International Journal of Pattern Recognition and Artificial Intelligence

A classifier ensemble is a combination of diverse and accurate classifiers. Generally, a classifier ensemble performs better than any single classifier in the ensemble. Naive Bayes classifiers are simple but popular classifiers for many applications. As it is difficult to create diverse naive Bayes classifiers, naive Bayes ensembles are not very successful. In this paper, we propose Random Subclasses (RS) ensembles for Naive Bayes classifiers. In the proposed method, new subclasses for each class are created by using 1-Nearest Neighbor (1-NN) framework that uses randomly selected points from the training data. A classifier considers each subclass as a class of its own. As the method to create subclasses is random, diverse datasets are generated. Each classifier in an ensemble learns on one dataset from the pool of diverse datasets. Diverse training datasets ensure diverse classifiers in the ensemble. New subclasses create easy to learn decision boundaries that in turn create accurate naive Bayes classifiers. We developed two variants of RS, in the first variant RS(2), two subclasses per class were created whereas in the second variant RS(4), four subclasses per class were created. We studied the performance of these methods against other popular ensemble methods by using naive Bayes as the base classifier. RS(4) outperformed other popular ensemble methods. A detailed study was carried out to understand the behavior of RS ensembles.

Vehicle detection, as an important technology for urban intelligent transportation system, is having attracted increasingly interests of researchers in recent years. For the time cost problem of traditional road vehicles testing approach, a moving region extraction method based on Gaussian model is used to reduce the scanning area of the window, exclude some background noise and improve test speed. For the problem of traditional single feature, relatively lower detection rate and lack of ability to adapt to complex environment, a method based on the combination of Haar-like and 2bitBP (2bit Binary Pattern) features is adopted. Feature integration method enhances the expression of features. As a result, the improved classification performance of classifiers enables it to be adapted to different traffic environment. Firstly, a Gaussian mixture model is established to detect moving targets in overall region and then the Haar-like and 2bitBP features extraction are carried out in the region. At the end the action of cascading classification on samples achieve the detection of moving vehicles. The experimental results show that the method is effective for vehicle detection.

In order to accurately detect various micro particles in patient's urinary samples, a urinary micro particle detection system based on spatial coordinate tracking method was developed. A hierarchical microscopic image analysis method was described and the detailed spatial coordinate tracking algorithm was designed. The architecture and the procedure of proposed system were presented. Moreover, the prototype system was introduced and six patients' urinary samples were analyzed to verify the reliability and simplicity of unattended operation.

Analysis and classification for remote sensing landscape based on remote sensing imagery is a popular research topic. In this paper, we propose a new remote sensing data classifier by incorporating the support vector machine (SVM) learning information into the K-nearest neighbor (KNN) classifier. The SVM is well known for its extraordinary generalization capability even with limited learning samples, and it is very useful for remote sensing applications as data samples are usually limited. The KNN has been widely used in data classification due to its simplicity and effectiveness. However, the KNN is instance-based and needs to keep all the training samples for classification, which could cause not only high computation complexity but also overfitting problems. Meanwhile, the performance of the KNN classifier is sensitive to the neighborhood size $K$ and how to select the value of the parameter $K$ relies heavily on practice and experience. Based on the observations that the SVM can contribute to the KNN on the problems of smaller training samples size as well as the selection of the parameter $K$, we propose a support vector nearest neighbor (abbreviated as SV-NN) hybrid classification approach which can simplify the parameter selection while maintaining classification accuracy. The proposed approach is consist of two stages. In the first stage, the SVM is performed on the training samples to obtain the reduced support vectors (SVs) for each of the sample categories. In the second stage, a nearest neighbor classifier (NNC) is used to classify a testing sample, i.e. the average Euclidean distance between the testing data point to each set of SVs from different categories is calculated and the NNC identifies the category with minimum distance. To evaluate the effectiveness of the proposed approach, firstly experiments of classification for samples from remote sensing data are evaluated, and then experiments of identifying different land covers regions in the remote sensing images are evaluated. Experimental results show that the SV-NN approach maintains good classification accuracy while reduces the training samples compared with the conventional SVM and KNN classification model.

High utility sequential patterns (HUSP) refer to those sequential patterns with high utility (such as profit), which play a crucial role in many real-life applications. Relevant studies of HUSP only consider positive values of sequence utility. In some applications, however, a sequence consists of items with negative values (NIV). For example, a supermarket sells a cartridge with negative profit in a package with a printer at higher positive return. Although a few methods have been proposed to mine high utility itemsets (HUI) with NIV, they are not suitable for mining HUSP with NIV because an item may occur more than once in a sequence and its utility may have multiple values. In this paper, we propose a novel method High Utility Sequential Patterns with Negative Item Values (HUSP-NIV) to efficiently mine HUSP with NIV from sequential utility-based databases. HUSP-NIV works as follows: (1) using the lexicographic quantitative sequence tree (LQS-tree) to extract the complete set of high utility sequences and using I-Concatenation and S-Concatenation mechanisms to generate newly concatenated sequences; (2) using three pruning methods to reduce the search space in the LQS-tree; (3) traversing LQS-tree and outputting all the high utility sequential patterns. To the best of our knowledge, HUSP-NIV is the first method to mine HUSP with NIV, which is shown efficient on both synthetic and real datasets.

With the advances in technique for high throughput data gathering such as microarrays, DNA sequencing machines and continuous development of human genome project, the traditional physical and chemical methods have been more difficult to meet the requests of time consuming and results accuracy. Exploring and understanding the causal relationship of complex gene regulatory networks and transforming the massive data of large-scale biological research to useful biological knowledge are the present challenge. As a result, there are two typical applications both the confidence value prediction of DNA sequence and influence degree prediction of gene expression which have become the hot issues in our daily life. In this paper, two extended fuzzy Petri nets approaches are proposed, based on the existing fuzzy Petri net model, to model and analyze for the hot issues respectively. One is the fuzzy colored Petri net, which combines fuzzy Petri net with colored Petri net to model fuzzy rule-based reasoning and determine confidence values for bases called in DNA sequence. The other is extended fuzzy Petri net, which integrates reverse reasoning into fuzzy Petri net and is proposed to model gene regulatory network. It can predict the change in expression level of target based on the input expression level of activator/repressor. Compared with the method of fuzzy Petri net, the two extended fuzzy Petri nets models perform more accurately in the following typical experiment reasoning outcomes and show that the proposed methods are feasible and available.

This paper presents a new algorithm for fast exemplar-based image inpainting by using image pyramid and a novel pruning scheme. The proposed technique separates the inpainting process into three major steps. First the model determines which patch belonging to the inpainting region should be filled first. Then the pruning system analyzes the selected patch and weeds out regions having no chance to be the best matching patch. Finally, it performs a fast texture synthesis based on greedy search method in the remaining regions to fill the selected patch. Compared with previous methods, the proposed algorithm is much more efficient computationally. Experiments on synthetic and natural images show the advantages of the proposed algorithm, where it is shown to compare favorably with contemporary state-of-the-art schemes.

Improving the resolution of the image is convenient for people to study the local details of the image, and plays an important role in computer vision. The problem of generating a corresponding super-resolution (SR) image from a single low-resolution (LR) image is addressed via the join operation in the paper. Firstly, the LR image is partitioned into some patches, each patch is represented as the sphere granule set. Secondly, the join operation between two adjacent image patches is used to compensate the pixel value of SR image. Experimental results showed the feasibility and superiority via join operation by root mean square errors (RMSE) between the reconstructed SR image and the original image compared with bicubic interpolation and NNLasso.

The research of the mechanical properties of concrete, a kind of heterogeneous composite material, was previously established on basis of the mathematical model of random aggregate, which is used to study and analyze the mesoscopic damage mechanism of concrete. Although the shape and distribution of aggregate in the model built by this method are closer to the real structure of concrete, there is still a big difference between them and the real concrete specimen. In order to solve the problem of large amount of redundant computation in the CT reconstruction of full size cube space, a fast reconstruction method based on ray-casting algorithm is proposed. First, a method integrating the new bounding box technology with the plane intersection algorithm clusters were adopted to cut the body data and ray-casting effectively, and then, the polygon scanning and conversion was utilized to reduce the number of cast rays, finally, the adaptive sampling method was used to avoid repeatedly sampling same voxel so that the reconstruction efficiency of whole algorithm and the feasibility of numerical calculation can be enhanced. The experimental results demonstrate that the proposed algorithm can greatly improve the 3D rendering speed of concrete CT without affecting the image quality. It provides a more effective and reliable approach for correctly analyzing the mesoscopic damage mechanism and mechanical characteristics of concrete.

In view of requirements of low-resource consumption and high-efficiency in real-time Ambulatory Electrocardiograph Diagnosis (AED) applications, a novel Cardiac Arrhythmias Detection (CAD) algorithm is proposed. This algorithm consists of three core modules: an automatic-learning machine that models diagnostic criteria and grades the emergency events of cardiac arrhythmias by studying morphological characteristics of ECG signals and experiential knowledge of cardiologists; a rhythm classifier that recognizes and classifies heart rhythms basing on statistical features comparison and linear discriminant with confidence interval estimation; and an arrhythmias interpreter that assesses emergency events of cardia arrhythmias basing on a two rule-relative interpretation mechanisms. The experiential results on off-line MIT-BIH cardiac arrhythmia database as well as online clinical testing explore that this algorithm has 92.8% sensitivity and 97.5% specificity in average, so that it is suitable for real-time cardiac arrhythmias monitoring.

Memory demand is associated with increased mental workload. The objective of the present study was to examine the effects of visuospatial memory secondary tasks on driving performance. Memory tasks for the unknown word-figure pairs and recognition tasks for word-figure pairs at two-level difficulties were employed separately to represent working memory’s process and long-term memory’s process. A simulator study was conducted based on the simulation of the standard environment of Lane change test (LCT). The performance of lane keeping, lane change, and secondary tasks was measured by statistical methods. The comprehensive appraisal model was constructed to quantify total driving performance. The results showed that the mean path deviation, steering angle, and lane excursion times increased, and the proportion of correct lane change decreased, with the perceived workload increasing and the total driving performance decreasing in dual-task driving condition. Compared with the simple working memory group, as the difficulty of tasks increased in difficult working memory group, lane change performance degraded and the perceived workload increased. In contrast to difficult working memory group, the performance of lane keeping and lane change increased, while the perceived workload decreased and the total performance increased by about 50% in difficult recognition group. There were few differences between the simple working memory group and simple recognition group. The difficult working memory group had the lowest total driving performance. The results indicate that as the secondary task’s difficulty increases, driving performance will degrade. Performance improves significantly when the working memory process is converted to the recognition process. This trend is more obvious when the memory task assumes to be more difficult.

Nerve injury can cause lower limb paralysis and gait disorder. Currently lower limb rehabilitation exoskeleton robots used in the hospitals need more power to correct abnormal motor patterns of stroke patients’ legs. These gait rehabilitation robots are powered by cumbersome and bulky electric motors, which provides a poor user experience. A newly developed gait rehabilitation exoskeleton robot actuated by low-cost and lightweight pneumatic artificial muscles (PAMs) is presented in this research. A model-free proxy-based sliding mode control (PSMC) strategy and a model-based chattering mitigation robust variable control (CRVC) strategy were developed and first applied in rehabilitation trainings, respectively. As the dynamic response of PAM due to the compressed air is low, an innovative intention identification control strategy was taken in active trainings by the use of the subject’s intention indirectly through the estimation of the interaction force between the subject’s leg and the exoskeleton. The proposed intention identification strategy was verified by treadmill-based gait training experiments.