Search

Video captioning refers to the automatic generation of natural language sentences for a given video. There are two open problems for this task: how to effectively combine the multimodal features to better represent video content, and how to effectively extract useful features from complex visual and linguistic information to generate more detailed descriptions. Considering these two difficulties together, we propose a video captioning method named ACF-Net: Appearance-guided Content Filter Network, utilizing appearance information as a Content Filter to guide the network to aware discrimination information from both motion information and object information. Specifically, we propose a new multimodal fusion method to alleviate the problem of insufficient video information fusion. Distinguished with previous feature fusion methods that directly concatenate features, our fusion mechanism fuses relevant content information through content filters to form unified multimodal features. Moreover, a hierarchical decoder with temporal semantic aggregation is proposed, which can dynamically aggregate visual and linguistic features while generating corresponding words, focusing on the most relevant temporal and semantic information. Extensive experiments are conducted on two benchmark datasets, including MSVD and MSR-VTT, which demonstrate the effectiveness of our proposed method. [ABSTRACT FROM AUTHOR]

Anomaly detection in surveillance videos aiming to locate anomalous events is a very challenging task, since when training a detector, only normal video samples are available. And thus, most existing approaches address this problem in a semi-supervised way by either predicting or reconstructing the video frames and then compute anomaly scores by comparing the generated frames with reference frames. However, reconstruction-based methods usually lead to mis-detection due to the excessively powerful reconstruction abilities yet the incapable capturing of temporal information, while prediction-based methods are able to perceive temporal information but insufficient to produce realistic future frames. To overcome these problems, we propose a novel Mutuality-Oriented Reconstruction and Prediction Hybrid Network (MORPH-Net) for detecting anomalous events. In the MORPH-Net, a new Mutuality-oriented Training (MO-Training) mechanism is introduced to better combine the advantages of prediction-based models and reconstruction-based models. Compared to traditional single training mechanism or simple fusion mechanism, the MO-Training mechanism can prompt the generator module to produce temporally discriminative and realistic frames which benefit the anomaly detection. The experiments evaluated on three large-scale video surveillance datasets show the efficacy of our method compared with the state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Rotating machines are one of the most common equipments in modern industry, effective fault detection and diagnosis methods are vital to equipment health monitoring. In industrial production, the known information of fault types is insufficient generally, especially for constructing complex equipment and components. In previous studies of equipment fault detection, accurate fault classification and diagnosis methods have been presented, while seldom takes the condition of paucity of fault data into account. Therefore, this paper presents a novel antibody population optimization based artificial immune system (APO-AIS) for rotating equipment anomaly detection. The proposed approach can detect abnormal events while monitoring the operating condition. Meanwhile, an antigen-based antibody selecting method, a density-based antibody screening method and an optimized judgment rule based on individual difference are presented for improving the iteration evolution. The presented methods and optimized judgment rule enhance the robustness and reduces training burden for the proposed approach, which leads to accurate anomaly detection in strong background noise and in practical industrial environment. The effectiveness and robustness of the proposed method has been proven experimentally by bearing fault diagnosing and centrifugal pump condition monitoring in this paper. [ABSTRACT FROM AUTHOR]

Multi-target tracking in complex scenes is challenging because target appearance features generate partial or significant variations frequently. In order to solve the problem, we propose a multi-target tracking method with hierarchical data association using main-parts and spatial-temporal feature models. In our tracking framework, target feature models and tracklets are initialized when the new targets appear. Main-parts feature model is presented to represent target with partial or no appearance variations. It is established by partitioning a target template into several parts and formulating appearance variation densities of these parts. For the target with significant appearance variations, the tracker learns its global spatial-temporal feature model by integrating appearance with histogram of optical flow features. During tracking, tracklet confidence is exploited to implement hierarchical data association. According to different tracklet confidence values, main-parts and global data association are respectively performed by employing main-parts and spatial-temporal feature models. As a result, our approach uses the Hungarian algorithm to obtain optimal associated pairs between target tracklets and detections. Finally, target feature models and tracklets are updated by the association detections for subsequently tracking. Experiments conducted on CAVIAR, Parking Lot and MOT15 datasets verify the effectiveness and improvement of our multi-target tracking method. [ABSTRACT FROM AUTHOR]

Rolling-element bearings (REBs) faults are one of the most common breakdowns of rotating machines, thus proposing effective bearing fault diagnosis and classification methods is vital. In previous studies, lots of bearing fault classification methods have been proposed to solve the problem in low signal-to-noise ratio (SNR) conditions. Though satisfactory classification results have been obtained, in consideration of the practicability and application scenarios, there are still many aspects to improve, such as the complexity of method and the classification ability in lower SNR conditions. Therefore, this paper presents a novel method that combines lower-order moment spectrum with support vector machine (SVM) for bearing fault classification in low SNR conditions. The lower-order moment spectrum reduces influence of Gaussian noise and enhances the quality of fault feature. A bandpass filter group (BPFG) has been used to reduce the dimension of the lower-order moment spectra (LOMS) as feature vectors. And a following SVM has been applied as the fault classifier, due to the mature application and satisfactory performance in fault classification. The proposed method is demonstrated to have strong ability of classification in low SNR conditions experimentally. [ABSTRACT FROM AUTHOR]

To the question of feature selection and multi-feature fusion in facial expression recognition, a novel fusion model is proposed in this paper based on evidence theory and local feature operator. First, the facial image is divided into several regions with significant recognition features, and the Local Binary Pattern (LBP) textural features of the regions are extracted. Then, the LBP histograms in the local regions are connected into a single histogram list, and Chi-square distance is used as the similarity measure to establish the guidelines for evidence synthesis. Finally, the Dempster-Shafer evidence inference theory (D-S evidence theory) is adopted to accomplish the feature vector fusion of all components and the class judgment of facial expression is performed. Experiment shows that the method is simple and effective, which has a high recognition rate and can improve the performance of the facial expression recognition system to some extent. [ABSTRACT FROM AUTHOR]

Function packet sets (P-sets) is a novel model with law and dynamic characteristics, which is composed of function internal P-set $$S^{{\bar{F}}}$$ and function outer P-set $$S^{F}$$ . Using the structure, law and dynamic characteristic of function P-sets, we present the concepts of P-information law and band information law and their generation. Subsequently, the attribute theorem of P-information, the hiding and hiding theorem of P-information in band information law and the recovery theorem of information law are studied. Based on the above theory results, an application example of information law hiding is given and validated. [ABSTRACT FROM AUTHOR]

Gives information on proposed methods for improving the wettability of liquid metals on solids. Inclusion of physical and chemical methods; What is the advantage of direct current; Detail on experiments.

Medical image captioning has been recently attracting the attention of the medical community. Also, generating captions for images involving multiple organs is an even more challenging task. Therefore, any attempt toward such medical image captioning becomes the need of the hour. In recent years, the rapid developments in deep learning approaches have made them an effective option for the analysis of medical images and automatic report generation. But analyzing medical images that are scarce and limited is hard, and it is difficult even with machine learning approaches. The concept of transfer learning can be employed in such applications that suffer from insufficient training data. This paper presents an approach to develop a medical image captioning model based on a deep recurrent architecture that combines Multi Level Transfer Learning (MLTL) framework with a Long Short-Term-Memory (LSTM) model. A basic MLTL framework with three models is designed to detect and classify very limited datasets, using the knowledge acquired from easily available datasets. The first model for the source domain uses the abundantly available non-medical images and learns the generalized features. The acquired knowledge is then transferred to the second model for the intermediate and auxiliary domain, which is related to the target domain. This information is then used for the final target domain, which consists of medical datasets that are very limited in nature. Therefore, the knowledge learned from a non-medical source domain is transferred to improve the learning in the target domain that deals with medical images. Then, a novel LSTM model, which is used for sequence generation and machine translation, is proposed to generate captions for the given medical image from the MLTL framework. To improve the captioning of the target sentence further, an enhanced multi-input Convolutional Neural Network (CNN) model along with feature extraction techniques is proposed. This enhanced multi-input CNN model extracts the most important features of an image that help in generating a more precise and detailed caption of the medical image. Experimental results show that the proposed model performs well with an accuracy of 96.90%, with BLEU score of 76.9%, even with very limited datasets, when compared to the work reported in literature. [ABSTRACT FROM AUTHOR]

Equalization can compensate channel distortion caused by channel multipath effects, and effectively improve convergent of modulation constellation diagram in optical wireless system. In this paper, the subspace blind equalization algorithm is used to preprocess M-ary phase shift keying (MPSK) subcarrier modulation signal in receiver. Mountain clustering is adopted to get the clustering centers of MPSK modulation constellation diagram, and the modulation order is automatically identified through the k-nearest neighbor (KNN) classifier. The experiment has been done under four different weather conditions. Experimental results show that the convergent of constellation diagram is improved effectively after using the subspace blind equalization algorithm, which means that the accuracy of modulation recognition is increased. The correct recognition rate of 16PSK can be up to 85% in any kind of weather condition which is mentioned in paper. Meanwhile, the correct recognition rate is the highest in cloudy and the lowest in heavy rain condition. [ABSTRACT FROM AUTHOR]

In the technique of video multi-target tracking, the common particle filter can not deal well with uncertain relations among multiple targets. To solve this problem, many researchers use data association method to reduce the multi-target uncertainty. However, the traditional data association method is difficult to track accurately when the target is occluded. To remove the occlusion in the video, combined with the theory of data association, this paper adopts the probabilistic graphical model for multi-target modeling and analysis of the targets relationship in the particle filter framework. Experimental results show that the proposed algorithm can solve the occlusion problem better compared with the traditional algorithm. [ABSTRACT FROM AUTHOR]