A5086162739- Remote-Sensing Image Classification
- Remote Sensing and Land Use
- Advanced Image and Video Retrieval Techniques
- Advanced Image Fusion Techniques
- Face and Expression Recognition
- Remote Sensing and LiDAR Applications
- Advanced Chemical Sensor Technologies
- Remote Sensing in Agriculture
- Image Retrieval and Classification Techniques
- Topic Modeling
- Advanced X-ray and CT Imaging
- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Image Processing Techniques and Applications
- Image and Signal Denoising Methods
- Advanced Radiotherapy Techniques
Hangzhou City University
2024
Changzhou City Planning and Design Institute
2024
Zhejiang University
2019-2023
Hyperspectral classification is a widely discussed problem in the remote sensing field. Many researchers have reported good results of hyperspectral classification. However, when applied to real world, strong demand for labeled data will be big obstacle. To address this problem, explored few-shot learning and semisupervised methods variety papers. We propose siamese network composed three-dimensional convolutional neural networks named 3DCSN. design structure 3DCSN that combines contrast...
With the development of deep learning, benchmark hyperspectral imagery classification is constantly improving, but there are still significant challenges for few-shot scenes. This letter proposes an active-learning-based prototypical network (ALPN), which uses to extract representative features from a few samples. Moreover, it combines semisupervised clustering and active learning methods select request labels valuable examples actively. In this way, feature extraction ability gradually...
With the development of deep learning, hyperspectral image classification (HSIC) has improved rapidly in recent years. Unsupervised feature learning algorithms play an important role extracting features from images (HSIs). This letter proposed a random-occlusion-based Bootstrap-Your-Own-Latent network (ROBYOL), combining new augmentation method and superior contrastive algorithm for extraction. The consists self-supervised part classifier as downstream task. It can be proved by experimental...
The noise corruption problem commonly exists in hyperspectral images (HSIs) and severely affects the accuracy of unmixing algorithms. formulation existing HSIs is relatively complex would change conjunction with different devices imaging settings. For real applications, applying denoising approaches without accurate close-to-reality modeling before may not improve, but rather degrade performance. This study proposes a robust method practical learning-based image denoising. We formulated...
Generative pre-trained large language models (LLMs) have demonstrated impressive performance over a wide range of tasks, thanks to the unprecedented amount data they been trained on. As established scaling laws indicate, LLMs' future improvement depends on computing and sources we can leverage for pre-training. Federated learning (FL) has potential unleash majority planet's computational resources, which are underutilized by data-center-focused training methodology current LLM practice. Our...
In the field of deep learning, finetuning pretrained networks to get a good classifier is common way transfer learning. Unlike traditional way, we insert deformable convolutional layers into networks, and finetune new networks. As result, find it performs as well normal one in classification, when construct plane detection pipeline based on two classifiers respectively, with convolution shows better result than other.
Scaling large language models (LLMs) demands extensive data and computing resources, which are traditionally constrained to centers by the high-bandwidth requirements of distributed training. Low-bandwidth methods like federated learning (FL) could enable collaborative training larger across weakly-connected GPUs if they can effectively be used for pre-training. To achieve this, we introduce Photon, first complete system end-to-end LLM training, leveraging cross-silo FL global-scale with...
Although there are many state-of-the-art methods for hyperspectral classification, data deficiency is a problem that should be addressed before popularizing technology. To solve this problem, it worth exploring based on small datasets. Inspired by the advanced deep learning classification and autoencoder structure, we propose structure named three-dimensional convolutional adversarial combines two processes semisupervised classification. Our experiments show its utility in data-deficient...