Multi-feature Fusion Strategy Research Articles

Detection of Straw Coverage under Conservation Tillage Based on an Improved Mask Regional Convolutional Neural Network (Mask R-CNN)

Conservation tillage, a crucial method for protecting soil fertility, depends heavily on maintaining adequate straw coverage. The current method of straw coverage detection relies primarily on manual measurement, which is both time-consuming and laborious. This paper introduces a novel straw coverage detection approach based on an improved mask regional convolutional neural network (Mask R-CNN) algorithm. Several images of wheat straw-covered fields were taken, and the dataset was augmented using techniques like image inversion, contrast enhancement, Gaussian noise addition, and translation after cropping the original images. These fields use a crop rotation cycle of wheat and corn. Subsequently, the straw images were annotated using the Labelme annotation tool to obtain the available straw instance segmentation dataset. The Mask R-CNN algorithm was improved by refining the mask generation network structure through a multi-feature fusion strategy, which interweaves features from both the encoder and the mask generation network, enhancing the model’s ability to capture detailed and shape information of the straw. Lastly, using the mask information output by the improved Mask R-CNN algorithm, the straw coverage was calculated by counting the proportion of pixels within each segmented region. In the results, compared to the original Mask R-CNN algorithm, our improved Mask R-CNN algorithm achieved an average improvement of 7.8% in segmentation accuracy, indicating that the improved Mask R-CNN algorithm offers superior segmentation performance. Thus, the new algorithm can achieve straw coverage detection with higher accuracy and can provide a reference for other agricultural applications.

A stacked deep multi-kernel learning framework for blast induced flyrock prediction

Blasting operations are widely and frequently used for rock excavation in Civil and Mining constructions. Flyrock is one of the most important issues induced by blasting operations in open pit mines, and therefore needs to be well predicted in order to identify the safety zone to prevent the potential injuries. For this purpose, 234 sets of blasting data were collected from Sungun Copper Mine site, and a stacked deep multi-kernel learning (SD-MKL) framework was proposed to estimate the blast induced flyrock with confidence accuracy. The proposed model uses the stacking-based representation learning framework (S-RL) to achieve deep learning on small-scale training sets. A multi-kernel learning model (MKL) is used as the base module of S-RL framework, which uses a multi-feature fusion strategy to generate multiple kernels with different kernel length in order to reduce the effort in tuning hyperparameters. In addition, this study further enhanced the predictive capability of SD-MKL by introducing the boosting method into the S-RL framework and hence proposed a boosted SD-MKL model. For comparison purpose, several existing machine learning models were implemented, i.e., kernel ridge regression (KRR), support vector machine (SVM), random forest (RF), gradient boosting decision tree (GBDT), ensemble deep random vector functional link (edRVFL), SD-KRR and SD-SVM. Our experimental results showed that the proposed boosted SD-MKL achieved the best overall performance, with the lowest RMSE of 0.21/1.73, MAE of 0.08/0.78, and the highest VAF of 99.98/99.24.

MFTrans: A multi-feature transformer network for protein secondary structure prediction

In the rapidly evolving field of computational biology, accurate prediction of protein secondary structures is crucial for understanding protein functions, facilitating drug discovery, and advancing disease diagnostics. In this paper, we propose MFTrans, a deep learning-based multi-feature fusion network aimed at enhancing the precision and efficiency of Protein Secondary Structure Prediction (PSSP). This model employs a Multiple Sequence Alignment (MSA) Transformer in combination with a multi-view deep learning architecture to effectively capture both global and local features of protein sequences. MFTrans integrates diverse features generated by protein sequences, including MSA, sequence information, evolutionary information, and hidden state information, using a multi-feature fusion strategy. The MSA Transformer is utilized to interleave row and column attention across the input MSA, while a Transformer encoder and decoder are introduced to enhance the extracted high-level features. A hybrid network architecture, combining a convolutional neural network with a bidirectional Gated Recurrent Unit (BiGRU) network, is used to further extract high-level features after feature fusion. In independent tests, our experimental results show that MFTrans has superior generalization ability, outperforming other state-of-the-art PSSP models by 3 % on average on public benchmarks including CASP12, CASP13, CASP14, TEST2016, TEST2018, and CB513. Case studies further highlight its advanced performance in predicting mutation sites. MFTrans contributes significantly to the protein science field, opening new avenues for drug discovery, disease diagnosis, and protein.

A two-stage computational framework for identifying antiviral peptides and their functional types based on contrastive learning and multi-feature fusion strategy.

Antiviral peptides (AVPs) have shown potential in inhibiting viral attachment, preventing viral fusion with host cells and disrupting viral replication due to their unique action mechanisms. They have now become a broad-spectrum, promising antiviral therapy. However, identifying effective AVPs is traditionally slow and costly. This study proposed a new two-stage computational framework for AVP identification. The first stage identifies AVPs from a wide range of peptides, and the second stage recognizes AVPs targeting specific families or viruses. This method integrates contrastive learning and multi-feature fusion strategy, focusing on sequence information and peptide characteristics, significantly enhancing predictive ability and interpretability. The evaluation results of the model show excellent performance, with accuracy of 0.9240 and Matthews correlation coefficient (MCC) score of 0.8482 on the non-AVP independent dataset, and accuracy of 0.9934 and MCC score of 0.9869 on the non-AMP independent dataset. Furthermore, our model can predict antiviral activities of AVPs against six key viral families (Coronaviridae, Retroviridae, Herpesviridae, Paramyxoviridae, Orthomyxoviridae, Flaviviridae) and eight viruses (FIV, HCV, HIV, HPIV3, HSV1, INFVA, RSV, SARS-CoV). Finally, to facilitate user accessibility, we built a user-friendly web interface deployed at https://awi.cuhk.edu.cn/∼dbAMP/AVP/.

Music-driven generative dance movement teaching game based on a multi-feature fusion strategy

The study aims to develop a music-driven dance teaching game that enhances players' dance learning. By utilizing a multi-feature fusion strategy, the game creates dance movements inspired by music. The model effectively combines music and dance features to generate various dance sequences. In performance tests, the research model successfully matches the tempo and intensity of the music with the generated dance. Overall, this approach improves players' dance movement learning ability. The teaching effectiveness of the research method is significantly superior, with an average value of 95.23 points in the teaching evaluation. In terms of running time, for simple dances, the average generation time is 10 s; for ordinary categories of dances, the average generation time is 15 s; for complex dances, the average generation time is 17 s. The fluency of the generated dances is 97.21 %. The study's findings confirm the effectiveness of the research method. The game's use of multi-feature fusion algorithms allows for adaptive adjustments in difficulty and complexity based on the player's learning situation and ability level. This provides a more diverse and immersive gaming experience, while also presenting innovative ideas for future game development and teaching design.

A Study on Detection Methods of Multi-Featured Fusion Teaching Factors in Blended College English Teaching

Abstract The detection of teaching factors is mainly from the perspectives of both teachers and students, and the multi-feature fusion of the teaching factor detection method is used to add possibilities for the improvement of the quality of blended teaching of college English. The use of auditory features in this paper is to separate the teaching scene video and obtain the teacher and student video scenes. Teacher and student voice clustering algorithm based on teaching rules is proposed to complete the feature division, and the combination of visual and auditory features is used to extract the blended teaching behaviors of college English and classify the types of teacher and student behaviors and behavioral probabilities. Create a multi-feature fusion teaching factor detection model by optimizing the multi-feature fusion process and multi-feature fusion strategy. Design an experimental configuration to detect multi-feature teaching factors that affect students’ concentration, attention, and teaching state. The multi-feature fusion model’s recognition accuracy for excellent concentration, excellent concentration, and poor concentration all reached 75.29% and above, and the probability of correctly recognizing excellent concentration was the greatest, gaining 96.80%.

IF-AIP: A machine learning method for the identification of anti-inflammatory peptides using multi-feature fusion strategy

Background:The most commonly used therapy currently for inflammatory and autoimmune diseases is nonspecific anti-inflammatory drugs, which have various hazardous side effects. Recently, some anti-inflammatory peptides (AIPs) have been found to be a substitute therapy for inflammatory diseases like rheumatoid arthritis and Alzheimer’s. Therefore, the identification of these AIPs is an emerging topic that is equally important. Methods:In this work, we have proposed an identification model for AIPs using a voting classifier. We used eight different feature descriptors and five conventional machine-learning classifiers. The eight feature encodings were concatenated to get a hybrid feature set. The five baseline models trained on the hybrid feature set were integrated via a voting classifier. Finally, a feature selection algorithm was used to select the optimal feature set for the construction of our final model, named IF-AIP. Results:We tested the proposed model on two independent datasets. On independent data 1, the IF-AIP model shows an improvement of 3%–5.6% in terms of accuracies and 6.7%–10.8% in terms of MCC compared to the existing methods. On the independent dataset 2, our model IF-AIP shows an overall improvement of 2.9%–5.7% in terms of accuracy and 8.3%–8.6% in terms of MCC score compared to the existing methods. A comparative performance analysis was conducted between the proposed model and existing methods using a set of 24 novel peptide sequences. Notably, the IF-AIP method exhibited exceptional accuracy, correctly identifying all 24 peptides as AIPs. The source code, pre-trained models, and all datasets are made available at https://github.com/Mir-Saima/IF-AIP.

Novel Multi-Feature Fusion Facial Aesthetic Analysis Framework

Machine learning has been used in facial beauty prediction studies. However, the integrity of facial geometric information is not considered in facial aesthetic feature extraction, and the impact of other facial attributes (expression) on aesthetics. We propose a novel multi-feature fusion facial aesthetic analysis framework (NMFA) to overcome this problem. First, we designed a facial shape feature, which is an intuitive, visual quantitative description, based on B-spline. Second, we designed a representative low-dimensional facial structural feature to establish the theoretical basis of the facial structure, based on facial aesthetic structure and expression recognition theory. Next, we designed texture and holistic features based on Gabor and VGG-face network. Finally, we used a multi-feature fusion strategy to fuse them for aesthetic evaluation. Experiments were conducted on four databases. The results revealed that the proposed method realizes the visualization of facial shape features, enriches geometric information, solves the problem of lack of facial geometric information and difficulty to understand, and achieves excellent performance with fewer parameters.

A Multi-Feature Fusion and Situation Awareness-Based Method for Fatigue Driving Level Determination

The detection and evaluation of fatigue levels in drivers play a crucial role in reducing traffic accidents and improving the overall quality of life. However, existing studies in this domain often focus on fatigue detection alone, with limited research on fatigue level evaluation. These limitations include the use of single evaluation methods and relatively low accuracy rates. To address these issues, this paper introduces an innovative approach for determining fatigue driving levels. We employ the Dlib library and fatigue state detection algorithms to develop a novel method specifically designed to assess fatigue levels. Unlike conventional approaches, our method adopts a multi-feature fusion strategy, integrating fatigue features from the eyes, mouth, and head pose. By combining these features, we achieve a more precise evaluation of the driver’s fatigue state level. Additionally, we propose a comprehensive evaluation method based on the analytic hierarchy process (AHP) and fuzzy comprehensive evaluation, combined with situational prediction. This approach effectively evaluates the fatigue state level of drivers at specific moments or stages and provides accurate predictions. Furthermore, we optimize the gated recurrent unit (GRU) network using an enhanced marine predator algorithm (MAP), which results in significant improvements in predicting fatigue levels during situational prediction. Experimental results demonstrate a classification accuracy of 92% across various scenarios while maintaining real-time performance. In summary, this paper introduces a novel approach for determining fatigue driving levels through multi-feature fusion. We also incorporate AHP-fuzzy comprehensive evaluation and situational prediction techniques, enhancing the accuracy and reliability of fatigue level evaluation. This research holds both theoretical and practical significance in the field of fatigue driving.

Continuous Estimation of Upper Limb Joint Angle Based on Stacked Denoising Autoencoder

In the human-robot interaction system of the rehabilitation robot for stroke rehabilitation, surface electromyography (sEMG) signal-based continuous joint angle estimation has essential significance and implementation value. However, the existing intra-subject mode is time-consuming and lacks generality, while the adoption of the new inter-subject mode tests the model’s generalization ability; at the same time, the often-adopted multi-feature fusion strategy makes the feature dimensionality too high and increases the computational pressure of the system. In this regard, firstly, four time-domain features of multi-channel sEMG are extracted as the initial features; then, a stacked denoising autoencoder (SDAE) network is constructed to encode the initial set of sEMG features in low dimensions and extract more robust low-dimensional features; finally, an LSTM network is introduced as the regression network between sEMG features and joint angles. The results indicate that the feature extraction method proposed is superior to other methods and can be used for the control of the rehabilitation robot with a stable and accurate continuous joint angle estimation during motion.

Fault Diagnosis of a Solenoid Valve Based on Multi-Feature Fusion

This paper presents a non-intrusive solenoid valve fault diagnosis method, which adopts the eigenvalue extraction scheme and combines the time-frequency analysis and time-domain analysis of the current signal through the solenoid valve fault pattern recognition method. Specifically, the research group selected the solenoid valve driving end current value and its second-order change rate curve as the characteristic curve, and then it extracted the time-domain parameters and the corresponding frequency band energy of the second-order change rate as the characteristic value. Then, the research group conducted time-frequency analysis on the basis of the current signal, created a multi-feature fusion feature vector and then input it into a multi-class support vector machine (SVM) for solenoid valve fault diagnosis. To evaluate the method, the research group developed a working solenoid valve model that integrated a magnetic circuit, circuit and spool motion, and the research group examined four typical solenoid valve states (normal, spring broken, spool stuck and spool slightly stuck), for which the research group collected the current signals during solenoid valve energization and closing. The trials involved analyzing the current characteristics of the drive end, both in simulation and experimental setups, and the research group demonstrated that the proposed drive end current detection diagnosis method can improve the accuracy of recognition by 15.22% compared with traditional neural network recognition methods. The SVM strategy exploiting a multi-feature fusion strategy can improve recognition accuracy by 8.7% and verification accuracy by 42.11% compared with SVM on the basis of the energy feature values.

Multi–feature fusion tracking algorithm based on peak–context learning

Object tracking is considered a critical process in most applications of computer vision. Recently, tracking algorithms that include correlation filter in their frameworks have gained massive popularity due to their high efficiency. The previous algorithms aim to learn the correlation filter by leveraging over all features of the target and its neighbors. However, in this paper, a new tracking algorithm that merges an elastic net constraint and a contextual information into the training scheme is proposed to estimate the target location successfully. The novel optimization problem can significantly strengthen the peak value of the target and effectively eliminate the distractive features. Moreover, most of the correlation filter trackers only use one single feature, which has poor ability under a sophisticated environment. For this reason, a multi–feature fusion strategy is proposed in the framework that embeds multiple features to enhance the tracking performance. Consequently, a multi–scale adaptive model is implemented to improve the tracking stability through scale variations. Besides, an updating mechanism is applied within the proposed framework to reduce the tracking drift. Extensive quantitative and qualitative experiments on challenging benchmarks show that this unified tracker model achieves impressive performance compared to correlation filter and deep trackers.

IThermo: A Sequence-Based Model for Identifying Thermophilic Proteins Using a Multi-Feature Fusion Strategy.

Thermophilic proteins have important application value in biotechnology and industrial processes. The correct identification of thermophilic proteins provides important information for the application of these proteins in engineering. The identification method of thermophilic proteins based on biochemistry is laborious, time-consuming, and high cost. Therefore, there is an urgent need for a fast and accurate method to identify thermophilic proteins. Considering this urgency, we constructed a reliable benchmark dataset containing 1,368 thermophilic and 1,443 non-thermophilic proteins. A multi-layer perceptron (MLP) model based on a multi-feature fusion strategy was proposed to discriminate thermophilic proteins from non-thermophilic proteins. On independent data set, the proposed model could achieve an accuracy of 96.26%, which demonstrates that the model has a good application prospect. In order to use the model conveniently, a user-friendly software package called iThermo was established and can be freely accessed at http://lin-group.cn/server/iThermo/index.html. The high accuracy of the model and the practicability of the developed software package indicate that this study can accelerate the discovery and engineering application of thermally stable proteins.

Object Tracking in Satellite Videos Based on Correlation Filter with Multi-Feature Fusion and Motion Trajectory Compensation

As a new type of earth observation satellite approach, video satellites can continuously monitor an area of the Earth and acquire dynamic and abundant information by utilizing video imaging. Hence, video satellites can afford to track various objects of interest on the Earth's surface. Inspired by the capabilities of video satellites, this paper presents a novel method to track fast-moving objects in satellite videos based on the kernelized correlation filter (KCF) embedded with multi-feature fusion and motion trajectory compensation. The contributions of the suggested algorithm are multifold. First, a multi-feature fusion strategy is proposed to describe an object comprehensively, which is challenging for the single-feature approach. Second, a subpixel positioning method is developed to calculate the object’s position and overcome the poor tracking accuracy difficulties caused by inaccurate object localization. Third, introducing an adaptive Kalman filter (AKF) enables compensation and correction of the KCF tracker results and reduces the object’s bounding box drift, solving the moving object occlusion problem. Based on the correlation filtering tracking framework, combined with the above improvement strategies, our algorithm improves the tracking accuracy by at least 17% on average and the success rate by at least 18% on average compared to the KCF algorithm. Hence, our method effectively solves poor object tracking accuracy caused by complex backgrounds and object occlusion. The experimental results utilize satellite videos from the Jilin-1 satellite constellation and highlight the proposed algorithm's appealing tracking results against current state-of-the-art trackers regarding success rate, precision, and robustness metrics.

Generative Adversarial Network Based on Multi-feature Fusion Strategy for Motion Image Deblurring

<p>Deblurring of motion images is a part of the field of image restoration. The deblurring of motion images is not only difficult to estimate the motion parameters, but also contains complex factors such as noise, which makes the deblurring algorithm more difficult. Image deblurring can be divided into two categories: one is the non-blind image deblurring with known fuzzy kernel, and the other is the blind image deblurring with unknown fuzzy kernel. The traditional motion image deblurring networks ignore the non-uniformity of motion blurred images and cannot effectively recover the high frequency details and remove artifacts. In this paper, we propose a new generative adversarial network based on multi-feature fusion strategy for motion image deblurring. An adaptive residual module composed of deformation convolution module and channel attention module is constructed in the generative network. Where, the deformation convolution module learns the shape variables of motion blurred image features, and can dynamically adjust the shape and size of the convolution kernel according to the deformation information of the image, thus improving the ability of the network to adapt to image deformation. The channel attention module adjusts the extracted deformation features to obtain more high-frequency features and enhance the texture details of the restored image. Experimental results on public available GOPRO dataset show that the proposed algorithm improves the peak signal-to-noise ratio (PSNR) and is able to reconstruct high quality images with rich texture details compared to other motion image deblurring methods.</p> <p> </p>

HCDC-SRCF tracker: Learning an adaptively multi-feature fuse tracker in spatial regularized correlation filters framework

Integrating multi-feature based on multi-layer features from the convolutional network or based on multiple hand-crafted features has been proved to be an effective way for improving tracking performance. In this work, we investigate how to integrate multi-layer convolutional features with hand-crafted features. Specifically, an adaptive multi-feature fusion strategy is proposed based on convolutional features from ResNet-101 and hand-crafted features from HOG as well as Grayscale in spatial regularized correlation filter framework. We fully consider the complementary advantages of multi-layer convolutional features and hand-crafted features to construct a robust and reliable appearance representation of the target. Comprehensive experimental results on benchmark datasets demonstrate that our tracker has achieved significant performance improvements in various challenging environments. Compared to the trackers based only on multi-layer convolutional features or complete hand-crafted fusion features, the most important is that our proposed tracker obtains more competitive tracking performance. Our tracker is publicly available. You can find open-sourced code of our tracker at https://github.com/binger1225/HCDC-SRCF.

Robust visual object tracking using context-based spatial variation via multi-feature fusion

With the emergence of camera technology, visual tracking has witnessed great attention in the field of computer vision. For instance, numerous discriminative correlation filter (DCF) methods are broadly used in tracking, nevertheless, most of them fail to efficiently find the target in challenging situations which leads to tracking failure throughout the sequences. In order to handle these issues, we propose contextual information based spatial variation with a multi-feature fusion method (CSVMF) for robust object tracking. This work incorporates the contextual information of the target to determine the location of the target accurately, which utilizes the relationship between the target and its surroundings to increase the efficiency of the tracker. In addition, we integrate the spatial variation information which measures the second-order difference of the filter to avoid the over-fitting problem caused by the changes in filter coefficient. Furthermore, we adopt multi-feature fusion strategy to enhance the target appearance by using different metrics. The tracking results from different features are fused by employing peak-to-sidelobe ratio (PSR) which measures the peak strength of the response. Finally, we conduct extensive experiments on TC128, DTB70, UAV123@10fps, and UAV123 datasets to demonstrate that the proposed method achieves a favorable performance over the existing ones.

A scalable parallel Chinese online encyclopedia knowledge denoising method based on entry tags and Spark cluster

Because of the open-collaborative of online encyclopedias, a large number of knowledge triples are improperly classified in online encyclopedia systems, and it is necessary to denoise and refine the open-domain encyclopedia Knowledge Bases (KBs) to improve the quality and precision. However, the lack and inaccuracy of triple semantic features lead to a poor refinement effect. In addition, considering large-scale encyclopedia KBs, the processing of massive knowledge will lead to too much computing time and poor scalability of the algorithm. To solve the problems of knowledge denoising in the Chinese encyclopedia system, first, based on data field theory, this paper proposes a new Cartesian product mapping-based method (TripleES) to calculate the semantic similarity of entity triples, based on which a method for quantifying the quality of entry tags is proposed. Second, to further improve the denoising effect on KBs, this paper proposes a new method (TriplePV) to compute the potential value of triple based on multi-feature fusion strategy to calculate the semantic distance between the “out-of-vocabulary” entry tags and embeds it into the potential function. Third, to ensure our algorithms have good scalability, the proposed denoising algorithms are implemented and optimized in parallel based on the Spark cluster-computing framework. Specifically, Spark-based TripleES (ES_Spark) and Spark-based TriplePV (PV_Spark) algorithms are proposed to calculate the semantic similarity and potential value of triples respectively. Finally, a comprehensive comparative analysis is performed on the denoising effect and time efficiency with the state-of-the-art distributed Chinese encyclopedia knowledge denoising algorithm. The experimental results on real-world datasets show that the parallel denoising algorithm proposed in this paper can improve the efficiency of knowledge denoising and the accuracy of KBs, which outperforms the state-of-the-art methods.

Bipartite Network of Interest (BNOI): Extending Co-Word Network with Interest of Researchers Using Sensor Data and Corresponding Applications as an Example

Traditional co-word networks do not discriminate keywords of researcher interest from general keywords. Co-word networks are therefore often too general to provide knowledge if interest to domain experts. Inspired by the recent work that uses an automatic method to identify the questions of interest to researchers like “problems” and “solutions”, we try to answer a similar question “what sensors can be used for what kind of applications”, which is great interest in sensor- related fields. By generalizing the specific questions as “questions of interest”, we built a knowledge network considering researcher interest, called bipartite network of interest (BNOI). Different from a co-word approaches using accurate keywords from a list, BNOI uses classification models to find possible entities of interest. A total of nine feature extraction methods including N-grams, Word2Vec, BERT, etc. were used to extract features to train the classification models, including naïve Bayes (NB), support vector machines (SVM) and logistic regression (LR). In addition, a multi-feature fusion strategy and a voting principle (VP) method are applied to assemble the capability of the features and the classification models. Using the abstract text data of 350 remote sensing articles, features are extracted and the models trained. The experiment results show that after removing the biased words and using the ten-fold cross-validation method, the F-measure of “sensors” and “applications” are 93.2% and 85.5%, respectively. It is thus demonstrated that researcher questions of interest can be better answered by the constructed BNOI based on classification results, comparedwith the traditional co-word network approach.

Graph Theory-based Fast Linear Iterative Clustering Multi-feature Fusion Model for Art Image Segmentation

ABSTRACT Image segmentation is the pre-processing stage of image analysis. It divides the image into different re-gions for the subsequent image analysis. The traditional segmentation algorithms can not deal with the images with complex background and uneven gray level effectively. Especially, for the art images seg-mentation with high resolution, the distinguishment degree between foreground of the object to be seg-mented and the background is small, which can lead to the incomplete segmentation effect. Therefore, this paper proposes a graph theorybased Fast Linear Iterative Clustering (FLIC) multi-feature fusion model for art image segmentation. First, the FLIC segmentation algorithm is used to pre-segment the su-perpixel of raw image. Second, the HOG feature, Lab color feature and spatial location feature are ex-tracted. Third, A multi-feature fusion strategy based on superpixel is designed. Four, the fast image seg-mentation based on multi-feature fusion is realized by using the graph theory. The graph theory is opti-mized by time domain convolution. Finally, the comparison experiments with other state-of-the-art methods are conducted on the public datasets: Berkeley Segmentation Dataset and Culture Gene Online . The results show that the proposed algorithm has better effect in terms of evaluation indexes (USE, BR, ASA and Time).