A5005526135- Natural Language Processing Techniques
- Speech Recognition and Synthesis
- Concrete and Cement Materials Research
- Innovative concrete reinforcement materials
- Magnesium Oxide Properties and Applications
- Structural Behavior of Reinforced Concrete
- Speech and dialogue systems
- Concrete Corrosion and Durability
- Topic Modeling
- Digital Media Forensic Detection
- Image and Signal Denoising Methods
- Advanced Steganography and Watermarking Techniques
- Neural Networks and Applications
- Advanced Image Processing Techniques
- Concrete Properties and Behavior
- Building materials and conservation
- Generative Adversarial Networks and Image Synthesis
- Advanced Neural Network Applications
- Speech and Audio Processing
- Drilling and Well Engineering
- Energetic Materials and Combustion
- Nuclear materials and radiation effects
- Thermal properties of materials
- Natural Fiber Reinforced Composites
- Vehicle Dynamics and Control Systems
Chinese Academy of Surveying and Mapping
2024
Dalian University of Technology
2011-2023
Sichuan University
2023
Apple (United Kingdom)
2021-2022
Lanzhou University of Technology
2019-2020
Chery Automobile (China)
2012
Neural image compression often faces a challenging trade-off among rate, distortion and perception. While most existing methods typically focus on either achieving high pixel-level fidelity or optimizing for perceptual metrics, we propose novel approach that simultaneously addresses both aspects fixed neural codec. Specifically, introduce plug-and-play module at the decoder side leverages latent diffusion process to transform decoded features, enhancing low quality without altering original...
This study investigated the frost resistance of fiber-reinforced concrete with different contents polyvinyl alcohol (PVA) fiber under freeze–thaw cycles, and to chloride penetration. The cycle experiment was carried out using rapid method. Mass loss, relative dynamic modulus elasticity, compressive strength, flexural strength were investigated, influence ion content diffusion coefficient after cycles analyzed. results showed that PVA can effectively reduce mass elastic loss improve strength....
Neural text-to-speech (TTS) synthesis can generate speech that is indistinguishable from natural speech. However, the synthetic often represents average prosodic style of database instead having more versatile variation. Moreover, many models lack ability to control output prosody, which does not allow for different styles same text input. In this work, we train a non-autoregressive parallel neural TTS front-end model hierarchically conditioned on both coarse and fine-grained acoustic...
Due to special climatic and environmental factors in the cold regions of north China, degradation concrete durability caused by freeze–thaw cycles has been a widespread concern among researchers. Freeze–thaw damage seriously affected safe use long-term operation structures. In this study, steel fiber polypropylene double-blended were selected for an investigation on law mass loss relative dynamic elastic modulus changes from rapid method. Based Weibull probability method, model was...
In the cold regions of north China, due to special climatic and environmental factors, degradation concrete durability caused by freeze–thaw cycles has been a widespread concern among researchers. Freeze–thaw damage is considered main form that causes deteriorate. Aiming at investigating influence on service life structures in building structures, test was carried out this work study frost resistance formed under different fiber blending modes. The analyzed, microstructure evaluated scanning...
Recent advances in text-to-speech (TTS) synthesis, such as Tacotron and WaveRNN, have made it possible to construct a fully neural network based TTS system, by coupling the two components together. Such system is conceptually sim-ple only takes grapheme or phoneme input, uses Mel-spectrogram an intermediate feature, directly generates speech samples. The achieves quality equal close natural speech. However, high computational cost of issues with robustness limited their usage real-world...
The aim of the present study is to investigate effects recycled coarse and fine aggregates on resistance sulfate (SO 4 2− ) attack concrete under dry–wet cycles. Six groups different ratios tests with were designed determine dynamic elastic modulus mass loss aggregate concrete. A reliability analysis was carried out by using Wiener probabilistic method; finally, product analyzed scanning electron microscopy–X-ray diffraction microscopic test. results show that in process attack, increase...
Recent parallel neural text-to-speech (TTS) synthesis methods are able to generate speech with high fidelity while maintaining performance.However, these systems often lack control over the output prosody, thus restricting semantic information conveyable for a given text.This paper proposes hierarchical TTS system prosodic emphasis by learning latent space that directly corresponds change in emphasis.Three candidate features compared: 1) Variance of pitch and duration within words sentence,...
Abstract In this study, the influence of dosage optimization steel fibers (SFs), basalt (BFs), and polypropylene (PPFs) on mechanical properties hybrid fiber‐reinforced concrete (HFRC) is investigated, including cube compressive strength, splitting tensile prism strength. The orthogonal method was used to design fiber mixtures optimize performance specimens. IBM SPSS Statistics software analyze significance sequence influencing factors SFs, BFs, PPFs through range variance. degree...
With the development of computer vision and remote sensor devices, object detection in aerial images has drawn considerable attention because its ability to provide a wide field view large amount information.Despite this, is challenging task owing densely packed objects, oriented diversity, complex background.In this study, we optimized three aspects YOLOv5 algorithm detect arbitrary objects sensing images, including head structure, features from backbone, angle prediction.To improve...
Neural image compression often faces a challenging trade-off among rate, distortion and perception. While most existing methods typically focus on either achieving high pixel-level fidelity or optimizing for perceptual metrics, we propose novel approach that simultaneously addresses both aspects fixed neural codec. Specifically, introduce plug-and-play module at the decoder side leverages latent diffusion process to transform decoded features, enhancing low quality without altering original...
Network coverage is a key research problem in wireless sensor network field. Sensor node energy generally supplied by batteries, and the limited, so an effective balance strategy can extend networks' life process of coverage. A chaotic particle swarm optimization control algorithm proposed that based on concept balance. It utilizes to improve function, effectively avoiding shortcomings "premature" optimization. also could realize among nodes The simulation results show has good performances...
Steganography techniques, such as audio steganography, have been widely used in covert communication. However, the deep neural network, especially convolutional network (CNN), has greatly threatened security of steganography. Besides, existing adversarial attacks-based countermeasures cannot provide general perturbation, and transferability against unknown steganography detection methods is weak. This paper proposes a cover enhancement method for based on universal perturbations with sample...
Recently, adversarial perturbations have been used to reassign cost which can enhance the security of steganography, called as embedding. However, existing methods selected costs be modified by self-defined rules were hard achieve optimal against steganalyzers. In this paper, we propose an automatic embedding scheme RLAE (deep Reinforcement Learning-based content-adaptive Adversarial Embedding). RLAE, agent network utilizes a generative generates policy for reassignment automatically...
Temperature is one of the most important factors affecting hydration process and crystallinity cement pastes, its effect on gel phase cannot be ignored. The Fe(OH)3 an indispensable in calcium sulfoaluminate cement, which significantly contributes to desirable properties CSA (especially ferric-rich (FR-CSA) cement). In order thoroughly study influence temperature nanostructure phase, XRD, TGA, FTIR, SEM, TEM Mössbauer spectroscopy were used investigate synthesized by sol–gel method ferric...
Sulfate erosion is one of the important factors causing decline concrete durability, and it also most environmentally damaging influential types environmental erosion. In view problem that durability fibre-reinforced decreased due to sulfate attack, according natural conditions in western China, rapid corrosion test fibre (layered hybrid) was carried out using method accelerated saturated solution immersion. The loss strength, mass relative dynamic elastic modulus studied as law degradation....
Recent advances in text-to-speech (TTS) synthesis, such as Tacotron and WaveRNN, have made it possible to construct a fully neural network based TTS system, by coupling the two components together. Such system is conceptually simple only takes grapheme or phoneme input, uses Mel-spectrogram an intermediate feature, directly generates speech samples. The achieves quality equal close natural speech. However, high computational cost of issues with robustness limited their usage real-world...