A5065336897- Advanced Optical Imaging Technologies
- Advanced Image and Video Retrieval Techniques
- Domain Adaptation and Few-Shot Learning
- Photonic and Optical Devices
- Remote-Sensing Image Classification
- Image Processing and 3D Reconstruction
- Optical Polarization and Ellipsometry
- Optical Coherence Tomography Applications
- Semiconductor Lasers and Optical Devices
- Advanced Neural Network Applications
- Image Retrieval and Classification Techniques
- Magneto-Optical Properties and Applications
- Handwritten Text Recognition Techniques
- Multimodal Machine Learning Applications
- Semantic Web and Ontologies
- Advanced optical system design
- Advanced Computational Techniques and Applications
- Image Processing Techniques and Applications
- 3D Surveying and Cultural Heritage
- Photorefractive and Nonlinear Optics
- Advanced Fiber Optic Sensors
- Image and Signal Denoising Methods
- Infrared Target Detection Methodologies
- 3D Shape Modeling and Analysis
- Photocathodes and Microchannel Plates
Aerospace Information Research Institute
2021-2024
Chinese Academy of Sciences
2015-2024
University of Chinese Academy of Sciences
2015-2024
Institute of Hydrobiology
2015
Institute of Geographic Sciences and Natural Resources Research
2015
ZheJiang Academy of Agricultural Sciences
2012
Pennsylvania State University
1983-2009
Port Washington Public Library
1992
Park University
1990
Deep learning approaches have contributed to the rapid development of remote sensing (RS) image interpretation. The most widely used training paradigm is use ImageNet pretrained models process RS data for specified tasks. However, there are issues such as domain gap between natural and scenes poor generalization capacity models. It makes sense develop a foundation model with general feature representation. Since large amount unlabeled available, self-supervised method has more significance...
The emergence of large-scale large language models, with GPT-4 as a prominent example, has significantly propelled the rapid advancement artificial general intelligence and sparked revolution Artificial Intelligence 2.0. In realm remote sensing (RS), there is growing interest in developing vision models (VLMs) specifically tailored for data analysis this domain. However, current research predominantly revolves around visual recognition tasks, lacking comprehensive, image-text datasets that...
Few-shot object detection (FSOD) has received numerous attention due to the difficulty and time-consuming of labeling objects. Recent researches achieve excellent performance in a natural scene by only using few instances novel classes fine-tune last prediction layer model well-trained on plentiful base data. However, compared with objects single direction small size variety, remote sensing images (RSIs) vary greatly. The methods proposed for cannot be directly applied RSIs. In this article,...
Few-shot object detection, expecting detectors to detect novel classes with a few instances, has made conspicuous progress. However, the prototypes extracted by existing meta-learning based methods still suffer from insufficient representative information and lack awareness of query images, which cannot be adaptively tailored different images. Firstly, only support images are involved for extracting prototypes, resulting in scarce perceptual Secondly, all pixels treated equally when...
Existing deep learning models usually assume that all data obeys independent identically distribution, which is unreasonable in remote sensing. Due to the differences camera parameters, spectral ranges, resolutions, and so on, images acquired by sensing sensors may be greatly diverse, causing face catastrophic forgetting when they are trained on new only. Thus, incremental introduced. An ideal model should expanded as number of tasks increases, have enough ability adapt changes data....
Few-shot segmentation of point cloud remains a challenging task, as there is no effective way to convert local information global representation, which hinders the generalization ability features. In this study, we propose bidirectional feature globalization (BFG) approach, leverages similarity measurement between features and prototype vectors embed perception in fashion. With point-to-prototype (P02PrG), BFG aggregates prototypes according weights from dense sparse prototypes....
Due to the differences in feature distribution between classes, when model learns a continuous data stream, it will encounter catastrophic forgetting. The incremental learning methods have shown great potential solve this problem. However, most existing based on task-incremental are difficult adapt characteristics of remote sensing scenes with few appearance but large features, which is not conducive artificially distinguish task-identity document (ID). Thus, we propose class-incremental...
A joint transform correlation system based on wavelet transforms is introduced. The selection of wavelets and the optical images enables this correlator to identify specific features distinguish similar characters. Preliminary experimental results are given.
A novel scheme for optical realization of wavelet transform a one-dimensional signal is described. Using commercially available components, the proposed system can perform in real time. Some preliminary experimental results are demonstrated.
Figures are very important non-textual information contained in scientific documents. Current digital libraries do not provide users tools to retrieve documents based on the available within figures. We propose an architecture for retrieving by integrating figures and other information. The initial step enabling integrated document search is categorize into a set of pre-defined types. several categories their functionalities scholarly articles. have developed machine-learning-based approach...
In this paper, we introduce the 2020 Gaofen Challenge and relevant scientific outcomes.
Large scale digitization projects have been conducted at digital libraries to preserve cultural artifacts and provide permanent access. The increasing amount of digitized resources, including scanned books scientific publications, requires development tools methods that will efficiently analyze manage large collections resources. In this work, we tackle the problem extracting metadata from volumes journals. Our goal is extract information describing internal structures content volumes, which...
With urgent application requirements, such as satellite in-orbit processing and unmanned aerial vehicle tracking, knowledge distillation (KD) following the teacher–student teaching mechanism has shown great potential to obtain lightweight detectors. However, compact students have limited accuracy due interference of large-scale variations blurred boundaries in remote sensing objects. Specifically, previous methods mostly force responses from layer same depth scale align. Stereotyped manual...
Recently, there has been a growing interest in few-shot incremental object detection (FSIOD). It learns new tasks with limited data while mitigating catastrophic forgetting on previous tasks. However, existing FSIOD methods experience parameter changes after training tasks, causing competition issue among Additionally, the background information differs various and using common weight for all results shift. Constrained by these two issues, only alleviate cannot wholly prevent performance...
Abstract. GEOBIA (Geographic Object-Based Image Analysis) is not only a hot topic of current remote sensing and geographical research. It believed to be paradigm in GIScience. The lack systematic approach designed conceptualize formalize the class definitions makes highly subjective difficult method reproduce. This paper aims put forward framework for based on geographic ontology theory, which could implement "Geographic entities - objects Geographic objects" true reappearance. consists...
A real-time large-capacity rapid-scanning optical correlator utilizing a rotating grating concept is described. We have shown that the proposed scanning (OSC) capable of processing memories with rapid spectrum scanning. With implementation closed-circuit TV system, OSC system can be applied in real-world situations. also experimentally tested overall correlation sensitivity due to scanning, object orientation, scale changes, and tilting grating. Several experimental results obtained this are...
Figures in digital documents contain important information. Current libraries do not summarize and index information available within figures for document retrieval. We present our system on automatic categorization of extraction data from 2-D plots. A machine-learning based method is used to categorize into a set predefined types image features. An automated algorithm designed extract values solid line curves The semantic type extracted plots can be integrated with textual provide more...
Crop fertilization recommendation system involves using models to calculate the needed amount of variety nutrients during crop growth, choosing suitable fertilizers, and arranging time. Whether it can be used widely or not, key point is that parameters in customized easily according with local agricultural production practices. To help address these issues, an infrastructure knowledge base its application proposed. This paper firstly focuses on decomposition model by method object-oriented...
A programmable optical system that can perform binary Boolean logic operations with a microchannel spatial light modulator (MSLM) is presented. The MSLM used as an adder, inverter, subtractor, or buffer. We three liquid-crystal televisions computer input interfacing devices. Through the use of its feedback loop, consecutively execute series operations. Experimental results are provided.