Search

According to the problem of the current RGB-D feature extraction network used for pose estimation was too large, this paper proposed a lightweight improved XYZNet RGB-D feature extraction network. First, this paper designed a lightweight sub-network BaseNet to replace ResNet18 in XYZNet, which made the network scale significantly reduced and obtained more powerful performance. Then, this paper proposed a re-parameterized multiscale convolutional attention (Rep-MSCA) sub-module based on depth separable convolution, which enhanced the ability of BaseNet to extract contextual information of different scales, and constrained the amount of parameters in the model. Finally, in order to improve the geometric feature extraction ability of PointNet in XYZNet with a small parameter cost, this paper designed a re-parameterized residual multi-layer perceptron (Rep-ResP) module. The floating point operations (FLOPs) and parameters of the improved network are 60.8% and 64.8% lower, the inference speed is 21.2% higher, and the accuracy of the mainstream datasets LineMOD and YCB-Video is 0.5% and 0.6% higher. The proposed model is more suitable for deployment in scenarios where hardware resources are tight. [ABSTRACT FROM AUTHOR]

In the intelligent transformation of the mechanical manufacturing, a large amount of spare parts models is inevitably generated, which brings huge challenges to the efficient retrieval of data. Considering the user-friendly and lightweight characteristics of design sketches, this paper conducted model retrieval of mechanical parts based on design sketches by constructing a deep cross-domain representation model. Aiming at the problem of cross-modal information association between sketches and 3D models, this paper proposed a feature joint learning method, which controlled the differences within and between classes of retrieved models. By this method, feature descriptors could acquire single-domain features as well as fuse the cross-domain information, and thus, the method established a consistent association representation of cross-modal data in a co-embedding space. Finally, it constructed the index table by using hash coding to realize fast retrieval of massive data. The experimental results on the parts data show that the proposed retrieval method based on the design sketch can not only achieve the most accurate retrieval results, but also have higher retrieval efficiency among the comparison methods. The method not only improves the accuracy of cross-modal spare parts information retrieval, but also improves the efficiency of data management, thereby indirectly improving the efficiency and convenience of product design. The related systems have been applied in some enterprises and have received good feedback. [ABSTRACT FROM AUTHOR]

Effectively representing features of modalities is a hot issue in cross-modal recipe retrieval. The current methods generally adopt two independent neural networks to extract the features of images and recipes respectively, achieving retrieval through cross-modal alignment. However, these methods mainly focus on the intra-modal information and ignore the intermodal interactions, resulting in the loss of some effective modality information. To address the problem, this paper proposed a cross-modal recipe retrieval method to enhance modality semantics through multimodal encoders. Firstly, it used a pre-trained model to extract initial semantic features of images and recipes and utilizing modality alignment to reduce the inter-model differences. Secondly, it employed the pairwise cross-modal attention to repeatedly reinforce the features of one modality by using features from another modality, extracted valid information. Thirdly, it used the self-attention mechanism to modal the internal features of modalities to capture rich modality-specific semantic information and potential associations. Finally, it introduced the triplet loss to minimize the distance between similar samples, achieved cross-modal retrieval learning. Experimental results on Recipe 1M dataset show that the proposed approach outperforms the current mainstream methods in terms of median ranking(MedR) and recall rate at top K(R@K), providing a powerful solution for cross-modal retrieval tasks. [ABSTRACT FROM AUTHOR]

Autonomous vehicle can calculate the movement track of surrounding vehicles according to the track prediction algorithm, and make response to reduce driving risk, while the traditional track prediction model will produce large errors in the case of long-term series prediction. To address this issue, this paper proposed a trajectory prediction model based on the Informer algorithm, and used the publicly available dataset NGSIM to conduct experimental analysison. Firstly, it filtered the original data by using symmetric exponential moving average method(sEMA), and added a joint normalization layer to the original Informer encoder to extract features from different vehicles, reducing the motion error between different vehicles, and improving the prediction accuracy by considering the speed information of the vehicle itself and the vehicle movement information of the surrounding environment. Finally, it got the vehicle trajectory distribution at the future time through the decoder. The results show that the trajectory prediction error of the model is less than 0.5 m. Through the analysis of MAE and MSE results of trajectory prediction, when the prediction time exceeds 0.3 s, the trajectory prediction effect of Informer model is obviously better than other algorithms, which verifies the effectiveness of the model and algorithm. [ABSTRACT FROM AUTHOR]

Most of the existing user profile methods lack different granularity text information representation, and there is a noise problem in the feature extraction stage, resulting in the inaccurate construction of the profile. To address these issues, this paper proposed a user profile method based on multi-granularity information fusion, called UP-MGIF. Firstly, it integrated the character-level granularity and the word-level granularity representation vectors in the embedding layer to expand feature content. Secondly, based on the improved bi-directional gated recurrent unit network(Bi-GRU), it designed a hybrid feature extraction model called Bi-GRU-DAE-Attention by combining denoising autoencoder (DAE) and attention mechanism to achieve feature denoising and semantic enhancement. Finally, it input the robust feature vectors into the classifier to achieve user profile generation. Experiments show that the user profile generation method achieves higher classification accuracy than other baseline methods on two profile datasets in the medical and Internet domains, and validate the effectiveness of each module through ablation experiments. [ABSTRACT FROM AUTHOR]

Aiming at the issue of insufficient utilization of temporal information, speaker information, and multi-modal information in existing conversational emotion recognition methods, this paper proposed a multi-modal temporal-aware DAG model(MTDAG). The designed temporal-aware unit optimized the discourse weight setting in chronological order and collected historical emotional cues to achieve more effective utilization of temporal and historical information based on recency effect. The context and speaker information fusion module achieved the full utilization of discourse information by extracting the deep joint information of contextual context and speaker self-context. By setting the DAG subgraphs to capture multi-modal information and constrain the interaction direction, the model achieved full utilization of multi-modal information while reducing the introduction of noise. Extensive experiments conducted on two benchmark datasets, IEMOCAP and MELD, demonstrate that the model exhibits excellent performance in emotion recognition. [ABSTRACT FROM AUTHOR]

In image retrieval based on deep hashing, to solve the problems of low feature extraction efficiency in convolutional neural networks (CNN ) and underutilization of feature correlation, this paper proposed a novel method combining sparse difference network and multi-supervised hashing( SDNMSH ), and used it for efficient image retrieval. SDNMSH took pairs of images as training inputs, and guided hash codes learning through an elaborately designed sparse difference convolutional neural network and a supervised hash function. The sparse difference convolutional layer and the vanilla convolutional layer composed the sparse difference convolutional neural network. The sparse difference convolutional layer could quickly extract rich feature information, to achieve efficient feature extraction of the entire network. At the same time, in order to make full use of the pairwise correlation of semantic information and features, so as to promote the feature information extracted by the network to be more effectively transformed into discriminative hash codes, and then to achieve efficient image retrieval by using SDNMSH, this paper adopted a multi-supervised hash( MSH) function and designed an objective function for this purpose. Extensive experimental results on three widely used datasets MNIST,CIFAR-10 and NUS-WIDE show that SDNMSH achieves better retrieval performance, compared with the state-of-the-arts. [ABSTRACT FROM AUTHOR]

Mild cognitive decline is the early stage of Alzheimer's disease, and the feature extraction and classification of mild cognitive decline using EEG signals is an important method for diagnosing mild cognitive decline. In the automatic detection technology of mild cognitive decline based on electroencephalogram artificial intelligence, the existing research only extracted a certain feature in the electroencephalogram signal or simply concatenates multiple features, which caused these methods to fail to well consider the correlation between different features and it would cause the problem of dimensional disaster. This paper proposed a convolutional neural network based automatic detection algorithm for resting state electroencephalogram data of mild cognitive decline. By extracting the power spectrum and brain network features of the electroencephalogram, it fused the two features by matrix operation, and designed a convolutional neural network to classify the fused features. This method achieves a high accuracy rate on the data set collected by a hospital in Shanghai. In addition, by inputting different subsets of the feature set, this method found the few groups of features that contribute the most to mild cognitive decline, thereby it also had a certain interpretability. Experiments on the dataset of this paper, it p roves the advantages of the power brain network for the automatic diagnosis of mild cognitive decline. [ABSTRACT FROM AUTHOR]

In the process of registering point clouds with low feature recognition, traditional methods based on local feature extraction and matching are usually not accurate, while the accuracy and efficiency of methods based on global feature matching are also hard to guarantee. In response to this problem, this paper proposed an improved local feature matching method. In initial registration, it designed a key point extraction method based on normal vector projection covariance analysis. Then it used fast point feature histogram( FPFH) descriptor to characterize these key points,and defined multiple matching conditions to screen the feature points. Finally, it took the sum of the nearest distance of the corresponding points as the optimization goal for rough matching. In fine registration, this paper used the improved iteration closest point(ICP) algorithm, which took the minimum distance from point to plane as the object of iterative optimization, for accurate registration. Experimental results show that, compared with the other three registration methods, the proposed method can maintain high registration accuracy while reducing registration time. [ABSTRACT FROM AUTHOR]

The sentence level extraction method is insufficient to solve the problem of Chinese event element dispersion. To solve this problem, this paper proposed a document level event extraction method based on context fusion. Firstly, the paper divided the document into paragraphs, and used bidirectional long and short memory network to extract sequence features of paragraphs. Secondly, the method used self-attention mechanism to capture the interaction information of paragraph context. Then the method combined the document sequence features with the interaction information to update the semantic representation. Finally, the method used sequence annotation to extract event elements and match event types. Compared with other event extraction methods on the same Chinese data set, the experimental results show that this method can effectively extract scattered event elements from documents, and improve the extraction performance of the model. [ABSTRACT FROM AUTHOR]

In order to solve the problem that the siamese neural network is sensitive to position, complex background and intra-class differences due to the use of image-level feature metrics, this paper proposed a Dual Metric Siamese Neural Network (DM-SiameseNet) . DM-SiameseNet used image-level features and local features (local descriptors) to jointly represent each image, then learned feature maps based on two different levels of similarity measures, and finally used an adaptive fusion strategy to adaptively integrate two different measurement result represented by the level feature. Experimental results showed that the accuracy of the improved model was increased by 5.04% and 9.66%respectively, and was higher than the measurement methods that only use image-level feature representation or local descriptor representation on miniImageNet, TieredImageNet, Stanford Dogs, Stanford Cars and CUB-200 datasets. The experimental results prove that the proposed model not only considers the global features of the image, but also captures more effective local feature information in the image, which improves the generalization ability of the model. [ABSTRACT FROM AUTHOR]

There are many fields choose to fuse LiDAR point cloud and camera images for applying. The accurate extrinsic calibration is a prerequisite for fusing their information. The point cloud feature extraction is a key step in extrinsic calibration. However, the low resolution and low quality of point clouds can affect the accuracy of calibration results. To address these problems, this paper proposed a LiDAR and camera extrinsic calibration method based on edge correlation point cloud. Firstly, it extracted the edge correlation point cloud of the calibration plate by using dual echoes. Then, it obtained the corner points from the edge correlation point cloud by the optimization method, where the corner points were compatible with the actual calibrated plate size. Finally, it matched the corner points in the point cloud with corner points in the image, and solved the extrinsic parameter between the LiDAR and the camera by the multi-point perspective method. The experimental results show that the reprojection error of the method is 1. 602 px, which is lower than that of comparison methods, verifying the effectiveness and accuracy of the method. [ABSTRACT FROM AUTHOR]

Federated learning is designed to solve the problem of data fragmentation and data isolation based on privacy protection in distributed machine learning. In the federated learning system, participants collaboratively train a model. Each participant uses local data to train the local model, and uploads the trained local model to the server for aggregation. In the real application environment, the data distribution between nodes is often very different, resulting in the accuracy of federated learning model is low. In order to solve the influence of non-independent identically distributed data on the accuracy of the model, this paper proposed a clustering federated learning framework by using the similarity of data distribution between different nodes. Extensive experiments were conducted on Synthetic, CIFAR-10 and FEMNIST standard datasets. Compared with other federated learning methods, clustering federated learning based on data distribution greatly improves the accuracy of the model and requires less computation. [ABSTRACT FROM AUTHOR]

Copyright of Application Research of Computers / Jisuanji Yingyong Yanjiu is the property of Application Research of Computers Edition and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

To improve the correct rate of image classification by convolutional neural network, this paper proposed a multimodel fusion convolutional neural network after research on the network structure. By extracting the output feature vectors of a single model and then fusing them, it obtained the new output feature vectors, and then set up a single classifier to classify the images, and improved the accuracy of the classification. By comparing the classification accuracy of single model and multimodel fusion, the class ification accuracy of multi-model fu sion convolutional neural network was improved. This paper analyzed the weight distribution of the last layer of the convolutional neural network, and found that the weight distribution curve of the same model on different data sets was similar and the weight distribution curve of the network model with better classification effect was more gentle. [ABSTRACT FROM AUTHOR]

In the large data environment, there are many problems in the parallel deep convolution neural network（DCNN） algorithm, such as excessive data redundancy, slow convolution layer operation and poor convergence of loss function. This paper proposed a parallel deep convolution neural network optimization algorithm based on the Im2 col method. First, the algorithm proposed a parallel feature extraction strategy based on Marr-Hildreth operator to extract target features from data as input of convolution neural network, which can effectively avoid the problem of excessive data redundancy. Secondly, the algorithm designed a parallel model training strategy based on the Im2 col method. The redundant convolution kernel is removed by designing the Mahalanobis distance center value, and the convolution layer operation speed is improved by combining the MapReduce and Im2 col methods. Finally, the algorithm proposed an improved small-batch gradient descent strategy, which eliminates the effect of abnormal data on the batch gradient and solves the problem of poor convergence of the loss function. The experimental results show that IA-PDCNNOA algorithm performs well in deep convolution neural network calculation under large data environment and is suitable for parallel DCNN model training of large datasets. [ABSTRACT FROM AUTHOR]

As the bearing fault signal is non-stationary and non-linear in nature, this paper used the wavelet packet decomposition( WPD) and auto-regressive( AR) spectrum estimating method to extract features. In order to improve the accuracy of diagnosis, this paper proposed the deep believe network to train diagnostic models. Firstly, it used the WPD and AR spectrum estimating method to calculate the energy of different signal bands to realize the feature extraction of vibration signals. Then, it used the extracted eigenvalues as input vectors of the deep belief network to conduct model training. Finally, it used the trained model to carry out bearing fault diagnosis. In order to verify the effectiveness of the proposed method, used the rotating bearing data set provided by the Case Western Reserve University in the United States, the algorithm was compared with three fault diagnosis methods. Experimental results show that the proposed method has better diagnostic performance. [ABSTRACT FROM AUTHOR]

This paper introduced the t-SNE.a manifold learning algorithm.into the field of bionic olfaction,and proposed a method of odor classification and identification using t-SNE( t-distributed stochastic neighbor embedding) +LDA( linear discriminant analysis) technique. Odorant characteristic data were collected from PEN3 electronic nose. Then the method reduced these nonlinear and high-dimensional odorant response data to low-dimensional space by using t-SNE. Finally.it used LDA for classification and recognition in the low dimensional data. This paper implemented comparative experiments (t-SNE, PCA + LDA and t-SNE + LDA) of the classification and identification for five different ingredient textile material. The result shows that t-SNE + LDA has higher ability of distinguishing and recognizing common textile ingredient materials and according to the output t-SNE + LDA has smaller within-class scatter and larger between-class scatter. Consequently,t-SNE + LDA is supposed to be a new method for odor classification and identification in bionic olfaction. [ABSTRACT FROM AUTHOR]

This paper designed and implemented the novel interactive solutions based on the touch screen of the smartphone and the built-in sensor considering the interaction of images as an information carrier's demanding to the environments. First, gesture information were acquired through smartphone built-in MEMS sensors and touch screens. Then the gestures information (angular velocity, acceleration ,touch screen data) were extracted in real time. The kinetic feature of gestures were analyzed. Meanwhile, the multivariate decision trees classifier was used to classify and identify gestures. Finally, this paper provided a prototype system. The experimental results show that the proposed method can meet the requirements of real-time human-computer interaction. [ABSTRACT FROM AUTHOR]

To deal with the problem of signal identification in satellite communication, this paper developed a new identification algorithm based on bispectrum cascade feature. It could obtain signal feature by analyzing the diagonal slice of the spectrum. By using Chirp-Z transform, the algorithm enlarged the diagonal slice of the spectrum from the frequency domain to the Z-plane. It extracted bispectrum cascade feature according to the extended Bhattacharyya distance measurement and obtained the most powerful detachable features as the feature parameters. By support vector machine (SVM), this paper simulated the processing of signal identification. The results show that the method is suitable for different types of communication signals. The algorithm is not sensitive to noise and can achieve high identification accuracy. [ABSTRACT FROM AUTHOR]

In order to improve the speed and precision of the content-based image retrieval system， this paper proposed a new convolutional encoding image retrieval method based on t-SNE. This paper firstly employed a high precision convolutional neural network model to extract image features from different layers， and then selected the optimal features by quantitative analysis of retrieval performance. Secondly， it adopted the t-SNE method to encode the best features of the network， and this step could reduce the feature dimension and remove the noise of the image features. Finally， it used the encoding features to build a content-based image retrieval system. The experimental results demonstrate that with the decrease of feature dimension， this method not only does not reduce the retrieval precision， but also can improve the retrieval accuracy in some cases. The retrieval accuracy of 16 dimensional convolutional encoding features can surpass the traditional method of 128 dimensional encoding features. Once the feature dimension is 8 times lower， the storage space will decrease 8 times and the retrieval efficiency will be greatly improved. Therefore， this method can effectively improve the speed and precision of content-based image retrieval system. [ABSTRACT FROM AUTHOR]

Because eigen-analysis method can’t work when the input sequence is too long or environment is nonstationary for blind estimate of PN sequence synchronous direct sequence code-division multiple access (DS-CDMA) signals with different power level. This paper introduced an on-line unsupervised learning neural network (NN) for adaptive feature extraction via principal component analysis (LEAP) method. Firstly,it chose the period segmented DS-CDMA signals as NN input and used the symlxil function of each weight vector to represent the PN sequence of each user. Then through the continuous input signal. it trained the weight vectors of NN repeatedly until convergence. Finally. the PN sequence of each user could he rebuilt by the symbolic function of each weight vector. Furthermore, this paper introduced variable step to improve the convergence rate. Theoretical analysis and simulation experiments show that LEAP NN method is efficiency in condition with -20 dB signal to noise rations (SNR) and 10 users of synchronous DS-CDMA signals with different power level. Besides, its convergence rate is better than traditional Sanger NN method. [ABSTRACT FROM AUTHOR]

Capsule network emphasizes the encoding of image features' spatial relationship, while its feature extraction module has difficulty in handling complex scenarios. In order to improve the performance of CapsNet, this paper proposed a capsule network with a self-attention feature extraction module ( SA-CapsNet) to solve the above problem. At first, it improved CapsNet by reducing the dimension of the capsule and adding an intermediate layer. Then mapped the SA module to the feature extraction layer of the capsule network to increase feature extraction capability. The results worked on the MNIST, Fashion MNIST, and CIFARlO datasets show that the classification accuracy rates are 99. 67%, 92. 21 %, and 82. 51 %, respectively. Additionally, results demonstrate the effectiveness of this proposed network, and the overall performance has been dramatically improved compared with CapsNet [ABSTRACT FROM AUTHOR]

The identification of encryption algorithms is the primary task of cryptanalysis. The current identification schemes cover the identification of block ciphers, the identification of stream ciphers, and the identification of hybrid algorithms of sym-metric ciphers and public key ciphers. With the popularization of Chinese national cryptographic algorithms, their security has also attracted much attention. For the national secret algorithm, this paper proposed the SM4 block cipher system identification based on random characteristics, and identified the national secret SM4 algorithm with the main international standard block cipher algorithm. It used the randomness test method to perform feature extraction on the ciphertext to obtain feature vectors. These feature vectors formed the feature space. Finally, it used machine learning algorithms to train and test the feature space. Experimental results show that the recognition rate of SM4 and other block cipher algorithms are over 90%. This work provides new possibilities for cryptanalysis and at the same time increases the diversity of password identification schemes. [ABSTRACT FROM AUTHOR]

Aiming at the problems of low accuracy and poor robustness of existing LiDAR odometry methods in outdoor dynamic road scene, this paper proposed a scheme of odometry and mapping based on 3D LiDAR and MEMS IMU. The scheme was consist of four sequential module, namely data preprocessing module,feature extraction considering dynamic obstacles module, laser odometry module and laser mapping module. By using the KITTI data and real scene data, it carried out the qualitative and quantitative accuracy evaluation of the improved scheme and the current popular LOAM and Lego-LOAM schemes. The test results show that the accuracy of the improved scheme is better than LOAM scheme under the real-time requirements of the system, which is equivalent to Lego-LOAM scheme. [ABSTRACT FROM AUTHOR]

This paper introduced the actuality and the developing level of spontaneous facial expression recognition at the present time, and paid particular attention to the key technology on the research of spontaneous facial expression recognition. This paper aimed to arouse researchers' attention and interests into this new field, to participate in the study of the spontaneous facial expression recognition problems actively, and to achieve more successes correlated to this problem. [ABSTRACT FROM AUTHOR]

LGS algorithm extracts information from far pixels. It assigns weights without considering distance between surrounding pixels and center pixel. Thus, this paper proposed WCLGS. The algorithm defined a hybrid feature extract strategy. It constructed graph structure from four directions of central pixel. The structure caught symmetric and asymmetric information in vertical direction and diagonal direction respectively. WCLGS gave a larger weight to edges close to central pixel, and a smaller weight to edges far from central pixel. It paid close attention to differences of near neighbor elements and center pixel by extracting closer pixel information and reasonable weighting strategy. Information extracted on both sides of center point was more uniform and sufficient. Compared with existing local graph structure algorithms, WCLGS has better recognition rate and performance in ORL, AR and HD infrared face databases. [ABSTRACT FROM AUTHOR]

Knowledge graph is a useful tool when introducing it into the recommendation system as auxiliary information. It can effectively enhance the learning ability of the recommendation system, improving the system' s accuracy and user' s satisfaction. Aiming at the problem of the preference propagation on the knowledge graph, this paper proposed an entity recommendation model based on the user preference propagation of the knowledge graph. This model took the transmission intensity into consideration, while propagated the preference at the same time, thus improved the final effect of recommendation. It controlled the propagation intensity of user's preference on the knowledge graph by extracting the basic characteristics of different specific attributes, and iteratively calculated the historical preference data of each user to obtain the preference propagation model of user-item pair. Later, employing the sorting learning algorithm to get the top N recommendations. In the end, comparison experiments on three different kinds of datasets verified the effectiveness of the proposed model. This study shows that controlling the propagation intensity during the propagating process can significantly improve the accuracy rate, recall rate,as well as the F, value of the recommendation system, and this method also has strong flexibility and interpretability. [ABSTRACT FROM AUTHOR]

At present, most intrusion detection algorithm cannot achieve a good balance between intrusion detection rate and false positive rate, in order to effectively avoid such problems, this paper proposed an intrusion detection method based on ADEN. The method first initialized the parameters of the encoder part in the ADEN model by training the deep belief network, and initialized the parameters of the decoder part by using the normal distribution. Then it tuned the parameters of the asymmetric deep belief network model by calculating the reconstruction error, so that the model can obtain the optimal low-dimensional representation of the original data. Finally, it used the data obtained by the encoder as input data of the classifier and detected. The ADEN model can extract features that are more conducive to classification and save more test time in the model initialization phase. The experimental results show that the method can achieve better detection performance and achieve better detection accuracy for small categories of samples. [ABSTRACT FROM AUTHOR]

Face recognition has a wide range of applications, but it is vulnerable to spoofing attacks that affect security. How to design a method with high detection accuracy, strong generalization ability, and satisfying real-time requirements is the focus of current research. This paper divided the existing methods of face anti-spoofing into methods based on hand-crafted descriptor, deep learning and fusion strategy, and introduced the basic ideas, implementation steps and advantages and disadvantages of typical algorithms. Finally, it collated and explained the published databases of face anti-spoofing, summarized the development trend of face anti-spoofing and the problems that still further needed to be solved, which could provide reference for the future research of face anti-spoofing. [ABSTRACT FROM AUTHOR]

Deep learning is hard to train with industrial product samples due to their features are non-obvious and amount is extremely rare. This paper proposed a neighboring difference filtering generative adversarial network ( NDF-GAN) _The combination of NDF-GAN extracted more features from real data and generated more samples, which could augment the real data. Experimental results reveal that, the proposed method can effectively provide more new samples to train the classifier and improve classification performance. In conclusion,NDF-GAN can augment industrial product samples . [ABSTRACT FROM AUTHOR]

Aiming at the problem that the feature cascading method for face feature extraction based on Gabor wavelet amplitude and phase makes the feature vector dimension higher, this paper proposed an improved Gabor wavelet transform feature extraction algorithm. The algorithm calculated local amplitude and local phase features, and enhanced the local correlation of each pixel. Then the algorithm weighted fusion for amplitude and phase features by weighting coefficients which experiment selected. Experimental results show that the proposed algorithm reduces the dimension of the feature vector and improves the final face recognition rate compared with the pre improved algorithm. [ABSTRACT FROM AUTHOR]

Aiming at the problem that it is difficult for the traditional fault diagnosis method to extract effective features from QAR data, this paper proposed a dual-channel fusion model called CNN-LSTM, which combined the CNN and the LSTM. Respectively as two channels, it fused CNN and LSTM through the attention mechanism to make the model be able to simultaneously express the features of the data in both space dimension and time dimension. And it also verified the validity of the feature extraction of the fusion model through time series prediction. Results of the experiment show that when compared with single CNN or LSTM, the dual-channel fusion model can extract data features more effectively, make the errors of both single-step prediction and multi-step prediction reduce by an average of 35.3%. It provides a new research idea for fault diagnosis based on QAR data. [ABSTRACT FROM AUTHOR]

The defect detection of date code relies mainly upon manual inspection. Due to the high complexity of time and the low accuracy, the existing algorithms for detecting the defect of codes can not obtain a wide range of applications in industry.For such a situation, this paper proposed an algorithm named PRICP based on iterative registration. This algorithm first abstracted the date code as a point feature, and then registered the template and feature points to achieve the purpose of detecting the defect. Machine learning as an upgrade auxiliary function, it improved its accuracy, robustness and defect classification ability. This algorithm successfully applied to the inspection of the date code of Tsingtao. The operation results demonstrate that the algorithm insensitive to constraint conditions, background and noise is superior to the traditional method in terms of robustness and efficiency, can detect the defect of information codes of multiple connected domains fastly and be directly applied to other visual field. [ABSTRACT FROM AUTHOR]

This paper proposed an algorithm of extracting feature points based on multiple criterions, which aimed to extract boundary feature points and sharp feature points efficiently. Firstly, the algorithm built the point cloud topological structure based on a modified k-d tree approach to search for the K-nearest neighborhood of the sample point. Then, according to each K-nearest neighborhood points, it calculated three feature parameters based on vector angle criteria, kernel density criteria and field power criteria. Finally, according to the three parameters.it obtained feature discriminant parameters and the global fixed threshold. It recognized a point as the feature point when its value of discriminant parameter was bigger than the threshold. The experimental results show that the proposed algorithm can extract boundary points and sharp feature points effectively. [ABSTRACT FROM AUTHOR]

Given the sensitivity of human face angle, expression and attitude as well as the low recognition precision, this paper presented a new face recognition algorithm based on distance optimization method. There were two revised points in the proposed method: a) The algorithm used the LBP operator to extract the texture map of the face image and then converged it with the R, G and B channels of the original image, then the neural network could take the fused image matrix as input, which enriched the human face texture features; b) In the processing of training, it reconstructed the loss function to improve the performance, made use of the thresold and margin to constrain the distance of the feature vector, and built a new optimization target for the model, which could limit the faces of the same person have small euclidean distances and faces of distinct people have large distances. Experiments on the unconstrained scenes of the LFW face database show that the accuracy of the model is 99. 15%respectively, indict the model can improve the accuracy of face recognition with strong robustness. [ABSTRACT FROM AUTHOR]

Multi-label learning has been widely used in many application scenarios right now. In this kind of learning problem, each instance is simultaneously assigned with more than one class label. Since different class labels might had their own unique characteristics(such as label-specific feature) which would be more useful for label classification, so some multi-label learning approaches based on label-specific features had already been proposed. Therefore, aiming at the problem that redundant feature space caused by label-specific feature construction, this paper proposed a multi-label label-specific feature extraction algorithm named LIFT_RSM, which could improve the performance of classification by comprehensively using random subspace method and the thought of pair-wise constraint dimensionality reduction to extract effective feature information in labelspecific feature space. The experimental results on several datasets show that the proposed algorithm can achieve better classification results compared with several classical multi-label algorithms. [ABSTRACT FROM AUTHOR]

Aiming to provide useful reference for dental clinical diagnosis, orthodontic, restoration, etc., this paper proposed a novel dental hard tissue segmentation algorithm according to the characteristics of the dental video images, which employed the modified marker-controlled watershed transform. Firstly, it calculated the preliminary marker areas after dental image pre-processing and color feature analysis. Secondly, the algorithm introduced the support vector machine method and the corresponding morphological operations to optimize the preliminary marker areas. Finally.it used the modified marker-controlled watershed transform to segment the video images and divided them into homogeneous regions. After calculating the average regional characteristics, it could segment the final teeth hard tissue areas from the video images. Simulation experiment results show that the algorithm has good adaptability and robustness, and can increase the segmentation precision of the dental hard tissues. [ABSTRACT FROM AUTHOR]

For the difficulties of multimodal emotion feature extraction and fusion, this paper listed some widely-used datasets for multimodal emotion recognition, and the methods of feature extraction based on facial expression features and speech features. It mainly introduced the multimodal fusion technology, and stated some levels or types of fusion and methods of fusion and compared them. Finally, it explored potential issues and future research directions. The purpose is to develop a systematic knowledge system for readers to study multimodal emotion recognition and promotes the progress of the related issues. [ABSTRACT FROM AUTHOR]

Sparse representation(SR) can encourage sparsity between the within-cluster and between-cluster. But it would be highly complex in solution and may miss the important correlation structure in the data. In order to overcome those disadvantages, this paper proposed a novel subspace algorithm discriminant least squares locality preserving projection(DLSLPP).DLSLPP used least squares regression(LSR) to catch the complete incident matrix of data. DLSLPP had low complex since it could get numerical solution. In particular, DLSLPP utilized weighted-mean method to construct representative samples different from traditional methods, thus could increase the discriminant power for classification. Experiments on four image database and several algorithms show that DLSLPP can achieve a promising performance. [ABSTRACT FROM AUTHOR]

In order to highlight the randomness, periodicity and related trend of load fluctuation, and further improve the prediction accuracy, this paper proposed a prediction method based on EMD and load fluctuation mechanism. Firstly, it decomposed the EMD of the load to obtain the random component, the periodic component and the trend component. Secondly, it analyzed the correlation between the variation rule of each component and the candidate factor. It deduced the load fluctuation mechanism, and extracted the characteristic value of the load forecasting time. Finally, it de-redundanted the feature. It extracted the novelty of this method was that the time scale characteristic of feature. Taking the load data set of Guangdong Province as a case study, the experimental results show that the proposed method is effective. [ABSTRACT FROM AUTHOR]

Traditional text categorization usually uses word embedding directly as a representation of the document, ignoring the meaning of the word in the current context, and potentially believes that the same word has the same meaning in different texts. This paper proposed a novel text classification model called WSDCNN(word sense disambiguation convolutional neural network) to solve the above problems. It used the bidirectional long short term memory network(BLSTM) for modeling context, and got the document feature map after finishing word sense disambiguation. Then it used convolutional neural network(CNN) to further extract the local characteristics of the document which was the most important for text categorization. Compared with the state-of-the-art models, the proposed model achieves excellent performance on 2 out of 4 data sets, especially to the document level data set and the other two show the same result as the previous best method. [ABSTRACT FROM AUTHOR]

Due to restrictions brought by the low-resolution system,target classification and recognition with low-resolution ra­dars have not been solved effectively for a long time. Research shows that,radar clutter,such as ground clutter,sea clutter,air clutter, etc. .has fractal characteristics,and different types of targets will produce different effects on the echo fractal characte­ristic. However,in strong clutter background .the echo fractal characteristics mainly represent the characteristics of the clutter. As a kind of non-stationary signal analysis tool, fractional Fourier transform can reveal the feature details of the target echo and fully suppress the clutter,and has the fast algorithm. Therefore .this paper intended to perform characteristic analysis on aircraft echoes from low-resolution radars in fractional Fourier domain by means of the fractal analysis of measures from the view of nonlinear analysis, estimated and analyzed their fractal parameters,and investigated the application of fractional Fourier domain echo fractal features in aircraft target classification. The analysis results show that, under the optimal transform order, aircraft echoes from low-resolution radars have obvious fractal characteristics in fractional Fourier domain,which fully reflects the cha­racteristics of aircraft targets,and the fractal analysis of measures can reveal their dynamic evolution mechanism. The classifica­tion results indicate that fractal features of aircraft echoes in fractional Fourier domain can applied to classification and recogni­tion for aircraft targets effectively. [ABSTRACT FROM AUTHOR]

Aiming to the complex type of vehicles and various brands leads to the difficulty of vehicle retrieval, this paper presented a two-level vehicle retrieval method based on depth feature coding. By improving and constructing the deep convolution network to extract the vehicle image feature, and using the two level retrieval strategy and the similarity metric function to achieve the vehicle and vehicle brand search. The retrieval results of three models of cars, trucks and buses show this method can improve the efficiency of vehicle retrieval. [ABSTRACT FROM AUTHOR]

This paper studied the Elman neural network model used in time series data predictions. It proposed an enhanced Elman model with the introduction of time weight and random factor for the improvement of prediction accuracy. In addition,it also proposed a feature selection framework as a part of the enhanced model for time series data training, the framework was able to evaluate the joint effect of multiple features. On this basis,it gave an Internet financial risk prediction model. The result indicates that the model has better accuracies in the predictions of financial time series data. [ABSTRACT FROM AUTHOR]

The background of text regions in natural scene images was complex, and the orientation of the text-line was uncertain, to address this problem, this paper proposed a multi-orientation natural scene text detection method. First the method used the color-enhanced maximally stable extremal region（C-MSER）to create character candidates for natural scene images, and used the heuristic rules and the LIBSVM classifier to eliminate non-character regions; then it designed a position-color model to retrieve the false negative candidates, and used the centers of the character regions for linear fitting to get the inclination angle of text-lines; finally, the method trained a CNN classifier to obtain the accurate results. The method got the f-score of 0.81 and 0.82 on two standard benchmark datasets(including ICDAR2011 and ICDAR2013), which demonstrates the effectiveness of the proposed multi-orientation natural scene text detection method. [ABSTRACT FROM AUTHOR]

The traditional multi-view facial expression recognition method is adopted view-specific classifier to recognition view-specific sample. This approach ignores the fact that different views of a facial expression are just different manifestations of the same facial expression. The traditional multi-view facial expression feature extraction which needs a lot of time and does not meet incremental update. To this end, this paper proposed a new method of multi-view facial expression recognition (MVFER). Firstly the MVFER system extracted regression model of incremental update feature, then it used PC A to select the feature, finally adopted discriminative shared Gaussian process latent variable model to recognition multi-view facial expression. The experiment carried on CMU-PIE database and LFPW database. The result shows the effectiveness of this method. [ABSTRACT FROM AUTHOR]

To solve the problem that the loss of image feature information and improve the accuracy of image retrieval, when the convolutional neural network (CNN) was used to extract the feature information of the image, this paper proposed an image retrieval algorithm based on improved convolutional neural network LeNet-L. First, it improved LeNet-5 convolution neural network structure, increased depth of network structure. Then, it pre-trained the deep convolutional neural network LeNet-L to extract the high-level semantic features. At last, it obtained the similar images by distance function between the image being retrieved and the one in image database. In Corel dataset, compared with the original model method and the traditional image retrieval method based on SVM and active learning, the proposed method had a higher accuracy. The experimental results show that the improved convolutional neural network has a better retrieval effect. [ABSTRACT FROM AUTHOR]

The utilization of smart phone’ acceleration sensors to identify human behavior is a big topic in intelligence field. Traditional identification methods, such as Bayes, speed learning and decision tree, must first collected acquisition frequency domain features of acceleration sensor data and extracted preferred features. This paper utilized convolutional neural network algorithm to extract smart phone’ accelerate data in three dimensions, and then automatically found out patterns from the data. Finally it used decision tree to identify human behavior from the pattern. Experimental results show that recognition accuracy improves 1.1% to 5.2% in comparison with traditional machine learning methods, especially data set on the large -scale. [ABSTRACT FROM AUTHOR]

This paper proposed an algorithm of extracting feature points based on multiple parameters hybridization method,which aimed to solve the problem that previous algorithms existed,including sharp feature points extraction were incomplete,and could not retain boundary points. Firstly,this algorithm constructed a k-d tree to establish K-nearest neighborhood of the point cloud. Then,it calculated the data point curvature,average vector angle between the point and its K-nearest neighborhood points,distance from point to its neighborhood gravity center,average distance from point to its neighborhood points for each K-nearest neighborhood. Finally,according to the four parameters,it defined the characteristic threshold and feature discriminant parameter. A point recognized as the feature point when its value of discriminant parameter was bigger than the threshold. Compared with existing algorithms,the proposed algorithm can not only extract the steep feature points,but also have the ability to identify the boundary points. [ABSTRACT FROM AUTHOR]