A5100331710- Remote-Sensing Image Classification
- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Advanced Image Fusion Techniques
- Video Surveillance and Tracking Methods
- Infrared Target Detection Methodologies
- Anomaly Detection Techniques and Applications
- Synthetic Aperture Radar (SAR) Applications and Techniques
- Advanced SAR Imaging Techniques
- Multimodal Machine Learning Applications
- Human Pose and Action Recognition
- Remote Sensing and Land Use
- Image Retrieval and Classification Techniques
- Visual Attention and Saliency Detection
- Image and Signal Denoising Methods
- Remote Sensing and LiDAR Applications
- Machine Learning and Data Classification
- Advanced Optical Imaging Technologies
- Image Processing Techniques and Applications
- Geophysical Methods and Applications
- Optical Systems and Laser Technology
- Neural Networks and Applications
- Advanced Graph Neural Networks
- Brain Tumor Detection and Classification
Xidian University
2017-2025
Ministry of Education of the People's Republic of China
2022-2025
Stomatology Hospital
2025
Nanjing University
2018-2025
Lanzhou University
2024
Xi'an Jiaotong University
2024
Langley Research Center
2009-2024
Inner Mongolia University of Technology
2024
Zhejiang University
2009-2024
Shenzhen University
2024
Extracting effective features is always a challenging problem for texture classification because of the uncertainty scales and clutter textural patterns. For classification, spectral analysis traditionally employed in frequency domain. Recent studies have shown potential convolutional neural networks (CNNs) when dealing with task spatial In this article, we try combining both approaches different domains more abundant information proposed novel network architecture named contourlet CNN...
We present a novel fruit counting pipeline that combines deep segmentation, frame to tracking, and 3D localization accurately count visible fruits across sequence of images. Our works on image streams from monocular camera, both in natural light, as well with controlled illumination at night. first train Fully Convolutional Network (FCN) segment video images into non-fruit pixels. then track frames using the Hungarian Algorithm where objective cost is determined Kalman Filter corrected...
In this work we propose a novel framework named Dual-Net aiming at learning more accurate representation for image recognition. Here two parallel neural networks are coordinated to learn complementary features and thus wider network is constructed. Specifically, logically divide an end-to-end deep convolutional into functional parts, i.e., feature extractor classifier. The extractors of subnetworks placed side by side, which exactly form the DualNet. Then two-stream aggregated final...
A graph structure is a powerful mathematical abstraction, which can not only represent information about individuals but also capture the interactions between for reasoning. Geometric modeling and relational inference based on data long-standing topic of interest in computer vision community. In this article, we provide systematic review representation learning its applications vision. First, sort out evolution graphs, categorizing them into nonneural network neural methods way nodes are...
Multimodal Image fusion is becoming urgent in multi-sensor information utilization. However, existing end-to-end image frameworks ignore a priori knowledge integration and long-distance dependencies across domains, which brings challenges to the network convergence global perception complex scenes. In this paper, conditional generative adversarial with transformer (TCGAN) proposed for multimodal fusion. The generator generate fused source images content. discriminators are adopted...
To address the challenges of long-tailed classification, researchers have proposed several approaches to reduce model bias, most which assume that classes with few samples are weak classes. However, recent studies shown tail not always hard learn, and bias has been observed on sample-balanced datasets, suggesting existence other factors affect bias. In this work, we systematically propose a series geometric measurements for perceptual manifolds in deep neural networks, then explore effect...
Brain-inspired algorithms have become a new trend in next-generation artificial intelligence. Through research on brain science, the intelligence of remote sensing can be effectively improved. This paper summarizes and analyzes essential properties cognise learning recent advance interpretation. Firstly, this introduces structural composition brain. Then, five represent brain-inspired are studied, including multiscale geometry analysis, compressed sensing, attention mechanism, reinforcement...
For visual question answering on remote sensing (RSVQA), current methods scarcely consider geospatial objects typically with large-scale differences and positional sensitive properties. Besides, modeling reasoning the relationships between entities have rarely been explored, which leads to one-sided inaccurate answer predictions. In this article, a novel method called spatial hierarchical network (SHRNet) is proposed, endows (RS) (VQA) system enhanced visual–spatial capability. Specifically,...
Deep convolutional neural networks (CNNs) are significant in remote sensing. Due to the strong local representation learning ability, CNNs have excellent performance sensing scene classification. However, focus on location-sensitive representations spatial domain and lack contextual information mining capabilities. Meanwhile, classification still faces challenges, such as complex scenes differences target sizes. To address problems challenges above, more robust feature necessary. In this...
The convolutional neural network has achieved remarkable results in most medical image seg- mentation applications. However, the intrinsic locality of convolution operation limitations modeling long-range dependency. Although Transformer designed for sequence-to-sequence global prediction was born to solve this problem, it may lead limited positioning capability due insufficient low-level detail features. Moreover, features have rich fine-grained information, which greatly impacts edge...
Panchromatic (PAN) and multispectral (MS) imagery classification is one of the hottest topics in field remote sensing. In recent years, deep learning techniques have been widely applied many areas image processing. this paper, an end-to-end framework based on multiple instance (DMIL) proposed for MS PAN images' using joint spectral spatial information feature fusion. There are two instances framework: used to capture other describe MS. The features obtained by concatenated directly, which...
With the development of remote sensing imaging technology, images with high-resolution and complex structure can be acquired easily. The classification is always a hot challenging problem. In order to improve performance image classification, we propose an adaptive multiscale deep fusion residual network (AMDF-ResNet). AMDF-ResNet consists backbone network. including several blocks generates hierarchy features, which contain semantic information from low high levels. network, feature module...
Deep learning models such as convolutional neural network has been widely used in 3D biomedical image segmentation. However, most of them neither consider the correlations between different modalities, nor fully exploit depth information. To better leverage multi-modalities and information, we proposed an architecture for brain tumor segmentation multi- modal magnetic resonance images (MRI), named LSTM UNet. Experiments results on BRATS-2015 show that our method outperforms state-of-the-art...
Panchromatic (PAN) and multispectral (MS) images have coordinated paired spatial spectral information, which can complement each other make up for their shortcomings image interpretation. In this article, a novel classification method called the deep group spatial–spectral attention fusion network is proposed PAN MS images. First, processed by unpooling to obtain same resolution as that of image. Second, modules are extract features. The regarded input two modules, respectively. Third,...
We study self-supervised video representation learning, which is a challenging task due to 1) lack of labels for explicit supervision; 2) unstructured and noisy visual information. Existing methods mainly use contrastive loss with clips as the instances learn by discriminating from each other, but they need careful treatment negative pairs either relying on large batch sizes, memory banks, extra modalities or customized mining strategies, inevitably includes data. In this paper, we observe...
Deep learning models are modern tools for spatio-temporal graph (STG) forecasting. Though successful, we argue that data scarcity is a key factor limiting their recent improvements. Meanwhile, contrastive has been an effective method providing self-supervision signals and addressing in various domains. In view of this, one may ask: can leverage the additional from to alleviate scarcity, so as benefit STG forecasting? To answer this question, present first systematic exploration on...
Semantic segmentation based on deep learning has achieved impressive results in recent years, but these are supported by a large amount of labeled data which requires intensive annotation at the pixel level, particularly for high-resolution remote sensing (RS) images. In this work, we propose simple yet efficient semisupervised framework linear sampling self-training, named LSST, to improve performance RS image semantic segmentation. Specifically, classical pseudo-labeling-based...