A5076026731- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- 3D Shape Modeling and Analysis
- Image Processing and 3D Reconstruction
- Medical Image Segmentation Techniques
- Video Surveillance and Tracking Methods
- Computer Graphics and Visualization Techniques
- Remote-Sensing Image Classification
- Generative Adversarial Networks and Image Synthesis
- Video Analysis and Summarization
- Human Pose and Action Recognition
- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Advanced Vision and Imaging
- Robotics and Sensor-Based Localization
- Image Enhancement Techniques
- 3D Surveying and Cultural Heritage
- Spectroscopy and Chemometric Analyses
- Remote Sensing and Land Use
- Image and Object Detection Techniques
- Geochemistry and Geologic Mapping
- Advanced Manufacturing and Logistics Optimization
- Industrial Vision Systems and Defect Detection
- Optical measurement and interference techniques
- Visual Attention and Saliency Detection
China University of Petroleum, East China
2016-2025
Institute of Software
2022-2025
Qingdao Academy of Intelligent Industries
2025
China University of Petroleum, Beijing
2008-2023
Sichuan University
2015-2017
Guilin University of Electronic Technology
2015
Institute of Computing Technology
2004-2011
Chinese Academy of Sciences
2004-2011
North Carolina State University
2008
China University of Geosciences (Beijing)
2008
Abstract Objective. Retinal vessel segmentation from optical coherence tomography angiography (OCTA) volumes is significant in analyzing blood supply structures and the diagnosing ophthalmic diseases. However, accurate retinal 3D OCTA remains challenging due to interference of choroidal flow signals variations structure. Approach. This paper proposes a layer attention network (LA-Net) for 3D-to-2D segmentation. The comprises projection path 2D path. key component proposed multi-scale module,...
Hyperspectral image (HSI) contains rich spatial and spectral information, which has been widely used in resource exploration, ecological environment monitoring, land cover classification target recognition. However, the nonlinearity of HSI data strong correlation between bands also bring difficulties challenges to application. In particular, limited available hyperspectral training samples will lead accuracy cannot be improved. Therefore, making full use advantages data, through algorithms...
The movement data of curling targets is great significance for the analysis and research curling. However, in real-life competitions, volume limited easy to be occluded, venue background illumination complicated. To address these challenges, a target detection model, IFCD, based on Inverted Feature Extraction Network (IFNet) proposed. IFNet allocates more resources deal with high-resolution features without introducing additional computational burdens, thus avoiding feature loss caused by...
With the development of 3D geological modeling, implicit modeling methods have gradually gained popularity. However, existing potential field cannot directly represent unconformable interfaces. In response, an method based on a vector was proposed, which generates surface models through and generalized marching cubes algorithm, visualizes results. An experiment conducted study area certain mineral deposit, model with consistency no topological errors established, demonstrating effectiveness...
In this paper, we have developed a novel framework called <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">JustClick</i> to enable personalized image recommendation via exploratory search from large-scale collections of Flickr images. First, topic network is automatically generated summarize images at semantic level. Hyperbolic visualization further used interactive navigation and exploration the network, so that users can gain insights first...
Soil carbon content prediction based on hyperspectral images can achieve large-scale spatial measurement, which has the advantages of wide coverage and fast information collection, is more suitable for field data collection. However, research soil mainly focuses feature extraction spectral information, ignoring cannot well reveal intrinsic structural characteristics data. Aiming at lack features consideration in images, methods multi-scale fusion are proposed by image. At same time...
Changes in land cover will cause the changes climate and environmental characteristics, which has an important influence on social economy ecosystem. The main form of is different types soil. Compared with traditional methods, visible near-infrared spectroscopy technology can classify soil rapidly, effectively, nondestructively. Based technology, this paper takes six Qingdao, China orchards, woodlands, tea plantations, farmlands, bare lands, grasslands as examples establishes a convolutional...
In this paper, we have developed a novel scheme to incorporate topic network and representativeness-based sampling for achieving semantic visual summarization visualization of large-scale collections Flickr images. First, is automatically generated summarizing visualizing images at level, so that users can select more suitable keywords precise query formulation. Second, the diverse similarities between semantically-similar are characterized precisely by using mixture-of-kernels image...
Smoothing of the graph convolution is not conducive to characterizing local differences point cloud. To solve this problem, we propose a Differential Graph Convolutional Network (Differ-GCN) for cloud analysis. First, new construction strategy that can make similar nodes in space belong same graph, which better represent commonality. After that, features are extracted by similarity matrix. Some smoothing information removed optimize over-smoothing and combined with difference points get...
To achieve more effective solution for large-scale image classification (i.e., classifying millions of images into thousands or even tens object classes categories), a deep multi-task learning algorithm is developed by seamlessly integrating CNNs with over the concept ontology, where ontology used to organize large numbers categories hierarchically and determine inter-related tasks automatically. Our can integrate learn discriminative high-level features representation, it also leverage...
The applications of isometric 3-D objects have recently received sufficient attention and, thus, it is very attractive to retrieve such from large-scale collections. Although existing approaches presented some interesting ideas, their performance limited ability on feature representation. To improve the object (shape) recognition, recent algorithms prefer using complicated deep neural networks learn discriminative features, but they consume huge amounts computing resources. Instead, this...
In this paper, we propose a new approach for cross-scenario clothing retrieval and fine-grained style recognition. The query photos captured by cameras or other mobile devices are filled with noisy background while the product images online shopping usually presented in pure environment. We tackle problem two steps. Firstly, hierarchical super-pixel merging algorithm based on semantic segmentation is proposed to obtain intact item. Secondly, aiming at solving of recognition different...
Abstract Background Cerebrovascular segmentation is a crucial step in the computer‐assisted diagnosis of cerebrovascular pathologies. However, accurate extraction cerebral vessels from time‐of‐flight magnetic resonance angiography (TOF‐MRA) data still challenging due to complex topology and slender shape. Purpose The existing deep learning‐based approaches pay more attention skeleton ignore contour, which limits performance structure. We aim weight contour brain shallow features when...