Diffusion models (DMs) are advanced generative models, yet recent research reveals their vulnerability to backdoor attacks, which establish hidden associations between input patterns and targeted model behavior, potentially causing malicious outputs during inference. These attacks pose significant risks, including model owner reputation damage and harmful content generation. However, existing defense methods often fail against backdoor attacks on diffusion models. To address this gap, we propose Diff-Cleanse, a two-stage defense framework. The first stage introduces a novel trigger inversion method for backdoor detection, and the second stage applies a structural pruning-based method for backdoor removal. Experiments on 373 models poisoned by three state-of-the-art attacks show that Diff-Cleanse achieves > 97% detection accuracy, completely remove backdoors and maintains the models’ benign performance. Code is available at https://github.com/shymuel/diff-cleanse.

Learning-based Multi-View Stereo (MVS) methods heavily rely on feature extraction and cost volume construction to bridge 2D semantics and 3D spatial associations. However, many recent studies have increasingly shifted focus away from these two critical steps, often adopting a multi-stage cascaded framework that can propagate errors from earlier stages. To this end, we introduce TTINS-MVSNet, a novel progressive refinement framework that integrates a one-stage MVS module enhanced by two types of Epipolar Transformer (ET) and subsequent depth optimization modules. Specifically, we incorporate an intra-view ET into context-aware feature extraction and an inter-view ET into visibility-aware cost aggregation. A significant innovation in our approach is the introduction of implicit neural super-resolution module, designed to recover finer details. Experimental results on benchmark datasets demonstrate that our method outperforms all previous approaches with similar structures, highlighting its effectiveness and generalization capability. The code will be available after publication.

We propose an instance segmentation model, named CoMask, that effectively alleviates the scale variation issue and addresses the precise localization. Specifically, we develop a multi-scale feature extraction module (MSFEM) to exploit multi-scale spatial cues. Besides, the channel attention mechanism is also adopted to further enhance the discriminating ability. Equipped with MSFEMs, multi-scale and multi-level features can be extracted to better characterize objects of various sizes and provide affluent high-level semantic information. For precise localization, we propose a collaborative learning framework to compute coarse masks and regresses position-sensitive dense offsets. The foreground confidence of each position is then assigned as the weight of the corresponding bounding box to calculate a weighted average. Thus, we can mitigate interference of background regions. After obtaining the final regressed bounding boxes, finer foreground masks can be calculated. We conduct experiments on MS COCO dataset. Experimental results validate that CoMask is competitive compared with state-of-the-art models.

Research on designing local feature descriptors has gradually shifted to deep learning. Different from other computer vision tasks, the biggest challenge for local descriptor learning lies with the formulation of loss functions. Existing methods solve the problem by leveraging Siamese loss or triplet loss and improve the performance of the learned descriptors by a significant margin. However, the widely used Siamese loss and triplet loss cannot fully utilize the context information. In this paper, we propose a novel loss function to introduce more context information to facilitate training. After incorporating the proposed loss function into training, our learned descriptor demonstrates state-of-the-art performance in patch verification, image matching and patch retrieval benchmarks. The pretrained model will be publicly available at https://github.com/clelouch/CANet.

Attention-based Encoder-Decoder has the effective architecture for neural machine translation (NMT), which typically relies on recurrent neural networks (RNN) to build the blocks that will be lately called by attentive reader during the decoding process. This design of encoder yields relatively uniform composition on source sentence, despite the gating mechanism employed in encoding RNN. On the other hand, we often hope the decoder to take pieces of source sentence at varying levels suiting its own linguistic structure: for example, we may want to take the entity name in its raw form while taking an idiom as a perfectly composed unit. Motivated by this demand, we propose Multi-channel Encoder (MCE), which enhances encoding components with different levels of composition. More specifically, in addition to the hidden state of encoding RNN, MCE takes 1) the original word embedding for raw encoding with no composition, and 2) a particular design of external memory in Neural Turing Machine NTM) for more complex composition, while all three encoding strategies are properly blended during decoding. Empirical study on Chinese-English translation shows that our model can improve by 6.52 BLEU points upon a strong open source NMT system: DL4MT1. On the WMT14 English-French task, our single shallow system achieves BLEU=38.8, comparable with the state-of-the-art deep models.

We investigate how to improve bilingual embedding which has been successfully used as a feature in phrase-based statistical machine translation (SMT). Despite bilingual embedding’s success, the contextual information, which is of critical importance to translation quality, was ignored in previous work. To employ the contextual information, we propose a simple and memory-efficient model for learning bilingual embedding, taking both the source phrase and context around the phrase into account. Bilingual translation scores generated from our proposed bilingual embedding model are used as features in our SMT system. Experimental results show that the proposed method achieves significant improvements on large-scale Chinese-English translation task.

Low voltage power line due to high noise, high attenuation and high time-varying characteristics not only the signal transmission distance was restricted, but also the reliability of communication was reduced, so the meter reading range and accuracy were affected. In this paper, typical topology structure and its characteristics of meter reading system were analyzed in detail, the ant colony genetic algorithm was proposed in this paper to solve the problem of power line carrier meter reading communication routing optimization, the signal transmission distance of meter reading system and the system invulnerability were enhanced by the dynamic routing network. The feasibility and robustness and invulnerability were tested by simulation experiments, the range of power line carrier meter reading was expanded and the reliability of power line communication was improved by algorithm.

The demand on electrical power quality is ever increasing along with the growing technological advancement. Voltage sag is the most frequently happening problem on power system and might cause great economic loss. To solve this problem, Dynamic Voltage Restorer (DVR) is the most efficient and economic device at present. Offers high-quality power is also a mark of national industrial, technological level and progress of social civilization. Therefore, it is a subject need to be solved In view of the needs of both theoretical and practical aspects. This paper based on the basic theory of DVR and the voltage sag detection algorithm and control compensation strategies, and then takes TMS320F2812 of TI Corporation as signal processer; we have a overall design of the dynamic voltage restorer. The operating experience and practice indicated: the Dynamic Voltage Restorer system has the merits, such as, the design and operation of a fast response, stable and reliable and low maintenance cost.

With the development of software technology, the new aircraft component testing platform has been widely applied in the field of aviation scientific research and design of new aircraft. As the needs of engineering applications, this paper studies two things to further improve the new aircraft component testing platform by solving the bottleneck of real-time and generality: the real-time collection and storage improvement of a large number of high-speed simulation data, the design of testing platform console's general interface. Starting from the hardware and software, this paper tests the enhancement of the acquisition and storage speed by hardware replacement and adding the buffer algorithm. Besides, the RTX real-time extension technology has also been analyzed for further real-time improvement. In the aspect of generality, this paper realizes the convenient generation and editing of console interface and function based on the establishment of MFC control class design, by taking XML as the bridge and using component-based software engineering.

Synthetic Aperture Radar (SAR) is an active microwave remote sensor, which could form real-time image around the target. Traditionally, the simulation of SAR imaging algorithm is based on Matlab which is totally software simulation and the final performance is difficult to guarantee. In this paper, the design of the seeker Equivalent Device of Missile-Borne Synthetic Aperture Radar is a device in hardware-in loop simulation of echo signal and imaging algorithm. Based on this device the target echo signal is simulated in a real environment which provides us a more reliable simulation platform for the echo signal and imaging algorithms. In practical testing, this device performs well in point target imaging simulation.

The traditional Canny edge detection method is widely used in gray image processing. However, this traditional algorithm is unable to deal with color images and the parameters in the algorithm are difficult to be determined adaptively. In this paper, an improved Canny algorithm is proposed to detect edges in color image. The proposed algorithm is composed of the following steps: quaternion weighted average filter, vector Sobel gradient computation, non-maxima suppression based on interpolation, edge detection and connection. Experimental results show that the proposed algorithm outperforms other color image edge detection methods and can be widely used in color image processing.

The on-line temperature monitoring for power cable joints can effectively prevent and reduce the accidents of cable joints. This article introduces a new type of temperature monitor and analysis system for urban power cable joints, based on GPRS data transmission. The platform has the functions of alarm and joint state assessment analysis. Power supply device was designed and developed based on supercapacitor, that is obtained power directly from the power line by tailor-made current transformer, it can solve the difficult point effectively.

A new method used in error and VI characteristics measurement of current transformer is developed based on the equivalent model of current transformer. Using this method, the up-flow unit and the standard current transformer are not necessary. And more characteristics of current transformer can be inferred. The current transformer calibrator developed based on this method can carry out rapid and precise measurement. It can be used in the calibration of current transformer according to the state regulations. It also has some expanded functions, such as drawing the VI characteristics curve.

High voltage circuit breakers play an important role in the power system. So it is necessary to implement the state detection of breakers in order to ensure stable and reliable running of the grid. The purpose of state detection is to provide reliable basis of maintenance by extracting mechanical state parameters accurately. This paper mainly focuses on the coil current signal feature extraction algorithm. To settle the problem of too much noise mixed with the current signal and signal distortion, the discrete wavelet transform algorithm is used to extract the coil current signal parameters. This paper also designs the FIR filter to extract stroke and speed parameters from travel-time waveform. The experiments show that the difference between the theoretical results and test results processed by the method in this paper is very small and the test results are able to accurately reflect operation states and mechanical features of high voltage circuit breakers.

A ceramic insulator has an excellent anti-pollution performance, but sometimes flashover occurs on it and then the outage of power network follows. This paper analyzes the formation mechanism of flashover and the result is that the RMS of leakage current(LC) and discharge pulses reflect the contamination severity approximately. This shows that to a large extent the flashover accident can be avoided by means of monitoring the leakage current value and the discharge pulses. Existing monitoring device for contaminated insulators inspects either leakage currents or discharge pulses, also, it is difficult to meet the requirements of the client for low-power consumption, low cost and high reliability. In this paper, a novel low-power and low-cost online monitoring system is presented, which is based on ARM Cortex-M and Zigbee wireless network. In addition, the monitoring system not only acquires the leakage current and discharge pulses of insulator strings, but also measures the environmental temperature and humidity and then comprehensively analyzes the severity of insulator pollution. After that the results will be displayed on the LCD or sent to the control center through the Zigbee network. Tests show that the monitoring system has good accuracy and good practical performance.