A decision based one-against-one method for multi-class support vector machine

Semi-Supervised Classification of PolSAR Images Based on Co-Training of CNN and SVM with Limited Labeled Samples

Recently, convolutional neural networks (CNNs) have shown significant advantages in the tasks of image classification; however, these usually require a large number of labeled samples for training. In practice, it is difficult and costly to obtain sufficient labeled samples of polarimetric synthetic aperture radar (PolSAR) images. To address this problem, we propose a novel semi-supervised classification method for PolSAR images in this paper, using the co-training of CNN and a support vector machine (SVM). In our co-training method, an eight-layer CNN with residual network (ResNet) architecture is designed as the primary classifier, and an SVM is used as the auxiliary classifier. In particular, the SVM is used to enhance the performance of our algorithm in the case of limited labeled samples. In our method, more and more pseudo-labeled samples are iteratively yielded for training through a two-stage co-training of CNN and SVM, which gradually improves the performance of the two classifiers. The trained CNN is employed as the final classifier due to its strong classification capability with enough samples. We carried out experiments on two C-band airborne PolSAR images acquired by the AIRSAR systems and an L-band spaceborne PolSAR image acquired by the GaoFen-3 system. The experimental results demonstrate that the proposed method can effectively integrate the complementary advantages of SVM and CNN, providing overall classification accuracy of more than 97%, 96% and 93% with limited labeled samples (10 samples per class) for the above three images, respectively, which is superior to the state-of-the-art semi-supervised methods for PolSAR image classification.

An SVM-Based Neural Adaptive Variable Structure Observer for Fault Diagnosis and Fault-Tolerant Control of a Robot Manipulator

A robot manipulator is a multi-degree-of-freedom and nonlinear system that is used in various applications, including the medical area and automotive industries. Uncertain conditions in which a robot manipulator operates, as well as its nonlinearities, represent challenges for fault diagnosis and fault-tolerant control (FDC) that are addressed through the proposed FDC technique. A machine-learning-based neural adaptive, high-order, variable structure observer for fault diagnosis (FD) and adaptive, modern, fuzzy, backstepping, variable structure control for use in a fault-tolerant control (FC) algorithm, are proposed in this paper. In the first stage, a variable structure observer is proposed as an FD technique for the robot manipulator. The chattering phenomenon associated with the variable structure observer(VSO) is solved using a high-order variable structure observer. Then, the dynamic behavior estimation performance in the high-order variable structure observer is improved by incorporating a neural network algorithm in the FD pipeline. This adaptive technique is also effective in improving the robustness of the fault signal estimation. Moreover, support vector machines (SVMs) that can derive adaptive threshold values are used to categorize faults. To design an effective fault-tolerant controller (FC), an adaptive modern fuzzy backstepping variable structure controller is used in this study. First, a new variable structure controller is designed. Next, to increase robustness and reduce high-frequency oscillations in uncertain conditions, a backstepping algorithm is used in parallel with the variable structure controller to design the backstepping variable structure controller. To design an effective hybrid controller, a fuzzy algorithm is integrated into the backstepping variable structure controller to create a fuzzy backstepping variable structure controller. Then, to improve the robustness and reliability of the FC, a neural adaptive. high-order. variable structure observer is applied to the fuzzy backstepping variable structure controller to design a modern fuzzy backstepping variable structure controller. An adaptive algorithm is used to fine-tune the variable structure coefficients and reduce the effect of faults on the robot manipulator. The effectiveness of the selected algorithm is validated using a PUMA robot manipulator. The neural adaptive. high-order variable structure observer improves the average performance for the identification of various faults by about 27% and 29.2%, compared with the neural high-order variable structure observer and variable structure observer, respectively.

Improved linear classifier model with Nyström

Most data sets consist of interlaced-distributed samples from multiple classes and since these samples always cannot be classified correctly by a linear hyperplane, so we name them nonlinearly separable data sets and corresponding classifiers are named nonlinear classifiers. Traditional nonlinear classifiers adopt kernel functions to generate kernel matrices and then get optimal classifier parameters with the solution of these matrices. But computing and storing kernel matrices brings high computational and space complexities. Since INMKMHKS adopts Nyström approximation technique and NysCK changes nonlinearly separable data to linearly ones so as to reduce the complexities, we combines ideas of them to develop an improved NysCK (INysCK). Moreover, we extend INysCK into multi-view applications and propose multi-view INysCK (MINysCK). Related experiments validate the effectiveness of them in terms of accuracy, convergence, Rademacher complexity, etc.

A fingertip force prediction model for grasp patterns characterised from the chaotic behaviour of EEG

A stable grasp is attained through appropriate hand preshaping and precise fingertip forces. Here, we have proposed a method to decode grasp patterns from motor imagery and subsequent fingertip force estimation model with a slippage avoidance strategy. We have developed a feature-based classification of electroencephalography (EEG) associated with imagination of the grasping postures. Chaotic behaviour of EEG for different grasping patterns has been utilised to capture the dynamics of associated motor activities. We have computed correlation dimension (CD) as the feature and classified with "one against one" multiclass support vector machine (SVM) to discriminate between different grasping patterns. The result of the analysis showed varying classification accuracies at different subband levels. Broad categories of grasping patterns, namely, power grasp and precision grasp, were classified at a 96.0% accuracy rate in the alpha subband. Furthermore, power grasp subtypes were classified with an accuracy of 97.2% in the upper beta subband, whereas precision grasp subtypes showed relatively lower 75.0% accuracy in the alpha subband. Following assessment of fingertip force distributions while grasping, a nonlinear autoregressive (NAR) model with proper prediction of fingertip forces was proposed for each grasp pattern. A slippage detection strategy has been incorporated with automatic recalibration of the regripping force. Intention of each grasp pattern associated with corresponding fingertip force model was virtualised in this work. This integrated system can be utilised as the control strategy for prosthetic hand in the future. The model to virtualise motor imagery based fingertip force prediction with inherent slippage correction for different grasp types ᅟ.

A Fast SVM-Based Tongue's Colour Classification Aided by k-Means Clustering Identifiers and Colour Attributes as Computer-Assisted Tool for Tongue Diagnosis

In tongue diagnosis, colour information of tongue body has kept valuable information regarding the state of disease and its correlation with the internal organs. Qualitatively, practitioners may have difficulty in their judgement due to the instable lighting condition and naked eye's ability to capture the exact colour distribution on the tongue especially the tongue with multicolour substance. To overcome this ambiguity, this paper presents a two-stage tongue's multicolour classification based on a support vector machine (SVM) whose support vectors are reduced by our proposed k-means clustering identifiers and red colour range for precise tongue colour diagnosis. In the first stage, k-means clustering is used to cluster a tongue image into four clusters of image background (black), deep red region, red/light red region, and transitional region. In the second-stage classification, red/light red tongue images are further classified into red tongue or light red tongue based on the red colour range derived in our work. Overall, true rate classification accuracy of the proposed two-stage classification to diagnose red, light red, and deep red tongue colours is 94%. The number of support vectors in SVM is improved by 41.2%, and the execution time for one image is recorded as 48 seconds.

In silico prediction of chemical acute oral toxicity using multi-classification methods

Chemical acute oral toxicity is an important end point in drug design and environmental risk assessment. However, it is difficult to determine by experiments, and in silico methods are hence developed as an alternative. In this study, a comprehensive data set containing 12, 204 diverse compounds with median lethal dose (LD₅₀) was compiled. These chemicals were classified into four categories, namely categories I, II, III and IV, based on the criterion of the U.S. Environmental Protection Agency (EPA). Then several multiclassification models were developed using five machine learning methods, including support vector machine (SVM), C4.5 decision tree (C4.5), random forest (RF), κ-nearest neighbor (kNN), and naïve Bayes (NB) algorithms, along with MACCS and FP4 fingerprints. One-against-one (OAO) and binary tree (BT) strategies were employed for SVM multiclassification. Performances were measured by two external validation sets containing 1678 and 375 chemicals, separately. The overall accuracy of the MACCS-SVM(OAO) model was 83.0% and 89.9% for external validation sets I and II, respectively, which showed reliable predictive accuracy for each class. In addition, some representative substructures responsible for acute oral toxicity were identified using information gain and substructure frequency analysis methods, which might be very helpful for further study to avoid the toxicity.

In-home hierarchical posture classification with a time-of-flight 3D sensor

A non-invasive technique for posture classification suitable to be used in several in-home scenarios is proposed and preliminary validation results are presented. 3D point cloud sequences were acquired using a single time-of-flight sensor working in a privacy preserving modality and they were processed with a low power embedded PC. In order to satisfy different application requirements (e.g. covered distance range, processing speed and discrimination capabilities), a twofold discrimination approach was investigated in which features were hierarchically arranged from coarse to fine by exploiting both topological and volumetric representations. The topological representation encoded the intrinsic topology of the body's shape using a skeleton-based structure, thus guaranteeing invariance to scale, rotations and postural changes and achieving a high level of detail with a moderate computational cost. On the other hand, using the volumetric representation features were described in terms of 3D cylindrical histograms working within a wider range of distances in a faster way and also guaranteeing good invariance properties. The discrimination capabilities were evaluated in four different real-home scenarios related with the fields of ambient assisted living and homecare, namely "dangerous event detection", "anomalous behaviour detection", "activities recognition" and "natural human-ambient interaction". For each mentioned scenario, the discrimination capabilities were evaluated in terms of invariance to viewpoint changes, representation capabilities and classification performance, achieving promising results. The two feature representation approaches exhibited complementary characteristics showing high reliability with classification rates greater than 97%.

Automatic sleep staging from ventilator signals in non-invasive ventilation

Non-invasive ventilation (NIV), a recognized treatment for chronic hypercapnic respiratory failure, is predominantly applied at night. Nevertheless, the quality of sleep is rarely evaluated due to the required technological complexity. A new technique for automatic sleep staging is here proposed for patients treated by NIV. This new technique only requires signals (airflow and hemoglobin oxygen saturation) available in domiciliary ventilators plus a photo-plethysmogram, a signal already managed by some ventilators. Consequently, electroencephalogram, electrooculogram, electromyogram, and electrocardiogram recordings are not needed. Cardiorespiratory features are extracted from the three selected signals and used as input to a Support Vector Machine (SVM) multi-class classifier. Two different types of sleep scoring were investigated: the first type was used to distinguish three stages (wake, REM sleep and nonREM sleep), and the second type was used to evaluate five stages (wake, REM sleep, N1, N2 and N3 stages). Patient-dependent and patient-independent classifiers were tested comparing the resulting hypnograms with those obtained from visual/manual scoring by a sleep specialist. An average accuracy of 91% (84%) was obtained with three-stage (five-stage) patient-dependent classifiers. With patient-independent classifiers, an average accuracy of 78% (62%) was obtained when three (five) sleep stages were scored. Also if the PPG-based and flow features are left out, a reduction of 4.5% (resp. 5%) in accuracy is observed for the three-stage (resp. five-stage) cases. Our results suggest that long-term sleep evaluation and nocturnal monitoring at home is feasible in patients treated by NIV. Our technique could even be integrated into ventilators.