A5074326967- Advanced Software Engineering Methodologies
- Domain Adaptation and Few-Shot Learning
- Industrial Vision Systems and Defect Detection
- Advanced Neural Network Applications
- Service-Oriented Architecture and Web Services
- Textile materials and evaluations
- EEG and Brain-Computer Interfaces
- Image and Object Detection Techniques
- Smart Agriculture and AI
- Advanced Image and Video Retrieval Techniques
- Remote-Sensing Image Classification
- Cloud Computing and Resource Management
- Plant Virus Research Studies
- Complex Network Analysis Techniques
- Advanced Memory and Neural Computing
- Model-Driven Software Engineering Techniques
- Remote Sensing and Land Use
- Software Engineering Research
- Advanced Sensor and Energy Harvesting Materials
- Disability Rights and Representation
- Neuroscience and Neural Engineering
- Multimodal Machine Learning Applications
- Face and Expression Recognition
- IoT and Edge/Fog Computing
- Robot Manipulation and Learning
Shanghai University
2023-2025
First Affiliated Hospital of Xinjiang Medical University
2025
Xinjiang Uygur Autonomous Region Disease Prevention and Control Center
2025
Xinjiang Medical University
2025
Guizhou University
2013-2024
East China Normal University
2013-2024
Zhengzhou University
2022-2024
Shanghai Jiao Tong University
2024
Xinjiang Academy of Agricultural Sciences
2021-2024
China Energy Engineering Corporation (China)
2024
Image classification has always been a hot research direction in the world, and emergence of deep learning promoted development this field. Convolutional neural networks (CNNs) have gradually become mainstream algorithm for image since 2012, CNN architecture applied to other visual recognition tasks (such as object detection, localization, semantic segmentation) is generally derived from network classification. In wake these successes, CNN-based methods emerged remote sensing scene achieved...
This study proposes a modified convolutional neural network (CNN) algorithm that is based on dropout and the stochastic gradient descent (SGD) optimizer (MCNN-DS), after analyzing problems of CNNs in extracting convolution features, to improve feature recognition rate reduce time-cost CNNs. The MCNN-DS has quadratic CNN structure adopts rectified linear unit as activation function avoid problem accelerate convergence. To address overfitting problem, uses an SGD optimizer, which implemented...
We adopted actual intelligent production requirements and proposed a tiny part defect detection method to obtain stable accurate real-time system solve the problems of manually setting conveyor speed industrial camera parameters in for factory products. First, we considered important influences properties parts environmental on its stability. Second, established correlation model between capability coefficient moving conveyor. Third, algorithm that are based single short detector network...
With the rapid development of machine learning, its powerful function in vision field is increasingly reflected. The combination and robotics to achieve same precise fast grasping as that humans requires high-precision target detection recognition, location reasonable grasp strategy generation, which ultimate goal global researchers one prerequisites for large-scale application robots. Traditional learning has a long history good achievements image processing robot control. CNN...
In the industrial field, anthropomorphism of grasping robots is trend future development, however, basic vision technology adopted by robot at this stage has problems such as inaccurate positioning and low recognition efficiency. Based on practical problem, in order to achieve more accurate objects, an object detection method for based improved YOLOv5 was proposed paper. Firstly, platform designed, wooden block image data set being proposed. Secondly, Eye-In-Hand calibration used obtain...
Resource management challenges frequently manifest in systems and networks as tough online decision tasks, for which the proper solution is dependent on an understanding of workload environment facilitates smooth use mobile edge cloud resources. Due to geographical dispersion resources, constrained resource capacity, unpredictable nature network hierarchy present such contexts, it difficult efficiently schedule jobs environments. Unfortunately, existing heuristic-based methods lack...
In response to the difficulty of plant leaf disease detection and classification, this study proposes a novel method called deep block attention SSD (DBA_SSD) for identification degree classification leaves. We propose three methods, namely, squeeze-and-excitation (Se_SSD), (DB_SSD), DBA_SSD. Se_SSD fuses feature extraction network mechanism channel, DB_SSD improves VGG network, DBA_SSD improved channel mechanism. To reduce training time accelerate process, convolutional layers trained in...
The purpose of mobile robot path planning is to produce the optimal safe path. However, robots have poor real‐time obstacle avoidance in local and longer paths global planning. In order improve accuracy prediction planning, shorten length reduce time, then obtain a better path, we propose decision model based on machine learning (ML) algorithms, an improved smooth rapidly exploring random tree (S‐RRT) algorithm, hybrid genetic algorithm‐ant colony optimization (HGA‐ACO). Firstly, algorithms...
Abstract Plant disease, a huge burden, can cause yield loss of up to 100% and thus reduce food security. Actually, smart diagnosing diseases with plant phenomics is crucial for recovering the most loss, which usually requires sufficient image information. Hence, being pursued as an independent discipline enable development high-throughput phenotyping disease. However, we often face challenges in sharing large-scale data due incompatibilities formats descriptions provided by different...
Primary thyroid lymphoma (PTL) is an extremely rare form of malignancy, merely accounting for 2-5% all cases. Owing to its low incidence and the absence concrete clinical manifestations, PTL frequently misdiagnosed as thyroiditis or cancer, thereby presenting a significant hurdle accurate diagnosis. This case study centered around 46-year-old female patient. We meticulously detailed diagnosis treatment process her primary mucosa-associated lymphoid tissue (MALT) marginal zone lymphoma....
Recent research has shown that the ubiquitous use of cameras and voice monitoring equipment in a home environment can raise privacy concerns affect human mental health. This be major obstacle to deployment smart systems for elderly or disabled care. study uses social robot detect embarrassing situations. Firstly, we designed an improved neural network structure based on You Only Look Once (YOLO) model obtain feature information. By focusing reducing area redundancy computation time, proposed...
With the development of multimedia technology, rapid increasing usage large image database becomes possible. To carry out its management and retrieval, Content-Based Image Retrieval (CBIR) is an effective method. This paper shows advantage content-based retrieval system, as well key technologies. Compare to shortcoming that only certain one feature used in traditional this introduces a method combines color, texture shape for advantage. Then focuses on extraction representation, several...
Robot manipulator trajectory planning is one of the core robot technologies, and design controllers can improve accuracy manipulators. However, most designed at this stage have not been able to effectively solve nonlinearity uncertainty problems high degree freedom In order overcome these performance manipulators, a method based on radial basis function (RBF) neural network proposed in work. Firstly, 6-DOF experimental platform was built. Secondly, overall framework designed, which included...
At present, deep neural networks have been widely used in various fields, but their vulnerability requires attention. The adversarial attack aims to mislead the model by generating imperceptible perturbations on source model, and although white-box attacks achieved good success rates, existing samples exhibit weak migration black-box case, especially some adversarially trained defense models. Previous work for gradient-based optimization either optimizes image before iteration or gradient...
Abstract Catastrophic forgetting in neural networks is a common problem, which lose information from previous tasks after training on new tasks. Although adopting regularization method that preferentially retains the parameters important to task avoid catastrophic has positive effect; existing methods cause gradient be near zero because loss at local minimum. To solve this we propose continuous learning with Bayesian parameter updating and weight memory (CL-BPUWM). First, based Bayes...
Fault diagnosis methods based on deep learning have progressed greatly in recent years. However, the limited training data and complex work conditions still restrict application of these intelligent methods. This paper proposes an bearing fault method, i.e., Siamese Vision Transformer, suiting conditions. The combining network is designed to efficiently extract feature vectors input samples high-level space complete classification fault. In addition, a new loss function Kullback-Liebler...
The complexity of the background and similarities between different types precision parts, especially in high-speed movement conveyor belts complex industrial scenes, pose immense challenges to object recognition parts due diversity illumination. This study presents a real-time method for 0.8 cm darning needles KR22 bearing machine under background. First, we propose an image data increase algorithm based on directional flip, establish two dataset, namely, real increased data. We focus...
In recent years, the use of convolutional neural networks (CNNs) and graph (GNNs) to identify hyperspectral images (HSIs) has achieved excellent results, such methods are widely used in agricultural remote sensing, geological exploration, marine sensing. Although many generalization classification algorithms designed for purpose learning a small number samples, there is often problem low utilization rate position information empty spectral domain. Based on this, GNN with an autoregressive...
Deep learning has been utilized for hyperspectral image (HSI) classification in recent years, with notable performance improvements. In particular, convolutional neural networks (CNNs) methods have achieved major advancements this area. However, there are some drawbacks to the existing CNN-based HSI approaches: 1) lack of effective and simple feature representations CNNs, which overlook effects spectral differences spatial contextual information; 2) model an enormous network complexity as a...