- Advanced Image Processing Techniques
- Advanced Vision and Imaging
- Image Processing Techniques and Applications
- Image Enhancement Techniques
- Image and Signal Denoising Methods
- Color Science and Applications
- Metamaterials and Metasurfaces Applications
- Plasmonic and Surface Plasmon Research
- Optical measurement and interference techniques
- Advanced Antenna and Metasurface Technologies
- Generative Adversarial Networks and Image Synthesis
- Terahertz technology and applications
- Advanced Neural Network Applications
- High-Voltage Power Transmission Systems
- Photonic and Optical Devices
- Advanced Image Fusion Techniques
- Color perception and design
- Industrial Vision Systems and Defect Detection
- Photonic Crystals and Applications
- Computer Graphics and Visualization Techniques
- Smart Agriculture and AI
- Advanced Algorithms and Applications
- Advanced Sensor and Control Systems
- Digital Media Forensic Detection
- Advanced Image and Video Retrieval Techniques
South China Robotics Innovative Research Institute
2025
Wuhan Polytechnic University
2022-2025
Northwestern Polytechnical University
2023-2024
Ministry of Industry and Information Technology
2023-2024
Beijing Forestry University
2019-2021
Xi'an University of Technology
2021
Research Institute of Forestry
2019-2020
Xinyang College of Agriculture and Forestry
2019-2020
Agriculture and Forestry University
2020
Institute of Experimental Physics of the Slovak Academy of Sciences
2020
Abstract Optical wavefront engineering is essential for the development of next‐generation integrated photonic devices. It used reflecting terahertz waves in a predesigned nonspecular direction with near‐unitary efficiency, which longstanding challenge high‐performance functional Recently, metagratings have offered an efficient solution beam steering at large angles without need discretization phase or impedance profile. Here, all‐dielectric fabricated using silicon cuboid complex lattice...
Recent data-driven deep learning methods for image reflection removal have made impressive progress, promoting the quality of photo capturing and scene understanding. Due to massive consumption computational complexity memory usage, performance these degrades significantly while dealing with high-resolution images. Besides, most existing can only remove patterns by downsampling input into a much lower resolution, resulting in loss plentiful information. In this paper, we propose novel...
Abstract The unique electromagnetic response characteristics of metasurfaces and their intrinsic physical mechanisms have attracted a lot attention. With the help metasurfaces, amplitude, phase, polarization, other information light waves can be effectively modulated. Fano resonance with asymmetric sharp line shape is sensitive to refractive index changes in environment, it realized through structure design metasurface. Terahertz wave located between microwave infrared used fields...
Predicting depth from a monocular image is an ill-posed and inherently ambiguous issue in computer vision. In this paper, we propose pyramidal third-streamed network (PTSN) that recovers the information using single given RGB image. PTSN uses structure images, which can extract multiresolution features to improve robustness of as input. The full connection layer changed into fully convolutional layers with new upconvolution structure, reduces parameters computational complexity. We loss...
We address the problem of depth estimation from a single monocular image in paper. Depth is an ill-posed and inherently ambiguous problem. In paper, we propose encoder-decoder structure with feature pyramid to predict map RGB image. More specifically, used detect objects different scales The encoder aims extract most representative information original through series convolution operations reduce resolution input adopt Res2-50 as important features. decoder section uses novel upsampling...
The semantic segmentation of outdoor images is the cornerstone scene understanding and plays a crucial role in autonomous navigation robots. Although RGB–D can provide additional depth information for improving performance tasks, current state–of–the–art methods directly use ground truth maps fusion, which relies on highly developed expensive sensors. Aiming to solve such problem, we proposed self–calibrated RGB-D image neural network model based an improved residual without relying sensors,...
The crux of image deraining stems from the challenge recognizing diverse rain patterns within rainy image. Most methods for remain visible residuals in restored image, which suffers insufficient modeling streaks. In this work, we propose contrastive learning-based generative network (CLGNet), follows a coarse-to-fine framework. coarse phase, our CLGNet employs hierarchical encoder–decoder structure to remove obvious patterns, and first generates background Then, introduce well-designed...
HIGHLIGHTS We propose a balanced feature pyramid network model to achieve automatic recognition of agricultural pests. and improved module address the problems large semantic gap before fusion in FPN (feature network), information loss during fusion, how utilize complementary strengthen features after fusion. Our method is evaluated on pest dataset achieves better results, mAP (mean Average Precision) reaches 90.04%, final combining with data enhancement strategy 92.56%. Abstract. In order...
We propose an encoder-decoder with densely convolutional networks model to recover the depth information from a single RGB image without need for sensors. The encoder part serves extract most representative original data through series of convolution operations and reduce resolution spatial input feature. use decoder section produce upsampling structure that improves output resolution. Our is trained scratch, any special tuning process, uses new optimization function adaptively learn rate....
Monocular depth estimation is a fundamental yet challenging task in computer vision as information will be lost when 3D scenes are mapped to 2D images. Although deep learning-based methods have led considerable improvements for this single image, most existing approaches still fail overcome limitation. Supervised learning model regression problem and, result, require large amounts of ground truth data training actual scenarios. Unsupervised treat the synthesis new disparity map, which means...
Color, which is effective information for computer vision, vulnerable to variations in illumination. It necessary achieve illumination invariance the description of an image’s color information. We propose two new spaces that are robust changes: space based on diagonal-offset model and edge space, channels derived from spherical linear transformation between dark channel bright channel. Then, we a moment invariant descriptor according Hu invariants spaces. test performance descriptors with...
The purpose of image motion deblur is to recover the underlying clear from corresponding blur image. In most traditional methods, recovery task formulated as a problem core estimation and use priori calculate. this paper we proposes generative adversarial network(GAN) model based on mobilenet-V3 network structure meet needs blurred mobile devices. Based evaluation indicators, propose new metric device. Extensive experiments show that our method superior competing methods.
The illumination estimation algorithm belongs to the field of color constancy, aiming restoring image through estimating RGB scene illumination. In different scenarios, performance a general varies greatly. If can be predicted, it inferred that scenarios related optimal algorithms is better than for this paper, novel based on outdoor classification was proposed: firstly, support vector machine (svm) classifiers used identify types , and then selected, finally values were calculated.
As one of the underlying pixel-based illumination estimation algorithms, White Patch algorithm is an for calculating global RGB value image based on specific assumption that maximum reflected light scene chromatic. The harsh assumptions illumination, and many images are difficult to satisfy this constraint. In paper, we propose improved method. Firstly, patch extracted by using sliding window method, then use white estimate color each patch, finally kernel density adopted obtain overall...
The critical challenge of image inpainting is to infer reasonable semantics and textures for a corrupted image. Typical methods are built upon some prior knowledge synthesize the complete One potential limitation that those often remain undesired blurriness or semantic mistakes in synthesized while handling images with large areas. In this paper, we propose Collaborative Contrastive Learning-based Generative Model ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML"...
Zinc oxide (ZnO) varistors is widely used in over voltage protection of various circuits for its advantages excellent nonlinear, fast response time and low power consumption. Commercial ZnO was selected dc pulse conductivity tested. Based on grain boundary barrier model, the analyzed Breakdown Pre-breakdown regions. At pulsed bias, capacitive current conduction obtained. The growth ratio greater than that current. tunneling electric field coefficient barrier, permittivity proportion...
Dielectric metamaterials with low ohmic losses and resonating in the local magnetic mode are preferable for enhancing material non-linearity. Here, we propose experimentally demonstrate broadband extraordinary electromagnetic transmission (EET) behavior, which is induced by coupling of modes two ceramic cuboids. It shown that behavior through a perforated metal sheet subwavelength aperture can be achieved exciting first-order Mie resonant these Our findings indicate bandwidth amplitude...
Abstract. With the integration and scale of pig breeding, frequency some diseases is also increasing. To automatically detect porcine reproductive respiratory syndrome (PRRS) during cultivation process, this article proposes an improved method for ear extraction that based on active contour model. Firstly, we use Gaussian space filtering piecewise linear transformation algorithm to highlight target zones interest. Secondly, a randomly picked image point reconstruct region combine model...