Search

In this article, a semisupervised weighting method for feature dimension based on entropy is proposed for classification, dimension reduction, and correlation analysis. For real-world data, different feature dimensions usually show different importance. Generally, data in the same class are supposed to be similar, so their entropy should be small; and those in different classes are supposed to be dissimilar, so their entropy should be large. According to this, we propose a way to construct the weights of feature dimensions with the whole entropy and the innerclass entropies. The weights indicate the contribution of their corresponding feature dimensions in classification. They can be used to improve the performance of classification by giving a weighted distance metric and can be applied to dimension reduction and correlation analysis as well. Some numerical experiments are given to test the proposed method by comparing it with some other representative methods. They demonstrate that the proposed method is feasible and efficient in classification, dimension reduction, and correlation analysis.

In general, vehicle images have varying resolutions due to vehicles’ movements and different camera settings. However, most existing vehicle re-identification models are single-resolution deep networks trained with pre-uniformly resizing vehicle images, which underestimates adverse effects of varying resolutions and leads to unsatisfactory performance. A straightforward solution for dealing with varying resolutions is to train multiple vehicle re-identification models. Each model is independently trained with images of a specific resolution. However, this straightforward solution requires significant overhead and ignores intrinsic associations among different resolution images. For that, an efficient multi-resolution network (EMRN) is proposed for vehicle re-identification in this paper. Firstly, EMRN embeds a newly-designed multi-resolution feature dimension uniform module (MR-FDUM) behind a traditional backbone network (i.e., ResNet-50). As a result, the whole model can extract fixed dimensional features from different resolution images so that it can be trained with one loss function of fixed dimensional parameters rather than training multiple models. Secondly, a multi-resolution image randomly feeding strategy is designed to train EMRN, making each mini-batch data of a random resolution during the training process. Consequently, EMRN can implicitly learn collaborative multi-resolution features via only a unitary deep network. Experiments on three large scale datasets, i.e., VeRi776, VehicleID, and VRIC, demonstrate that EMRN is superior to state-of-the-art vehicle re-identification methods.

Touch modality identification has attracted increasing attention due to its importance in human-robot interactions. There are three issues involved in the tactile perception for the touch modality identification, including the high dimensionality of tactile signals, complex tensor morphology of tactile sensing units, and the misalignment among different tactile time-series samples. In this article, we propose a novel kernel-based approach to deal with these three issues in a unified framework. Specifically, the techniques, including sparse principal component analysis and subsampling, are employed to reduce the feature dimension. Then, a singular value decomposition (SVD)-based kernel is proposed to preserve the spatial information of the tactile sensing elements. The sample misalignment issue is addressed via the employment of a global alignment kernel. Moreover, the merits of these two kernels are fused through an ideal regularized composite kernel, which simultaneously takes the label information of the training set into consideration. The effectiveness of the proposed kernel-based approach is verified on a public touch modality data set with a comprehensive comparison with the competing methods.

With the growing needs of practical applications such as security monitoring, partial face recognition is a challenging but important issue, because the captured faces in real-world surveillance videos may be occluded or with variations. Though current face recognition methods perform well in relatively constrained scenes, they may suffer from degradation for partial faces. In this paper, we propose a framework of model-based transfer learning and sparse coding (MTLSC) for partial face recognition. First, due to less information in partial face image, we exploit the mirrored image of an original probe sample as sample augment to provide further information. Considering the inadequacy of training face samples, we obtain face features based on model-based transfer learning VGGNet that is pre-trained on VGGFace dataset. Then we reconstruct face features by sliding window in view of different sizes of partial face hard to extract the same feature dimension. Finally we carry out sparse coding with rectification and calculate the minimum score of the probe and mirrored samples among all classes to get the results. Thus, by model-based transfer learning, sliding window for feature reconstruction and sparse coding with rectification, the proposed framework improves partial face recognition performance. Experimental results on three face databases (LFW, AR and NIR), and two person re-identification databases (iLIDS-VID and PKU-Reid) demonstrate our method is effective for partial face recognition.

Text non-text separation is one of the most essential pre-processing steps for any optical character recognition (OCR) system. As an OCR engine can only process texts, the non-texts present in an input document image are required to be suppressed at the initial level. Therefore, to build a complete OCR system, an efficient text non-text separation module needs to be developed. To this end, we have proposed a texture-based feature descriptor followed by a novel feature selection technique for region-based text non-text classification. First, we have incorporated rotation invariant property with local ternary pattern to form a new texture-based feature descriptor, rotation invariant local ternary pattern (RILTP). Next, a novel feature selection technique is proposed which is a modified version of binary particle swarm optimization (BPSO). For the evaluation of the proposed text non-text classification method, we have initially constructed a database consisting of 690 images of text and non-text regions extracted from 70 pages of RDCL 2015 and 75 pages of RDCL 2017 page segmentation competitions databases. In this database, each class contains 345 data samples. The proposed texture-based feature descriptor has obtained an accuracy of 97.09% on this database. Whereas, after applying BPSO, the feature dimension is reduced by approximately 55% and at the same time, the accuracy reaches 97.5%. Furthermore, in this work, another database is also created from Media team document pages to validate the robustness of this method. The second database comprises 100 text and 100 non-text images. The method has achieved 96.28% accuracy when it is trained with the first database and tested with the second database. The comparative study reveals the robustness and strength of the proposed method as it outnumbers many state-of-the-art texture-based features. Besides, the proposed feature selection method is also compared with various standard feature selection methods, and it has been observed that the proposed one outperforms all those methods considered here for comparison.

The linear separability effect refers to a benefit in search performance observed in a feature-search task, where target and distractor features vary along a continuous feature dimension: Search performance is best when there is a boundary in feature space that separates the distractor features from the target feature. However, the role that distractor heterogeneity plays in this effect is not well understood. Here, we reexamined this effect in the context of a new predictive procedure from Lleras et al. (2019) that quantifies the impact of distractor heterogeneity on search performance. Experiments 1A and 1B measured people's performance in homogeneous search conditions where they searched for the target among one type of distractor. The parameters observed in Experiments 1A and B were then used to predict search times in Experiments 2 and 3, where the target was presented in heterogeneous displays containing two types of distractors. The results show that total variance accounted for was 95% to 98%, without including any factor indexing the linear separability rule. The results demonstrate that heterogeneous search in orientation space is a function of target-distractor similarity and interitem interactions. The study highlights the robustness of the predictive procedure and demonstrates the generalizability of the method to estimate interitem interactions to new stimulus types. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

We propose a novel probabilistic label enhancement algorithm, called PLEA, to solve challenging label distribution learning (LDL) for multi-label classification problems. We adopt the well-known maximum entropy model based label distribution learner. However, unlike the existing LDL algorithms based on the maximum entropy model, we propose to use manifold learning to enhance the label distribution learner. Specifically, the supervised information in the label manifold is utilized in the feature manifold space construction to improve the accuracy of feature extraction, while dramatically reducing the feature dimension. Then the robust linear regression is employed to estimate the label distributions associated with the extracted reduced-dimension features. Using the enhanced reduced-dimension features and their associated estimated label distributions in the maximum entropy model, the unknown true label distributions can be estimated more accurately, while imposing considerably lower computational complexity. We evaluate the proposed PLEA method on a wide-range artificial and high-dimensional real-world datasets. Experimental results obtained demonstrate that our proposed PLEA method has advantages in LDL accuracy and runtime performance, compared to the latest multi-label LDL approaches. The results also show that our PLEA compares favourably with the state-of-the-arts multi-label learning algorithms for classification tasks.

Detection of dysphonia is useful for monitoring the progression of phonatory impairment for patients with Parkinson’s disease (PD), and also helps assess the disease severity. This paper describes the statistical pattern analysis methods to study different vocal measurements of sustained phonations. The feature dimension reduction procedure was implemented by using the sequential forward selection (SFS) and kernel principal component analysis (KPCA) methods. Four selected vocal measures were projected by the KPCA onto the bivariate feature space, in which the class-conditional feature densities can be approximated with the nonparametric kernel density estimation technique. In the vocal pattern classification experiments, Fisher’s linear discriminant analysis (FLDA) was applied to perform the linear classification of voice records for healthy control subjects and PD patients, and the maximum a posteriori (MAP) decision rule and support vector machine (SVM) with radial basis function kernels were employed for the nonlinear classification tasks. Based on the KPCA-mapped feature densities, the MAP classifier successfully distinguished 91.8% voice records, with a sensitivity rate of 0.986, a specificity rate of 0.708, and an area value of 0.94 under the receiver operating characteristic (ROC) curve. The diagnostic performance provided by the MAP classifier was superior to those of the FLDA and SVM classifiers. In addition, the classification results indicated that gender is insensitive to dysphonia detection, and the sustained phonations of PD patients with minimal functional disability are more difficult to be correctly identified.

Previous approaches to texture analysis and segmentation use multichannel filtering by applying a set of filters in the frequency domain or a set of masks in the spatial domain. This paper presents two new texture segmentation algorithms based on multichannel filtering in conjunction with neural networks for feature extraction and segmentation. The features extracted by Gabor filters have been applied for image segmentation and analysis. Suitable choices of filter parameters and filter bank coverage in the frequency domain to optimize the filters are discussed. Here we introduce two methods to optimize Gabor filter bank. First, a Gabor filter bank with a flat response is implemented and the optimal feature dimension is extracted by competitive networks. Second, a subset of Gabor filter bank is selected to compose the best discriminative filters, so that each filter in this small set can discriminate a pair of textures in a given image. In both approaches, multilayer perceptrons are employed to segment the extracted features. The comparisons of segmentation results generated using the proposed methods and previous research using Gabor, discrete cosine transform (DCT), and Laws filters are presented. Finally, the segmentation results generated by applying the optimized filter banks to textured images are presented and discussed.

In the field of pattern recognition, the problem of offline character recognition has long been a fascinating challenge. Several studies demonstrate that the process of image recognition is a little computationally expensive. The objective of this study includes an effective and reliable technique for the offline identification of Latin font characters produced by computers with a less number feature dimension. For which an effective representation technique for the character images is needed. The framework uses some machine learning techniques, SVM, random forest, RNN, and CNN, such as a standard recognition classifier. This study uses four feature extraction technologies to assess the efficiency of the proposed framework on a particular Chars74kFnt dataset. The extracted feature set originally introduced here is of 528 dimensions. Results show that CNNs deep neural network classifier outperforms the others without compromising efficiency. Later on, this dimension is reduced from 528 to just 220 by applying the Maximum Relevance Minimum Redundancy (mRMR) feature selection process. The results of these tests are shown in separate tables and also analyzed.

This work focuses on training difficulty and high computational cost caused by high-resolution input image when car plate character recognition based on support vector machine is applied. The use of singular value decomposition technology is proposed with improved optimal threshold analysis strategy to select retained features. The original data set is mapped in the low-dimensional feature space and the model is trained. The same test set is used to evaluate the models trained under different feature dimensions, and the optimal feature dimension is determined to 32 dimensional features. Results showed that the algorithm can accurately classify and detect license plate characters. The proposed model achieves 99.69% accuracy of the original data set, demonstrates excellent advantages, and remarkably simplifies the model design when only 96 dimensional features are used compared with traditional linear and convolutional neural network classifiers. New methods reduce the algorithm run time by 200%, and the number of parameter storage required by the algorithm to 25%. Hence, the proposed method can significantly improve the license plate recognition system usage in resource-constrained embedded devices.

Accurate precipitation prediction is very important for social life and economical activity. Prediction of the quantitative precipitation in semi-arid areas is difficult because of rain scarcity and volatility. In this study, the 3-h precipitation situation in the semi-arid region of Lanzhou is predicted, that is, the precipitation status after 3 h is forecasted on 5 levels: ‘no rain’, ‘light rain’, ‘moderate rain’, ‘heavy rain’ and ‘torrential rain’. We selected the meteorological data from 25 stations in and nearby Lanzhou, and processed the data with lag, difference and multiplication. Due to the large number of features, we use Mutual Information (MI) feature extraction method to reduce feature dimension, extract the features that are highly correlated with the target variable, and introduce spatio-temporal information in this way. Precipitation in semi-arid areas also has the problem of sample imbalance. We oversampled the data using Adaptive Synthetic (ADASYN) sampling approach and generated some minority class samples. Based on the MI feature extraction method and the ADASYN oversampling method, we constructed an Adaptive Synthesis and Mutual Information extraction Matrix (ASMI-M), which is the feature matrix used for model training. Then we proposed a Convolutional Neural Network and Long Short Term Memory (CNN-LSTM) model based on deep learning to predict the 3-h precipitation in Lanzhou City, which has achieved better prediction performance than traditional machine learning methods.

Three-dimensional (3D) shape recognition is a popular topic and has potential application value in the field of computer vision. With the recent proliferation of deep learning, various deep learning models have achieved state-of-the-art performance. Among them, multiview-based 3D shape representation has received increased attention in recent years, and related approaches have shown significant improvement in 3D shape recognition. However, these methods focus on feature learning based on the design of the network and ignore the correlation among views. In this article, we propose a novel progressive feature guide learning network (PGNet) that focuses on the correlation among multiple views and integrates multiple modalities for 3D shape recognition. In particular, we propose two information fusion schemes from visual and feature aspects. The visual fusion scheme focuses on the view level and employs the soft-attention model to define the weights of views for visual information fusion. The feature fusion scheme focuses on the feature dimension information and employs the quantified feature as the mask to further optimize the feature. These two schemes jointly construct a PGNet for 3D shape representation. The classic ModelNet40 and ShapeNetCore55 datasets are applied to demonstrate the performance of our approach. The corresponding experiment also demonstrates the superiority of our approach.

Spoken language identification (LID) is the process of determining and classifying natural language from a given content and dataset. Data must be processed to extract useful features to perform LID. The mel-frequency cepstral coefficient (MFCC) is one of the most popular feature extraction techniques in LID. The MFCC features are generated to serve as inputs for the classification stage. In this study, reduction in the MFCC feature dimension is investigated because large data size affects the computational time and resources (i.e., memory space) and slows the identification speed. The implementation of data reduction techniques to retain the most important feature parameters is also evaluated in this study. The investigation of data reduction is based on standard deviation (STD) calculation and principal component analysis (PCA). The features based on MFCC and the reduced dimensions based on STD and PCA results are then used as inputs to an optimized extreme learning machine (ELM) classifier called the optimized genetic algorithm-ELM (OGA-ELM). Several sets of data samples with one dimension of principal components (i.e., 119) are utilized for the evaluation. The results are generated using two different datasets. The first dataset is derived from eight separate languages, whereas the second dataset is a part of the National Institute of Standards and Technology Language Recognition Evaluation 2009 dataset. To evaluate the performance of the proposed method, this study utilizes several assessment measures, namely, accuracy, recall, precision, F-measure, G-mean, and identification time. The best LID performance is observed when the MFCC based on STD and PCA features with 119 feature dimensions is used with OGA-ELM as the classifier. The experimental results show that the proposed MFCC method achieves 99.38% accuracy using the first dataset. Additionally, it achieves accuracies of up to 97.60%, 96.80%, and 91.20% using the second dataset with durations of 30, 10, and 3 s, respectively. The proposed MFCC method exhibits the fastest computational time in all experiments, requiring only a few seconds to identify languages. Using a data reduction technique can substantially speed up the computational time, overcome resource limitations, and improve LID performance.

Background As proven to reflect the work state of heart and physiological situation objectively, electrocardiogram (ECG) is widely used in the assessment of human health, especially the diagnosis of heart disease. The accuracy and reliability of abnormal ECG (AECG) decision depend to a large extent on the feature extraction. However, it is often uneasy or even impossible to obtain accurate features, as the detection process of ECG is easily disturbed by the external environment. And AECG got many species and great variation. What’s more, the ECG result obtained after a long time past, which can not reach the purpose of early warning or real-time disease diagnosis. Therefore, developing an intelligent classification model with an accurate feature extraction method to identify AECG is of quite significance. This study aimed to explore an accurate feature extraction method of ECG and establish a suitable model for identifying AECG and the diagnosis of heart disease. Methods In this research, the wavelet combined with four operations and adaptive threshold methods were applied to filter the ECG and extract its feature waves first. Then, a BP neural network (BPNN) intelligent model and a particle swarm optimization (PSO) improved BPNN (PSO-BPNN) intelligent model based on MIT-BIH open database was established to identify ECG. To reduce the complexity of the model, the principal component analysis (PCA) was used to minimize the feature dimension. Results Wavelet transforms combined four operations and adaptive threshold methods were capable of ECG filtering and feature extraction. PCA can significantly deduce the modeling feature dimension to minimize the complexity and save classification time. The PSO-BPNN intelligent model was suitable for identifying five types of ECG and showed better effects while comparing it with the BPNN model. Conclusion In summary, it was further concluded that the PSO-BPNN intelligent model would be a suitable way to identify AECG and provide a tool for the diagnosis of heart disease.

Attentional mechanisms in perception can operate over locations, features, or objects. However, people direct attention not only towards information in the external world, but also to information maintained in working memory. To what extent do perception and memory draw on similar selection properties? Here we examined whether principles of object-based attention can also hold true in visual working memory. Experiment 1 examined whether object structure guides selection independently of spatial distance. In a memory updating task, participants encoded two rectangular bars with colored ends before updating two colors during maintenance. Memory updates were faster for two equidistant colors on the same object than on different objects. Experiment 2 examined whether selection of a single object feature spreads to other features within the same object. Participants memorized two sequentially presented Gabors, and a retro-cue indicated which object and feature dimension (color or orientation) would be most relevant to the memory test. We found stronger effects of object selection than feature selection: accuracy was higher for the uncued feature in the same object than the cued feature in the other object. Together these findings demonstrate effects of object-based attention on visual working memory, at least when object-based representations are encouraged, and suggest shared attentional mechanisms across perception and memory.

Accurate classification of Alzheimer’s disease (AD) and mild cognitive impairment (MCI), especially distinguishing the progressive MCI (pMCI) from stable MCI (sMCI), will be helpful in both reducing the risk of converting into AD and also releasing the burden on the family and even the society. In this study, a novel deep belief network (DBN) based multi-task learning algorithm is developed for the classification issue. In particular, the dropout technology and zero-masking strategy are exploited for getting over the overfitting problem and also enhancing the generalization ability and robustness of the model. Then, a new framework based on the DBN-based multi-task learning is established for accurate diagnosis of AD. After MRI preprocessing, not only the principal component analysis is utilized to reduce the feature dimension, but also multi-task feature selection approach is introduced to select the feature set related to all tasks as a result of taking the internal relevancy among multiple related tasks into consideration. Using data from the ADNI dataset, our method achieves satisfactory results in six tasks of health control (HC) vs. AD, HC vs. pMCI, HC vs. sMCI, pMCI vs. AD, sMCI vs. AD and sMCI vs. pMCI with the accuracies are 98.62%, 96.67%, 92.31%, 91.89%, 99.62% and 87.78%, respectively. Experimental results demonstrate that the DBN-based MTL algorithm developed in this study is an effective, superior and practical method of AD diagnosis.

The current mainstream methods of loop closure detection in visual simultaneous localization and mapping (SLAM) are based on bag-of-words (BoW). However, traditional BoW-based approaches are strongly affected by changes in the appearance of the scene, which leads to poor robustness and low precision. In order to improve the precision and robustness of loop closure detection, a novel approach based on stacked assorted auto-encoder (SAAE) is proposed. The traditional stacked auto-encoder is made up of multiple layers of the same autoencoder. Compared with the visual BoW model, although it can better extract the features of the scene image, the output feature dimension is high. The proposed SAAE is composed of multiple layers of denoising auto-encoder, convolutional auto-encoder and sparse auto-encoder, it uses denoising auto-encoder to improve the robustness of image features, convolutional auto-encoder to preserve the spatial information of the image, and sparse auto-encoder to reduce the dimensionality of image features. It is capable of extracting low to high dimensional features of the scene image and preserving the spatial local characteristics of the image, which makes the output features more robust. The performance of SAAE is evaluated by a comparison study using data from new college dataset and city centre dataset. The methodology proposed in this paper can effectively improve the precision and robustness of loop closure detection in visual SLAM.

The pathophysiology of major depressive disorder (MDD) has been explored to be highly associated with the dysfunctional integration of brain networks. It is therefore imperative to explore neuroimaging biomarkers to aid diagnosis and treatment. In this study, we developed a spatiotemporal graph convolutional network (STGCN) framework to learn discriminative features from functional connectivity for automatic diagnosis and treatment response prediction of MDD. Briefly, dynamic functional networks were first obtained from the resting‐state fMRI with the sliding temporal window method. Secondly, a novel STGCN approach was proposed by introducing the modules of spatial graph attention convolution (SGAC) and temporal fusion. A novel SGAC was proposed to improve the feature learning ability and special anatomy prior guided pooling was developed to enable the feature dimension reduction. A temporal fusion module was proposed to capture the dynamic features of functional connectivity between adjacent sliding windows. Finally, the STGCN proposed approach was utilized to the tasks of diagnosis and antidepressant treatment response prediction for MDD. Performances of the framework were comprehensively examined with large cohorts of clinical data, which demonstrated its effectiveness in classifying MDD patients and predicting the treatment response. The sound performance suggests the potential of the STGCN for the clinical use in diagnosis and treatment prediction., In this study, we developed a spatiotemporal graph convolutional network framework to learn discriminative features from functional connectivity for automatic diagnosis and treatment response prediction of major depressive disorder. Performances of the framework were comprehensively examined with large cohorts of clinical data, which demonstrated its effectiveness in classifying major depressive disorder patients and predicting the treatment response.

In 1993, Krishnapuram and Keller proposed possibilistic c -means (PCM) clustering, where the PCM had various extensions in the literature. However, the PCM algorithm with its extensions treats data points under equal importance for features. In real applications, different features in a dataset should take different importance with different weights. In this article, we first propose a feature-weighted PCM (FW-PCM). We then construct a feature-reduction framework. Therefore, we give a feature-weighted possibilistic c -means clustering with a feature-reduction framework, termed as a feature-weighted reduction PCM (FW-R-PCM) algorithm. The proposed FW-R-PCM can improve the clustering performance of PCM by calculating feature weights to identify important features, and so it can consequently eliminate these irrelevant features to reduce feature dimension. Its theoretical behavior and computational complexity are also analyzed. The effectiveness and usefulness of FW-R-PCM are demonstrated through experimental results using synthetic and real datasets, where comparisons of FW-R-PCM with PCM, FW-PCM, and some existing feature-weighted clustering algorithms are also made.

Cancer is one of the deadly diseases prevailing worldwide and the patients with cancer are rescued only when the cancer is detected at the very early stage. Early detection of cancer is essential as, in the final stage, the chance of survival is limited. The symptoms of cancers are rigorous and therefore, all the symptoms should be studied properly before the diagnosis. Thus, an automatic prediction system is necessary for classifying cancer as malignant or benign. Hence, this paper introduces the novel strategy based on the JayaAnt lion optimization-based Deep recurrent neural network (JayaALO-based DeepRNN) for cancer classification. The steps followed in the developed model are data normalization, data transformation, feature dimension detection, and classification. The first step is data normalization. The goal of data normalization is to eliminate data redundancy and to mitigate the storage of objects in a relational database that maintains the same information in several places. After that, the data transformation is carried out based on log transformation that generates the patterns using more interpretable and helps fulfill the supposition, and to reduce skew. Also, the non-negative matrix factorization is employed for reducing the feature dimension. Finally, the proposed JayaALO-based DeepRNN method effectively classifies cancer based on the reduced dimension features to produce a satisfactory result. Thus, the resulted output of the proposed JayaALO-based DeepRNN is employed for cancer classification. The proposed JayaALO-based DeepRNN showed improved results with maximal accuracy of 95.97%, maximal sensitivity of 95.95%, and maximal specificity of 96.96%.

Deep learning has been successfully applied in process monitoring in recent years due to its powerful feature extraction. However, these monitoring methods are difficult to extract intrinsic representations of the process data in complex nonlinear processes. A new deep neural network, residual attention convolutional autoencoder (RACAE) is proposed for process monitoring. The unsupervised learning method of RACAE can extract representative features from high-dimensional data, which can significantly improve process monitoring performance in nonlinear processes. RACAE effectively integrates convolution calculation with an autoencoder to perform effective feature extraction of multivariate data. Moreover, residual attention block is embedded in the autoencoder to select these key features and then reduce the feature dimension for detector. A new process monitoring model is proposed and two kinds of statistics are developed for fault detection. The effectiveness of RACAE in fault detection is evaluated through a numerical case and two benchmark processes. The convolutional autoencoder based on residual attention provides a new approach for feature learning and process monitoring of complex processes.

Introduction: In recent decades, the growing rate of cancer incidence is a big concern for most societies. Due to the genetic origins of cancer disease, its internal structure is necessary for the study of this disease. Methods: In this research, cancer data are analyzed based on DNA sequences. The transition probability of occurring two pairs of nucleotides in DNA sequences has Markovian property. This property inspires the idea of feature dimension reduction of DNA sequence for overcoming the high computational overhead of genes analysis. This idea is utilized in this research based on the Markovian property of DNA sequences. This mapping decreases feature dimensions and conserves basic properties for discrimination of cancerous and non-cancerous genes. Results: The results showed that a non-linear support vector machine (SVM) classifier with RBF and polynomial kernel functions can discriminate selected cancerous samples from non-cancerous ones. Experimental results based on the 10-fold cross-validation and accuracy metrics verified that the proposed method has low computational overhead and high accuracy. Conclusion: The proposed algorithm was successfully tested on related research case studies. In general, a combination of proposed Markovian-based feature reduction and non-linear SVM classifier can be considered as one of the best methods for discrimination of cancerous and non-cancerous genes.

Water body extraction from remote sensing images is an important task. Deep learning has become a more popular method for extracting water bodies from remote sensing images. However, these methods are usually aimed at a specific sensor and are not applicable. Thus, we proposed a new network, called the dense-local-feature-compression (DLFC) network aiming at extracting water body from different remote sensing images automatic. In this network, each layer of the network can receive the feature maps of all layers before it by the densely connected module of DenseNet. The concatenate operation on the feature dimension is used when connecting across layers. It can realize the different levels of features reuse. The local-feature-compression module is introduced before concatenate operation. It can obtain the more abstract features further by the convolution operation. Through the DLFC, we can fuse the spatial and spectral information for the remote sensing images that can extract water body from different remote sensing images. Besides, we construct a new water body dataset based on GaoFen-2 (GF-2) remote sensing images. The proposed DLFC achieved excellent performance with GF-2, GaoFen-6, Sentinel-2, and ZY-3 remote sensing images. Compared with the traditional water body extraction method and contemporary networks, the DLFC exhibits noticeable improvement. The results indicate that the DLFC can realize water body extraction from multisource remote sensing images automatically and rapidly.

Efficient and accurate monitoring of green tide is of great significance to marine disaster prevention and marine environment protection. A method is proposed in this article for the automatic extraction of the green tide from Chinese Gaofen-3 (GF-3) satellite synthetic aperture radar (SAR) images, which is based on feature selection and deep learning. In this article, since SAR images contain rich polarization information, we first employ H/A/α decomposition and other methods for the extraction of high-dimensional features from GF-3 SAR images. Second, a novel feature selection method for SAR images is designed by using the Bhattacharyya distance and the Separability index, which can select the optimal features subset with a strong ability for recognizing green tide and without correlation between features from the high-dimensional features of SAR. Then, to alleviate the model training burden and improve the prediction efficiency, a lightweight semantic segmentation network, called Mobile-SegNet, is designed based on MobileNets and SegNet. Finally, the selected optimal features and their labels are sent to Mobile-SegNet for training and obtaining the automatic recognition model of green tide, and in turn, the automatic extraction of green tide is achieved through model prediction. To verify the effectiveness of the proposed green tide extraction method, GF-3 SAR remote sensing images taken in 2020 that covered the Yellow Sea are collected and used in the green tide extraction experiment. The results show that the proposed method is available for an effective reduction of the feature dimension required for green tide extraction, and the improvement of the accuracy and efficiency of green tide detection. The overall accuracy, F1-score, mean intersection over union, and the kappa coefficient of the proposed method reached 99.52%, 95.76%, 92.19%, and 0.92, respectively.

The attention-based SeriesNet (A-SeriesNet) combined augmented attention residual learning module-based convolutional neural network (augmented ARLM-CNN) subnetwork with hidden state attention module-based recurrent neural network (HSAM-RNN) subnetwork for conditional time series prediction with high accuracy. The augmented ARLM-CNN subnetwork has defects in extracting latent features of the multi-condition series. The forecasting accuracy will decrease when the feature dimension of the multi-condition series becomes high. The same problem also occurs in the HSAM-RNN subnetwork of A-SeriesNet. The dual-stage attention recurrent neural network (DA-RNN) proved that the attention-based encoder-decoder framework is an effective model for dealing with the above problem. This paper applies the DA-RNN to the HSAM-RNN subnetwork of A-SeriesNet and presents the triple-stage attention-based recurrent neural network (TA-RNN) subnetworks. Furthermore, this paper considers a CNN-based encoder-decoder structure named dual attention residual learning module-based convolutional neural network (DARLM-CNN) subnetwork to improve the augmented ARLM-CNN subnetwork of A-SeriesNet. Finally, this paper presents the triple-stage attention-based SeriesNet (TA-SeriesNet), which uses a new concatenation method instead of the element-wise multiplication of A-SeriesNet to parallel connect the proposed subnetworks and reduce the dependence of forecasting results on a certain subnetwork. The experimental results show our TA-SeriesNet is superior to other deep learning models in forecasting accuracy evaluation metrics for high feature dimensional time series datasets.

Hyperspectral images (HSIs) include hundreds of spectral bands, which lead to Hughes phenomenon in classification task and decrease the classification accuracy. Feature selection can remove redundant and noisy features in the HSIs to overcome this phenomenon. In real applications, we may face a HSI scene with only a few labeled samples. Meanwhile, there are adequate labeled samples in a similar HSI scene. For example, they share the same land-cover classes. The shared information can be used to help the scene with a few labeled samples in feature selection. Traditional single-scene-based feature selection appears powerless in solving such problems. Cross-scene feature selection provides an attractive way to select feature subsets by simultaneously using the information from two HSI scenes. However, spectral distribution may change due to atmospheric conditions, yielding spectral shift. In order to tackle this problem, we propose a cross-domain algorithm based on hybrid whale optimization algorithm with simulated annealing (WOASA). The newly proposed algorithm is dubbed cross-domain WOASA (CDWOASA). CDWOASA simultaneously considers the separability of different land-cover classes and the consistency of selected features between two scenes, leading to discriminative and domain-invariant characters of selected feature subset. Moreover, since the original WOASA is not able to precisely control the dimension of selected features, we propose an improvement using a sorting strategy based on the fitness function value, thus making the output feature dimension precisely controlled. The experimental results on two cross-scene HSI datasets demonstrate the superiority of CDWOASA in cross-scene feature selection.

In multi-modal learning tasks such as video understanding, the most important operations are feature extraction, feature enhancement for single modality and feature aggregation between modalities. In this paper, we present two attention based algorithms, the Position-embedding Non-local (PE-NL) Network and the Multi-modal Attention (MA) feature aggregation method. Inspired by Non-local Neural Networks and Transformers, our PE-NL is a self-attention liked feature enhancement operation and it can capture long-range dependencies and model relative positions. The MA aggregation method merges visual and audio modals while reduces feature dimension and the number of parameters without losing too much accuracy. Both of PE-NL and MA blocks can be plugged into many multi-modal learning architectures. Our Gated PE-NL-MA network achieves competitive results on Youtube-8M dataset.

Recent discriminative trackers especially based on Correlation Filters (CFs) have shown dominant performance for visual tracking. This kind of trackers benefit from multi-resolution deep features a lot, taking the expressive power of deep Convolutional Neural Networks (CNN). However, distractors in complex scenarios, such as similar targets, occlusion, and deformation, lead to model drift. Meanwhile, learning deep features results in feature redundancy that the increasing number of learning parameters introduces the risk of over-fitting. In this paper, we propose a discriminative CFs based visual tracking method, called dimension adaption correlation filters (DACF). First, the framework adopts the multi-channel deep CNN features to obtain a discriminative sample appearance model, resisting the background clutters. Moreover, a dimension adaption operation is introduced to reduce relatively irrelevant parameters as possible, which tackles the issue of over-fitting and promotes the model effectively adapting to different tracking scenes. Furthermore, the DACF formulation optimization can be efficiently performed on the basis of implementing the alternating direction method of multipliers (ADMM). Extensive evaluations are conducted on benchmarks, including OTB2013, OTB2015, VOT2016, and UAV123. The experiments results show that our tracker gains remarkable performance. Especially, DACF obtains an AUC score of 0.698 on OTB2015.

The accuracy of behavioral interactive features is a key factor for improving the performance of rating prediction. In order to deeply explore the potential rules of user behavior and enhance the accurate representation of interactive features, this paper proposes two rating prediction models, based on the spatial dimension and distance measurement (SDDM), under the premise of taking the mean value of the user behavior history as a user feature, and obtaining the interactive features of an item and a user by calculating the distance between them in each feature dimension. In the proposed SDDM-Var and SDDM-PCC models, the variance and the Pearson correlation coefficient (PCC) are respectively utilized to evaluate the user’s attention to each feature dimension as to further obtain the weight vector of the interactive features. Finally, in order to improve the generalization ability of the proposed models, the rating prediction is accomplished by means of a specially designed multi-layer full-connection neural network. The conducted experiments with two public MovieLens datasets demonstrate the superior rating prediction performance of the proposed models in comparison with the existing baseline models, in terms of the root mean square error (RMSE), by achieving values of 0.865 and 0.872 on MovieLens 100K, and 0.839 and 0.832 on MovieLens 1M, respectively for SDDM-Var and SDDM-PCC.

In recent years, many researchers have explored different methods to obtain discriminative features for electroencephalogram-based (EEG-based) emotion recognition, but a few studies have been investigated on deaf subjects. In this study, we have established a deaf EEG emotion dataset, which contains three kinds of emotion (positive, neutral, and negative) with 15 subjects. Ten kinds of time–frequency domain features and eleven kinds of nonlinear dynamic system features were extracted from the EEG signals. To obtain the optimal feature combination and optimal classifier, an integrated genetic firefly algorithm (IGFA) was proposed. The multi-objective function with variable weight was utilized to balance the classification accuracy and the feature reduction ratio that are contradictory goals to find brighter fireflies in each generation. To retain the historical optimal solution and reduce the feature dimension, an optimal population protection scheme and subgroups generation scheme was carried out. The experimental results show that the averaged feature reduction rate of the proposed method is 0.959, and the averaged classification accuracy is 0.961. By investigating important brain regions, deaf subjects have common areas in the frontal and temporal lobes for EEG emotion recognition, while individual areas occur in the occipital and parietal lobes.

When a compound fault occurs, the randomness and ambiguity of the gearbox will cause uncertainty in the collected signal and reduce the accuracy of signal feature extraction. To improve accuracy, this research proposes a gearbox compound fault feature extraction method, which uses the inverse cloud model to obtain the signal feature value. First, EEMD is used to decompose the collected vibration signals of gearbox faults in normal and fault states. Then, the mutual information method is used to select the sensitive eigenmode function that can reflect the characteristics of the signal. Subsequently, the inverse cloud generator is used to extract cloud digital features and construct sample feature sets. On this basis, the concept of synthetic cloud is introduced, and the cloud-based distance measurement principle is used to synthesize new clouds, reduce the feature dimension, and extract relevant features. Finally, a simulation experiment on a rotating machinery unit with a certain type of equipment verifies that the proposed method can effectively extract the feature of gearbox multiple faults with less feature dimension. And comparing with the feature set extracted by the single cloud model, the results show that the method can better represent the fault characteristic information of the signal.

Face spoofing detection, also known as liveness detection, is a challenging and one of the most active research areas in computer vision. In this paper, a novel texture descriptor, namely equilibrium difference local binary pattern (ED-LBP), is proposed for the representation and recognition of face texture. First, the adjacent pixels discrepancy in a facial image is fully considered and the discrepancy information is encoded into LBP. This novel texture feature extraction method has the advantage of getting more elaborate texture information without increasing the feature dimension. Second, the texture signatures in different color channels are fully investigated. More specifically, the feature histograms are calculated over each image band separately. Third, by the integration of the chromatic ED-LBP histograms and the two-level spatial pyramid, the local structure information of face is encoded in our approach which can well describe the differences between facial videos of valid users and impostors. Finally, the ED-LBP histograms from different color spaces are usually cascaded into a united feature vector which is feed into SVM for classification identification. Extensive experiments on four challenging and publicly available face anti-spoofing databases, namely CASIA FASD, Replay-Attack, Replay-Mobile, and OULU-NPU, demonstrate the effectiveness of our proposed approach. The results indicate that our methods are superior to state-of-the-art techniques and can effectively resist photo and video spoofing attacks in face recognition (FR).

With the rapid development of the Internet, various forms of network attack have emerged, so how to detect abnormal behavior effectively and to recognize their attack categories accurately have become an important research subject in the field of cyberspace security. Recently, many hot machine learning-based approaches are applied in the Intrusion Detection System (IDS) to construct a data-driven model. The methods are beneficial to reduce the time and cost of manual detection. However, the real-time network data contain an ocean of redundant terms and noises, and some existing intrusion detection technologies have lower accuracy and inadequate ability of feature extraction. In order to solve the above problems, this paper proposes an intrusion detection method based on the Decision Tree-Recursive Feature Elimination (DT-RFE) feature in ensemble learning. We firstly propose a data processing method by the Decision Tree-Based Recursive Elimination Algorithm to select features and to reduce the feature dimension. This method eliminates the redundant and uncorrelated data from the dataset to achieve better resource utilization and to reduce time complexity. In this paper, we use the Stacking ensemble learning algorithm by combining Decision Tree (DT) with Recursive Feature Elimination (RFE) methods. Finally, a series of comparison experiments by cross-validation on the KDD CUP 99 and NSL-KDD datasets indicate that the DT-RFE and Stacking-based approach can better improve the performance of the IDS, and the accuracy for all kinds of features is higher than 99%, except in the case of U2R accuracy, which is 98%.

Wearable electroencephalography sensor integrated with seizure detection algorithm has become an emerging research topic in health management for patients with refractory epilepsy. To enhance seizure detection accuracy and reduce power consumption, this paper has proposed a selection method of optimal features customized for each patient. A feature pool consisting of several popular features is constructed. Then, the customized multi-domain features for each patient are selected from feature pool based on minimum redundancy maximum relevance. By removing redundant features, the computational complexity of on-line seizure detection can be greatly reduced, which is desired in wearable device where low-power and real-time operation are required. Four test datasets are used to evaluate the performance of proposed method, with detection accuracy calculated. Experiment results show that different datasets vary in optimal features. By using the proposed method, the detection accuracy and the value of area under receiver operating characteristic curve can be improved by 11.49% and 0.13 respectively, while feature dimension being reduced to 1.

Text non-text classification is an important research problem in the domain of document image processing. Undesirably, this is an almost ignored research topic, particularly, when we consider the unconstrained offline handwritten document images. For text non-text classification, many times researchers employ high dimensional feature vectors, which not only increase the computation time and storage requirement, but also reduce the classification accuracy due to the presence of redundant or irrelevant features. Here lies the application of some feature selection (FS) algorithms in order to find out the relevant subset of the features from the original feature vector. In this paper, our aim is two-fold. Firstly, application of coalition game based FS technique to find out an optimal feature subset for classifying the components present in a handwritten document image either as text or non-text. Secondly, five variants of a popular texture based feature descriptor, called Local Binary Pattern (LBP), along with its basic version are fed to the FS module for identifying the useful patterns only which can pinpoint the regions of an image as most informative in terms of the said classification task. To the best of our knowledge, the approach is completely novel where coalition game based FS technique is applied for locating the feature-rich regions to be used for text non-text classification. For experimentation, we have prepared an in-house dataset along with its ground truth information which consists of 104 handwritten engineering class notes as well as laboratory copies that include handwritten and printed texts, graphical components and tables etc. Experimental outcomes confirm that the proposed approach not only helps in reducing the feature dimension significantly but also increases the recognition ability of all six feature vectors.

Automatic classification of glaucoma from fundus images is a vital diagnostic tool for Computer-Aided Diagnosis System (CAD). In this work, a novel fused feature extraction technique and ensemble classifier fusion is proposed for diagnosis of glaucoma. The proposed method comprises of three stages. Initially, the fundus images are subjected to preprocessing followed by feature extraction and feature fusion by Intra-Class and Extra-Class Discriminative Correlation Analysis (IEDCA). The feature fusion approach eliminates between-class correlation while retaining sufficient Feature Dimension (FD) for Correlation Analysis (CA). The fused features are then fed to the classifiers namely Support Vector Machine (SVM), Random Forest (RF) and K-Nearest Neighbor (KNN) for classification individually. Finally, Classifier fusion is also designed which combines the decision of the ensemble of classifiers based on Consensus-based Combining Method (CCM). CCM based Classifier fusion adjusts the weights iteratively after comparing the outputs of all the classifiers. The proposed fusion classifier provides a better improvement in accuracy and convergence when compared to the individual algorithms. A classification accuracy of 99.2% is accomplished by the two-level hybrid fusion approach. The method is evaluated on the public datasets High Resolution Fundus (HRF) and DRIVE datasets with cross dataset validation.