Discriminative correlation filter‐ (DCF‐) based trackers are computationally efficient and achieve excellent tracking in challenging applications. However, most of them suffer low accuracy and robustness due to the lack of diversity information extracted from a single type of spectral image (visible spectrum). Fusion of visible and infrared imaging sensors, one of the typical multisensor cooperation, provides complementarily useful features and consistently helps recognize the target from the background efficiently in visual tracking. Therefore, this paper proposes a discriminative fusion correlation learning model to improve DCF‐based tracking performance by efficiently combining multiple features from visible and infrared images. Fusion learning filters are extracted via late fusion with early estimation, in which the performances of the filters are weighted to improve the flexibility of fusion. Moreover, the proposed discriminative filter selection model considers the surrounding background information in order to increase the discriminability of the template filters so as to improve model learning. Extensive experiments showed that the proposed method achieves superior performances in challenging visible and infrared tracking tasks.

Domain adaptation indeed promotes the progress of intelligent fault diagnosis in industrial scenarios. The abundant labeled samples are not necessary. The identical distribution between the training and testing datasets is not any more the prerequisite for intelligent fault diagnosis working. However, two issues arise subsequently: Feature learning in domain adaptation framework tends to be biased to the source domain, and unreliable pseudolabeling seriously impacts on the conditional domain adaptation. In this article, a new domain adaptation approach with self-supervised learning and feature clustering (DASSL-FC) is proposed, trying to alleviate the issues by unbiased feature learning and pseudolabels updating strategy. Taking different transformation methods as pretext, the transformed data and its pretext train a neural network in an SSL way. As to pseudolabeling, clusters are taken as the auxiliary information to correct the network predicted labels in terms of the “strong cluster” rule. Then, the updated pseudolabels and their confidence are enforced to further estimate the conditional distribution discrepancy and its confidence weight. To verify the effectiveness of the proposed method, the experiments are implemented on intraplatform and interplatforms for simulating the practical scenarios.

With the application of artificial intelligence in modern industry, the demand for intelligent fault diagnosis systems is increasing, which has attracted much attention of researchers. There is a popular methodology to use the knowledge or rules defined by experts for fault diagnosis. However, the expert defined information that needs to be manually annotated for faults by human with expertise knowledge costs too expensive in real industrial scenarios and is not always practicable to unknown fault category. In order to realize the zero-shot intelligent fault diagnosis and extend to a more general case, we propose an intelligent fault diagnosis system in this paper, using the statistical attributes automatically extracted from the vibration signal-self. To obtain the statistical attributes, the signals are decomposed by Ensemble Empirical Mode Decomposition (EEMD) to get their detailed time and frequency characteristics. And the statistical description of the time and frequency domains is taken as the attributes to relate the unseen categories with the seen ones. With the attributes, Gaussian models of the seen categories are established. And those of the unseen category are estimated by the attribute strength relation between the seen and unseen categories. The unseen category is finally predicted by Maximum A Posterior probability (MAP). The experiments on fault dataset show that the proposed system can obtain well recognition performance compared with the classic methods. Among them, the accuracy of zero-shot fault diagnosis can approach to 86%. And the performance can be further improved through attribute selection.

The automatic feature extraction capability of deep learning has led to its extensive usage in fault diagnosis applications. In engineering scenarios where the distribution between training and test sets is inconsistent, deep domain adaptation methods are commonly employed to solve cross-domain fault diagnosis problems. Despite achieving good performance for cross-domain diagnosis, there are some limitations to domain adaptation models. Firstly, most existing research has only focused on domain alignment between source and target domains while neglecting class information, which can result in incorrect alignment between classes of the two domains. Secondly, target samples that are distributed close to the boundaries of the clusters are easily misclassified by the classification decision boundary learned from the source domain. To address these issues, joint discriminative adversarial domain adaptation (JDADA) is proposed in this paper. The proposed method combines domain alignment and class alignment by introducing a class alignment module into the domain adversarial network. Furthermore, the discriminative discrepancy module is proposed to compact features of the same class and separate features of different classes to extract more discriminative features. Additionally, we propose a new pseudo-labelling strategy to address the problem of target training samples without labels. The proposed method is evaluated on the gearbox data set and bearing data set, and the results demonstrate its effectiveness and superiority over state-of-the-art domain adaptation methods. Specifically, JDADA achieves up to 5.0% accuracy improvement on the gearbox data set and 3.4% accuracy improvement on the bearing data set.

Emotion Recognition in Conversation (ERC) has gained considerable attention due to its importance in human-computer interaction. In the ERC task, the combination of multimodal information and contextual information is necessary since it can help the model understand the emotional changes in the context from multiple perspectives. As Graph Neural Networks (GNNs) have shown the superiority in relation modeling, many graph-based methods have been proposed to improve the performance of emotion recognition by utilizing the edges to mine the contextual relationship and multimodal relationship in a conversation. However, the existence of numerous redundant edges and excessively complex modality interaction in the graph hinders the model from capturing the truly effective dependency information for emotion recognition. In this paper, we propose a Hypergraph based Contextual Relationship Modeling Method (HyperCRM) to carry out the ERC task. HyperCRM models a conversation as a hypergraph instead of a graph, which defines two types of hyperedges, namely speaker-level hyperedge and sequence-level hyperedge, to represent the contextual relationship within the same speaker and the local sequence of the conversation, respectively. Multimodal information is leveraged here as the node feature representation by the feature concatenation. In addition, an improved hypergraph convolution method is designed to capture the long-range contextual information by the three-stage information propagation in the hypergraph, including node-hyperedge, hyperedge-hyperedge and node-hyperedge. The extensive experiments on two public datasets shows the new State-Of-The-Art (SOTA) results, to further demonstrate that the proposed method can simply make use of the multimodal information and effectively model the complex contextual relationships in the conversation.

Human care services, as one of the classical Internet of things applications, enable various kinds of things to connect with each other through wireless sensor networks (WSNs). Owing to the lack of physical defense devices, data exchanged through WSNs such as personal information is exposed to malicious attacks. Therefore, intrusion detection is urgently needed to actively defend against such attacks. Intrusion detection as a data mining procedure cannot control the size of rule sets and distinguish the similarity between normal and intrusion network behaviors. Therefore, in this paper, an evolving mechanism is introduced to extract the rules for intrusion detection. To extract diversified rules as well as control the quantity of rulesets, the extracted rules are examined according to the distance between the rules in the rule set of the same class and the rules in the rule set of different classes. Thereby, it alleviates the problem that the quantity of rules expands unexpectedly with the evolving genetic network programming. The simulations are conducted on a benchmark intrusion dataset, and the results show that the proposed method provides an effective solution to evolve the class association rules and improves the intrusion detection performance.