn view of the difficulty of distinguishing the color component in top dyed melange yarn due to the spectral overlap of the component colors, a novel color component analysis method based on support vector machine is presented. With this method, spectra data can be distinguished more accurately and effectively than with the traditional method—human‐eye detection—and therefore, the method will be very helpful for accurate color matching. In our work, the core idea was to convert the overlapped spectra data into linearly separable ones in a high dimension space, followed by recognition and determination of the composition of melange yarn by trained support vector machine classifier. The effects of four kernel functions, i.e., linear, radial basis kernel, sigmoid, and polynomial, as well as five spectral preprocessing methods, including amplification, first derivative, second derivative, principal components analysis, and L*a*b* values were studied. The results demonstrated that with the amplification factor of 100 of reflectance spectra coupled with L*a*b* as input data, and using radial basis kernel as kernel function, the highest recognition rate was achieved, with an average recognition rate of eight colors of 96.5%, indicating that it was a better color component analysis method for top dyed melange yarn. © 2015 Wiley Periodicals, Inc. Col Res Appl, 41, 636–641, 2016