A5100410001- Advanced Image Processing Techniques
- Anomaly Detection Techniques and Applications
- Industrial Technology and Control Systems
- Advanced Vision and Imaging
- Image and Signal Denoising Methods
- Network Security and Intrusion Detection
- Image Processing Techniques and Applications
- Aluminum Alloys Composites Properties
- Advanced Steganography and Watermarking Techniques
- Human Pose and Action Recognition
- AI in cancer detection
- Hydraulic and Pneumatic Systems
- Digital Media Forensic Detection
- Manufacturing Process and Optimization
- Seismic Imaging and Inversion Techniques
- Simulation and Modeling Applications
- Topic Modeling
- Advanced Statistical Process Monitoring
- Bone Tissue Engineering Materials
- Additive Manufacturing and 3D Printing Technologies
- CO2 Sequestration and Geologic Interactions
- Multimodal Machine Learning Applications
- Carbon Dioxide Capture Technologies
- Advanced Algorithms and Applications
- Domain Adaptation and Few-Shot Learning
Central South University
2007-2024
Sinopec (China)
2021-2023
Beihang University
2021-2023
Zhejiang University of Science and Technology
2023
Chongqing University of Posts and Telecommunications
2023
Beijing Jiaotong University
2022
Xi'an Jiaotong University
2022
Shanghai Jiao Tong University
2021
Hainan University
2020
Peking University
2020
Conventional steganography approaches embed a secret message into carrier for concealed communication but are prone to attack by recent advanced steganalysis tools. In this paper, we propose Image DisEntanglement Autoencoder Steganography (IDEAS) as novel without embedding (SWE) technique. Instead of directly the image, our approach hides it transforming synthesised and is thus fundamentally immune typical attacks. By disentangling an image two representations structure texture, exploit...
Exploring the characterization laws of image data and improving efficiency knowledge is essential to promote development Internet Things technology. Considering that images in real world usually contain multiple objects, objects are closely dependent. For these reasons, it brings great challenges robust representation learning multilabel images. In general, researchers model relationship between based on a class activation map use graph convolution mine dependencies objects. However,...
Face images are typically a key component in the fields of security and criminal investigation. However, due to lighting shooting angles, faces taken under low-light conditions often difficult recognize. super-resolution (FSR) technology can restore high-resolution based on low-resolution inputs. existing face methods rely prior knowledge inaccurate estimated from images. Faces restored by inputs may suffer problems such as low brightness many missing details. In this article, we proposed an...
Real-world applications often require the classification of documents under situations small number features, mis-labeled and rare positive examples. This paper investigates robustness three regularized linear methods (SVM, ridge regression logistic regression) above situations. We compare these in terms their loss functions score distributions, establish connection between optimization problems generalization error bounds. Several sets controlled experiments on Reuters-21578 corpus are...
Abstract The traditional impact test method needs a large number of sensors deployed on the entire structure, which cannot meet requirements rapid bridge testing. A new mobile is proposed by sequentially testing substructures then integrating data all for flexibility identification structure. novelty that quantum‐inspired genetic algorithm (QIGA) to improve computational efficiency transforming scaling factor sign determination problem an optimization problem. Experimental example...
We present a non-uniform recursive sampling technique for multi-class scatterplots, with the specific goal of faithfully presenting relative data and class densities, while preserving major outliers in plots. Our is based on customized binary kd-tree, which leaf nodes are created by recursively subdividing underlying density map. By backtracking, we merge until they encompass points all classes our subsequently applied outlier-aware strategy. A quantitative evaluation shows that approach can...
Coverless image steganography (CIS) has attracted significant attention because it can fundamentally resist steganalysis tools. Available CIS schemes are mainly divided into synthesis-based and mapping-based schemes. Compared with the former, methods ensure lossless information extraction stronger attack robustness. However, these still face challenges of incomplete secret mapping, trade-off between robustness distinguishability mapping features, demand for huge numbers images. To tackle...
We propose a magneto-optical diffractive deep neural network (MO-D2NN). simulated several MO-D2NNs, each of which consists five hidden layers made magnetic material that contains 100 × domains with domain width 1 µm and an interlayer distance 0.7 mm. The networks demonstrate classification accuracy > 90% for the MNIST dataset when light intensity is used as measure. Moreover, 80% obtained even small Faraday rotation angle π/100 rad polarization MO-D2NN allows to be rewritten, not possible...
How to use artificial intelligence technology mine human abnormal behavior from considerable video data generated by the Internet-of-Things system has been intensively studied for a long time. Existing deep learning anomaly detection algorithms deployed in cloud typically perform supervised based on constant kinds of data. However, this model with preset categories ignores diversity and unpredictability occurrences open scenarios. Thus, we propose an algorithm as edge network service...
A railway track gradually deviates from the originally designed vertical alignment in operation. To ensure safety and comfort, maintenance department must periodically recreate calibrate existing to it. The recreated should be as close possible measured points along track, while satisfying multiple constraints. consists of tangents curves. Identifying geometric element attribution accurately recreating efficiently with all constraints satisfied are key problems. Existing methods need manual...
Spatial time series imputation is of great importance to various real-world applications. As the state-of-the-art generative models, diffusion models (e.g. CSDI) have outperformed statistical and autoregressive based in imputation. However, may introduce unstable noise owing inherent uncertainty sampling, leading generated deviating from intended Gaussian distribution. Consequently, imputed data deviate real data. To this end, we propose a Self-adaptive Scaling Diffusion Model named SaSDim...
Pneumatic suspension is the most significant subsystem for an automobile. In this paper, a simplified and novel pneumatic spring structure with only conical rubber surface presented designed to reduce influence of external factors besides pneumatic. The nonlinear stiffness analyzed based on ideal gas model material mechanics. Natural frequency analysis transmission rate are obtained as two effect criteria dynamic model. vibration isolation system platform established in both simulation...
In order to effectively improve the resolution of image and restore more edge texture details, an super-resolution reconstruction algorithm based on FSRCNN residual network is proposed. Firstly, a deep channel constructed, network, simple added estimate high-frequency information, output used as low-frequency information be combined with obtain reconstructed image. Then, distillation module fused shallow further details. Finally, outputs two channels are final Experiments results illustrate...
Reference-based image super-resolution (RefSR) aims to reconstruct high-resolution (HR) images from low-resolution (LR) by introducing HR reference images. The key step of RefSR is transfer features LR features. However, existing methods still lack an efficient mechanism, resulting in blurry details the generated image. In this article, we propose a double-layer search module and adaptive pooling fusion group for reference-based super-resolution, called DLASR. Based on re-search strategy,...
Medical zero-shot relation triplet extraction, referred to as Med-ZeroRTE, requires the model extract triplets comprising entities and relations from medical sentences. Importantly, sentences include that were unseen during model's training phase. While Med-ZeroRTE had not been formally explored before this work, limited availability of datasets, influenced by privacy concerns annotation costs, emphasizes necessity exploring Med-ZeroRTE. This exploration faces two main challenges: Firstly,...
The objective of this study is to propose a solution for process plant upgradation becoming extinct due obsoleteness spares. will help in reliability, availability, and maintainability (RAM) based control system plants developing countries. Available options are compact control, modular, semiautomatic. RAM provides which high reliability availability (usually all parts replaced with upgraded compatible technology) easy maintain throughout the service life plant. Case stacker reclaimer cement...