- Video Surveillance and Tracking Methods
- Natural Language Processing Techniques
- Advanced Image Fusion Techniques
- Remote-Sensing Image Classification
- Anomaly Detection Techniques and Applications
- Topic Modeling
- Remote Sensing and Land Use
- Advanced Graph Neural Networks
- Human Pose and Action Recognition
- Infrared Target Detection Methodologies
- Face and Expression Recognition
- Automated Road and Building Extraction
- Cutaneous Melanoma Detection and Management
- Smart Agriculture and AI
- Brain Tumor Detection and Classification
- Face recognition and analysis
- Optical Systems and Laser Technology
- AI and Multimedia in Education
- Advanced Technologies in Various Fields
- Software Engineering Research
- Advanced Text Analysis Techniques
- Grit, Self-Efficacy, and Motivation
- Advanced Neural Network Applications
- Technology and Security Systems
- Energy Load and Power Forecasting
Jiangnan University
2023-2024
Shanghai Institute of Technology
2023-2024
Wuhan Sports University
2010-2022
Shanghai Research Institute of Sports Science
2022
China Southern Power Grid (China)
2021
Xiaomi (China)
2012
After the birth of deep learning, artificial intelligence has entered a vigorous period rapid development. In this process rising and growing, we have made one achievement after another. When learning is applied to fruit target detection, due complex recognition background, large similarity between models, serious texture interference, partial occlusion fruits, detection rate based on traditional methods low. order solve these problems, BCo-YOLOv5 network model proposed recognize detect...
As a hot research topic, sports video classification has wide range of applications in switched TV, on demand, smart and other fields is closely related to people’s lives. Under this background, aroused great interest people. However, the existing methods usually use manual classification, which workers themselves often influence. It challenging ensure accuracy results, leading wrong classification. Due these limitations, we introduce neural network technology automatic sports. This paper...
With the advancement of Internet technology and widespread use mobile smartphones, urban e-commerce is becoming increasingly saturated. Rural ushered in a new era development. How to improve rural logistics carrying capacity keep up with development better serve has become hot topic. The purpose this study predict demand context e-commerce. We develop an indicator system for forecasting Guangdong Province. GM (1,1) gray model weakening buffer operator was used forecast According results,...
Federated learning is a promising solution in several industries for co-training models among distributed clients via centralized servers without leaving private user data on the devices. Thus, federated can be seen as stimulus edge computing paradigm it supports collaborative and model optimization. In view of strict requirements security system reliability hyperspectral classification techniques surveillance, aerospace, military missions, this paper proposes novel stereo attention...
Understanding human emotions and psychology is a critical step toward realizing artificial intelligence, correct recognition of facial expressions essential for judging emotions. However, the differences caused by changes in expression are very subtle, different features less distinguishable, making it difficult computers to recognize accurately. Therefore, this paper proposes novel multi-layer interactive feature fusion network model with angular distance loss. To begin, multi-scale module...
In the era of rapid technological advancement, demand for sophisticated Multi-Object Tracking (MOT) systems in applications such as intelligent surveillance and autonomous navigation has become increasingly critical.However, existing models often struggle with accuracy efficiency densely populated or dynamically complex environments. Addressing these challenges, we introduce a novel deep learning-based MOT model that incorporates latest CT-DETR detection technology an advanced ReID module...
With the advent of Big Data era, specialized data in kill chain domain has increased dramatically, and engine-based method retrieving information can hardly meet users' need for more accurate answers. The includes four components: control equipment, sensor strike equipment (weapon platform), evaluator as well related which contain a large amount valuable such parameter contained each component. If these fragmented confusing are integrated effective query methods established, they help...
MLP-based weakly supervised crowd counting approaches have made significant advancements over the past few years. However, owing to limited datasets, current methods do not consider problem of region-to-region dependency in image. For this, we propose a method termed SR2. SR2 consists three parts: scale-reasoning module, scale-ranking and regression branch. In particular, module extracts fuses image multiple scale feature, then sends fused features branch obtain estimated counts; is used...
Do current large language models (LLMs) better solve graph reasoning and generation tasks with parameter updates? In this paper, we propose InstructGraph, a framework that empowers LLMs the abilities of by instruction tuning preference alignment. Specifically, first structured format verbalizer to unify all data into universal code-like format, which can simply represent without any external graph-specific encoders. Furthermore, stage is introduced guide in solving tasks. Finally, identify...
Efficient and accurate extraction of electrical parameters from circuit datasheets design documents is critical for accelerating in Electronic Design Automation (EDA). Traditional workflows often rely on engineers manually searching extracting these parameters, which time-consuming, prone to human error. To address challenges, we introduce DocEDA, an automated system that leverages advanced computer vision techniques Large Language Models (LLMs) extract seamlessly documents. The layout...
This paper analyzed several mistakes in the traditional 2D analytic theories and then introduced detail a new method basing on DLT theory from two aspects: foundation stone of theories, program design, also utilized direct inspection to analyze errors between old one. The results showed that could obtain more accurate than
MLP-based weakly supervised crowd counting methods have made significant progress in recent years. However, owing to the limited datasets, current do not consider problem of region-to-region dependency image and training complex networks overfitting. To solve above problems, we propose an method termed as SR2. SR2 consists three parts: scale reasoning module, ranking module regression branch. In particular, reasoningmodule extracts fuses multi-scale feature, then sends fused features branch...
Personalized federated learning has gained significant attention as a promising approach to address the challenge of data heterogeneity. In this paper, we relatively unexplored problem in learning. When model been trained and deployed, an unlabeled new client joins, providing personalized for becomes highly challenging task. To challenge, extend adaptive risk minimization technique into unsupervised setting propose our method, FedTTA. We further improve FedTTA with two simple yet effective...
Structural image content recognition (SICR) aims to transcribe a two-dimensional structural (e.g., mathematical expression, chemical formula, or music score) into token sequence. Existing methods are mainly encoder-decoder based and overlook the importance of feature selection spatial relation extraction in map. In this paper, we propose DEAL (shorted for Dynamic fEAture seLection) SICR, which contains dynamic selector extractor as two cornerstone modules. Specifically, novel loss function...
Retelling generation technology has important applications in many aspects, such as: question answering system, it can be used to improve the system's ability understand questions and optimize system performance; With continuous development maturity of various underlying technologies natural language processing, is possible study retelling, which attracted more attention. In this paper, a method generating retelling based on template matching proposed, effectively preserve structural...