A5101586835- Horticultural and Viticultural Research
- Multimodal Machine Learning Applications
- Image and Signal Denoising Methods
- Smart Agriculture and AI
- Image Enhancement Techniques
- Recommender Systems and Techniques
- Software Reliability and Analysis Research
- Topic Modeling
- Software Engineering Research
- Software Testing and Debugging Techniques
- Sparse and Compressive Sensing Techniques
- Domain Adaptation and Few-Shot Learning
- Advanced Image Processing Techniques
- Natural Language Processing Techniques
- Fermentation and Sensory Analysis
- Advanced Data Storage Technologies
- Multimedia Communication and Technology
- Expert finding and Q&A systems
- Color Science and Applications
- Remote Sensing in Agriculture
- Customer churn and segmentation
- Video Analysis and Summarization
- Spectroscopy and Chemometric Analyses
- Human Pose and Action Recognition
- Advanced Image and Video Retrieval Techniques
Tsinghua University
2023
Ministry of Agriculture and Rural Affairs
2021-2022
Northwest A&F University
2021-2022
Association for Computing Machinery
2021
Institute of Computing Technology
2021
University of Chinese Academy of Sciences
2021
Nanjing Tech University
2020
Kyowa Hakko Kirin (Singapore)
2018
Peking University
2015
Shanghai Fudan Microelectronics (China)
2014
Large pre-trained code generation models, such as OpenAI Codex, can generate syntax-and function-correct code, making the coding of programmers more productive. In this paper, we introduce CodeGeeX, a multilingual model with 13 billion parameters for generation. CodeGeeX is on 850 tokens 23 programming languages June 2022. Our extensive experiments suggest that outperforms models similar scale both tasks and translation HumanEval-X. Building upon HumanEval (Python only), develop HumanEval-X...
Accurately detecting and segmenting grape cluster in the field is fundamental for precision viticulture. In this paper, a new backbone network, ResNet50-FPN-ED, was proposed to improve Mask R-CNN instance segmentation so that detection performance can be improved under complex environments, shape variations, leaf shading, trunk occlusion, grapes overlapping. An Efficient Channel Attention (ECA) mechanism first introduced network correct extracted features better detection. To obtain detailed...
With the continuous expansion of wine grape planting areas, mechanization and intelligence harvesting have gradually become future development trend. In order to guide picking robot pick grapes more efficiently in vineyard, this study proposed a bunches segmentation method based on Pyramid Scene Parsing Network (PSPNet) deep semantic network for different varieties natural field environments. To end, Convolutional Block Attention Module (CBAM) attention mechanism atrous convolution were...
Unlike code errors, the presence of smell often does not affect behavior software system, but it will cause quality problems in terms readability, understandability, and efficiency. To improve reduce maintenance costs, developers need to detect smells rapidly make corresponding refactoring. In detection, recently, machine learning-based methods become more prevalent can overcome shortcomings heuristic-based methods, which mainly rely on manually designed rules. However, our best knowledge,...
Open-domain dialogue generation in natural language processing (NLP) is by default a pure-language task, which aims to satisfy human need for daily communication on open-ended topics producing related and informative responses. In this paper, we point out that hidden images, named as visual impressions (VIs), can be explored from the text-only data enhance understanding help generate better Besides, semantic dependency between an post its response complicated, e.g., few word alignments some...
This paper proposes a new user similarity measure to improve the collaborative filtering algorithm. We apply basic fractional function and an exponential calculate between users by taking both common features different into consideration. test our two measures on data sets, movie lens book-crossing sets. Experiment results show that slightly improves performance, while significantly outperforms other measures.
Intent detection and slot filling are two important tasks in Spoken Language Understanding. The Condition Random Fields (CRF) was introduced for the pretty much same fashion to deep neural networks. Recently, attention based encoder-decoder models have shown promising results joint intent spoken language understanding dialog systems. However, often trained separately. In this paper, we propose ACJIS, a novel Attentive Cross approach Joint Slot filling. We introduce cross enhance modeling...
This paper proposes a novel demosaicing algorithm based on improved gradients with color correlation. Compared to conventional method, especially region dense lines and textures, the proposed method may reduce probability edge direction judgment, which lead obvious artifacts. The adaptive calculation uses correlation information. can effectively suppress artifacts increase average PSNR R, G B channels by 0.59db, 0.69db 0.56db respectively.
This paper proposes an image green enhancement method based on luminance and hue. In the current algorithm, there are two main issues, i.e., presence of over-saturation color deviation, which can not satisfy people's demands for stronger enhancement. To solve these problems, correct mathematical model about chrominance constraint relations is proposed appropriate product factor list obtained. The MATLAB simulation images show that effectively enhances at same time avoid problems deviation.
Generalized Kittler and Illingworth minimum-error thresholding (GKIT) algorithm was proposed by G.Moser for change detection in synthetic aperture radar (SAR) images with non-Gussion distribution. In this paper, we present an improved GKIT approach unsupervised from amplitude relaxing the demand of same equivalent number looks (ENL) based on Nakagami model. Experimental results actual SAR are given to demonstrate validity our method.