Existing text-video retrieval methods mainly focus on singlemodal video content (i.e., visual entities), often overlooking heterogeneous scene text that is ubiquitous in human environments. Although scene text in videos provides finegrained semantics for cross-modal retrieval, effectively utilizing it presents two key challenges: (1) Temporally dense scene text disrupts sync with sparse video frames, obstructing video understanding;(2) Redundant scene text and irrelevant video frames hinder the learning of discriminative temporal clues for retrieval. To address them, we propose a temporal scene-text calibrating and distilling (TCD) network for textvideo retrieval. Specifically, we first design a window-OCR captioner that aggregates dense scene text into OCR captions to facilitate feature interaction. Next, we devise a heterogeneous semantics calibration module that leverages scene text as a self-supervised signal to temporally align window-level OCR captions and frame-level video features. Further, we introduce a context-guided temporal clue distillation module to learn the complementary and relevant details between scene text and video modalities, thereby obtaining discriminative temporal clues for retrieval. Extensive experiments show that our TCD achieves state-of-the-art performance on three scene-text related benchmarks.

In cross-modal retrieval, comprehensive image understanding is vital while the scene text in images can provide fine-grained information to understand visual semantics. Current methods fail to make full use of scene text. They suffer from the semantic ambiguity of independent scene text and overlook the heterogeneous concepts in image-caption pairs. In this paper, we propose a heterogeneous prompt-guided entity inferring and distilling (HOPID) network to explore the nature connection of scene text in images and captions and learn a property-centric scene text representation. Specifically, we propose to align scene text in images and captions via heterogeneous prompt, which consists of visual and text prompt. For text prompt, we introduce the discriminative entity inferring module to reason key scene text words from captions, while visual prompt highlights the corresponding scene text in images. Furthermore, to secure a robust scene text representation, we design a perceptive entity distilling module that distills the beneficial information of scene text at a fine-grained level. Extensive experiments show that the proposed method significantly outperforms existing approaches on two public cross-modal retrieval benchmarks.

The rod-fastened rotor is a critical component of heavy-duty gas turbines, and therefore, it is imperative to investigate its failure mechanisms and develop intelligent diagnostic techniques. The issue of effective cross-domain diagnosis has garnered significant interest considering the changing operating conditions of rod-fastened rotors. To mitigate the effects of limited data samples or high collection costs in the target domain, we introduce a fault diagnosis model named the task-generalization-based graph convolutional network (TG-GCN), which leverages adversarial perturbations within a meta-learning framework to enhance the robustness of prior knowledge incorporated into the graph convolutional network. Initially, the method involves generating adversarial training samples in proximity to the distribution of the source-domain data by optimizing the worst case formulation to enhance generalization to task distribution. Next, the support set and the query set are simultaneously fed into the feature extraction network, yielding corresponding feature vectors. Subsequently, similarities between feature vectors and the degree of dispersion are computed in the task space, serving as connection weights for constructing the graph network model aimed at executing fault diagnosis. The proposed TG-GCN framework employs the introduction of data augmentation during the iteration process, which helps to strengthen the vulnerable components of the network by connecting them with the worst case perturbations. Moreover, we employed a rod-fastened rotor experimental system to elucidate the underlying failure mechanisms and corroborate the superiority of the TG-GCN.

Currently, convolutional neural networks (CNNs) have shown great potential in the field of rotating machinery fault diagnosis. To maximize accuracy, the network architecture of novel models has become complex, and the number of parameters has also increased, which makes them not economical in terms of computational efficiency. Experience has shown that there is irreversible information loss in the process of excessive downsampling. This article proposes the CNN with mixed information (MIXCNN), a classification model that is efficient and lightweight for end-to-end fault diagnosis, and the convolutional layers maintain the same size of the output throughout the network. The MIXCNN uses depthwise convolution to increase the ability of discrimination in spatial locations and uses traditional convolution to achieve cross-channel interaction of information; then, the residual connection is introduced to reduce the loss of information on the convolutional layers. By appropriately increasing the depth of the network, MIXCNN pays more attention to features that are differentiated and highly recognizable, and more spatial information is mixed by a larger convolution kernel inside the MIXCNN. Experimental validation using publicly available datasets reveals that the proposed method has higher accuracy and generalization capability than state-of-the-art fault diagnosis methods.

Deflection of the vertical (DOV) has become a significant error source in inertial navigation systems (INS) and meantime widely used in geodesy, exploration, geoid refinement et.al. Hence, the efficient and reliable method for DOV estimation is crucial for those fields. In this paper, an ingenious and efficient method, the relative estimation means based on double-alignment (RE-DA), is proposed, which mainly includes attitude tracking of normal and orientation for plumb line two processes. Firstly, the influence mechanism of DOV in initial alignment based on the inertial method was briefly introduced and the conclusion was clarified that the alignment result, attitude matrix, was with respect to the plumb line. Moreover, the method for attitude tracking of normal based on the strapdown inertial navigation system and odometer integrated navigation (SINS/OD) using the Kalman filter was investigated. Finally, the process of the pro-posed RE-DA method was specifically introduced. In verifying the pro-posed method, several repeated land-vehicle experiments were conducted, where the statistical results show that the root means square (RMS) of the estimation values are 0.73ʺ and 0.69ʺ respectively, which provides a novel means for DOV high-efficient estimation.