DEtection TRansformer (DETR) marked a significant advancement in transformer-based object detection, aiming to modernize the field with a sophisticated approach. However, its computational efficiency lagged, a challenge DeformableDETR sought to address with multi-scale deformable attention. Despite faster convergence and enhanced results, DeformableDETR deviated from the standard encoder-decoder structure, leading to a skewed emphasis on the query component and underutilization of its potential in detailed feature representation. Responding to this imbalance, we introduced EnDETR equipped with Enhanced Attention Module, designed to optimize feature representation while preserving model complexity. EnDETR, enhanced with adversarial training, demonstrated robustness against adversarial attacks and natural perturbations. Our empirical studies across diverse datasets showcased EnDETR’s superiority, outperforming DeformableDETR with a notable 2.7 increase AP on clean datasets, and a more significant 3.8 increase on distorted datasets, especially achieving a 12.7 improvement in detecting small objects. These results emphasize EnDETR’s effectiveness in complex detection scenarios, establishing it as a potent, efficient solution in transformer-based object detection.

Digital Capital and its economic implications have gained significant attention in digital transformation era. In rural areas, external shocks can easily lead to widespread income fluctuation, making the enhancement of income resilience a central long-term task for rural development. This study addresses the critical issue of improving rural resident’s income resilience against external shocks in digital transformation by exploring the mechanism and effect of Digital Capital on rural resident’s income-increasing rate through the mediation effect of individual dynamic capabilities by the method of structural equation model (SEM). The Digital Capital and individual dynamic capabilities of rural residents were developed through a confirmatory factor analysis (CFA) and validated with a representative sample survey of 289 rural residents in Yantai City, China. The theoretical and empirical contributions to the literature lie in that: (1) clarifying the quantitative relationships among Digital Capital, rural resident’s income-increasing rate (concrete benefits in the offline arena) and individual dynamic capabilities. (2) exploring the mechanism by which Digital Capital operates on the specific outcomes. The findings indicate that rural resident's Digital Capital has a direct impact on their individual dynamic capabilities, and can influence rural resident's income-increasing rate from the perspective of total effect; Furthermore, the analysis demonstrates individual dynamic capabilities fully mediate the relationship between Digital Capital and the income-increasing rate of rural residents.

The excellent model obtained through natural data training in deep learning is easily tampered with by adversarial examples. After discovering that, adversarial training has become the best way to defend against adversarial attacks and improve the robustness of the model. Since it is expensive to frequently calculate adversarial examples in each epoch during the training process, most people prefer to choose a single-step adversarial training method. However, the single-step adversarial training method will cause catastrophic overfitting and make the model lose robustness forever. In this paper, we explain adversarial training from the perspective of data augmentation, using artificial binary data to explore the reason for the occurrence of this overfitting. We propose two methods, VFSAT(Various fixed-stepsize single-step adversarial training) and GradSum, to prevent the overfitting in term of local smoothing and improve the robustness of the model obtained by single-step adversarial training. Simultaneously, experiments on CIFAR-10 and Tiny ImageNet datasets were constructed and the proof that single-step adversarial training could also resist multi-step adversarial attacks was derived.

SysML state machine (SysML-STM) is a modeling tool used in the Systems Modeling Language (SysML) to describe the behavior of a system. It is widely used in model-driven development (MDD). Formal methods are mathematical techniques to ensure the correctness, reliability and safety of software systems and hardware designs. In this paper, we introduce formal methods into MDD by transforming a SysML-STM model into a Uppaal timed automata. By formally verifying the system at an early stage of the development life-cycle, we aim to enhance the system's robustness. We design the mapping rules between the two models and have developed a tool, STMTU, to transform them directly. Our tool effectively leverage the benefits of formal verification techniques to ensure the correctness and reliability of the system. And the direct transformation of these models not only reduces the learning cost for developers but also helps to promote the wider adoption of formal methods.

In research of uncertain data stream in sensor network, the probability density is usually used to describe the uncertainty of continuous uncertain object values. Current researches are mainly based on the simply assumption that the values of uncertain object meets some conventional distribution, such as Gaussian distribution. However, the statistical distribution of sensor data streams cannot be described accurately in many cases. In this paper, we propose an algorithm for probability density fitting about continuous uncertain sensor data streams, which based on GMM. Experiments show that this algorithm can meet the time requirement of actual sensing application and improve the fitting effect on the probability density of the value uncertain object.

Most face hallucination methods are usually limited to frontal face with small pose variations. This letter presents a simple and efficient multiview face hallucination (MFH) method to generate high-resolution (HR) multiview faces from a single given low-resolution (LR) one. The problem is addressed in two steps. A simple face transformation method is proposed by defining a constrained least square problem for LR multiview face transformation and a position-patch based face hallucination method is extended to incorporate HR multiview face details. Experimental results show that our approach has some advantages over existing MFH methods.

A full frequency diversity COFDM (FD-COFDM) scheme based on sub-channel multiplexing is proposed to improve the performance of COFDM without sacrificing spectral efficiency. Since the channel suffers the frequency selective fading, the received signal contains strong inter-symbol interference (ISI). To obtain the diversity- and code-gains, an iterative receiving algorithm is presented. By jointly using diversity combining, interference cancellation and soft decoding, the proposed iterative receiving algorithm properly decodes the information bits in the strong ISI environment. Consequently, all the subcarriers of FD-COFDM participate in diversity so that the system can achieve full frequency diversity in most practical channels. The new receiving algorithm is relatively simple compared to the maximum likelihood estimation algorithm, and is therefore a better alternative to balance the algorithm complexity and system performance.

In this paper, a multiuser detection (MUD) scheme employing joint adaptive filter and space-time signal processing is discussed and experimentally investigated for direct-sequence (DS) ultra-wideband (UWB) multiple-input multiple-output (MIMO) systems. In dense multipath environments, by taking advantage of the space-time filter structure in conjunction with simple transmit and/or receive diversity approaches, the investigated MUD scheme is capable of significantly improving the performance in combating multiuser interference (MUI) compared with the temporal MUD method based on single-antenna systems. The effects of transmit and/or receive diversity are also examined via the Monte-Carlo simulations.