检索结果-内蒙古大学图书馆

第35届中国过程控制会议

作者： Biao Yang Yang Chen Xiucai Chen Haoran Zhang Faculty of Information Engineering and Automation Kunming University of Science and Technology

This study employs the Deep Deterministic Policy Gradient(DDPG) algorithm in deep reinforcement learning to explore its potential in improving uniformity during microwave heating processes. Microwave heating is a crucial technology in the field of thermal processing, yet it often encounters issues with uneven temperature distribution in practical applications, which can affect product quality and processing efficiency. Traditional methods of microwave heating control often struggle to address this challenge. The advantage of the DDPG algorithm lies in its ability to learn policies in continuous action spaces, making it suitable for scenarios in microwave heating where continuous adjustment of power and radiation direction is necessary. In this study, we model the microwave heating system as a reinforcement learning environment, where an intelligent agent optimizes temperature distribution uniformity by adjusting microwave power and radiation direction. The agent uses temperature distribution uniformity as a reward signal and continuously optimizes its policy through the DDPG algorithm to achieve more uniform heating effects. Experimental results demonstrate significant performance advantages of the DDPG algorithm in microwave heating. Compared to traditional microwave heating control methods, we observe that DDPG agents converge more quickly to uniform temperature distributions, reducing energy waste and enhancing product quality and production efficiency. This provides new methods and insights for uniformity control in the field of microwave heating, highlighting the broad prospects of deep reinforcement learning in industrial heating processes.

关键词： microwave heating deep reinforcement learning temperature uniformity

来源：评论

学校读者我要写书评

暂无评论

Person Re-Identification Based on Spatial Feature Learning and Multi-Granularity Feature Fusion

引用

Journal of Shanghai Jiaotong university(science) 2025年第2期30卷 363-374页

作者： DIAO Zijian CAO Shuai LI Wenwei LIANG Jianan WEN Guilin HUANG Weici ZHANG Shouming Faculty of Information Engineering and Automation Kunming University of Science and TechnologyKunming 650500China Robotic Laboratory South China Robotics Innovation Research InstituteFoshan 528300GuangdongChina Foshan Zhiyouren Technology Co. Ltd.Foshan 528300GuangdongChina Robotic Laboratory Institute of Intelligent ManufacturingGuangdong Academy of SciencesGuangzhou 510070China

In view of the weak ability of the convolutional neural networks to explicitly learn spatial invariance and the probabilistic loss of discriminative features caused by occlusion and background interference in pedestrian re-identification tasks,a person re-identification method combining spatial feature learning and multi-granularity feature fusion was ***,an attention spatial transformation network(A-STN)is proposed to learn spatial features and solve the problem of misalignment of pedestrian spatial *** the network was divided into a global branch,a local coarse-grained fusion branch,and a local fine-grained fusion branch to extract pedestrian global features,coarse-grained fusion features,and fine-grained fusion features,*** them,the global branch enriches the global features by fusing different pooling *** local coarse-grained fusion branch uses an overlay pooling to enhance each local feature while learning the correlation relationship between multi-granularity *** local fine-grained fusion branch uses a differential pooling to obtain the differential features that were fused with global features to learn the relationship between pedestrian local features and pedestrian global ***,the proposed method was compared on three public datasets:Market1501,DukeMTMC-ReID and *** experimental results were better than those of the comparative methods,which verifies the effectiveness of the proposed method.

关键词： pedestrian re-identification spatial features attention spatial transformation network multi-branch network relation features

来源：评论

学校读者我要写书评

暂无评论

Constructing Negative Sample Pairs to Mitigate Language Priors in Medical Visual Question Answering 24

Constructing Negative Sample Pairs to Mitigate Language Prio...

引用

Proceedings of the 2024 5th International Conference on Intelligent Medicine and Health

作者： Hao Cong Lijun Liu Xiaobing Yang Wei Peng Li Liu Faculty of Information Engineering and Automation Kunming University of Science and Technology Kunming Yunnan China

ISBN: (纸本)9798400709616

Objective: Our research aims to design a new debiased model that utilizes Constructing Negative Sample Pairs (CNSP) to address the language priors problem in Medical Visual Question Answering (Med-VQA). Language priors can cause models to rely on linguistic cues rather than image content, potentially leading to misdiagnosis and inappropriate treatment recommendations in clinical decision-making. Methods: First, to avoid the language priors caused by the poor feature extraction ability and multimodal feature fusion effect of the model, we adopt a two-stage model of the pre-training stage and fine-tuning stage. In the pre-training stage, we combine Masked Image Modeling (MIM), Masked Language Modeling (MLM), and Image-Text Matching (ITM) pre-training tasks. Second, to solve the language priors caused by the imbalance of the dataset and the loss function design problem, we construct negative sample pairs through self-supervised learning to balance the number of positive and negative sample types and back-propagate the loss value obtained through the negative sample pairs to the fine-tuning stage model to curb the generation of language priors. Finally, to verify the transferability of our method, we added our method to other Med-VQA models and achieved significant improvement in results. Results: Our method significantly enhances the model's accuracy, achieving 78.7%, 84.3%, and 81.3% on the public datasets VQA-RAD, SLAKE, and VQA-MED-2019, respectively. Conclusion: Our CNSP model provides a novel and effective solution to the problem of language priors in Med-VQA, and as a plug-and-play approach, can be applied to other Med-VQA models.

关键词： Language priors

来源：评论

学校读者我要写书评

暂无评论

Unsupervised Social Media Event Detection Method combining Semantic Edge Pruning and Community Discovery 25

Unsupervised Social Media Event Detection Method combining S...

引用

Proceedings of the 4th International Conference on Computer, Artificial Intelligence and Control engineering

作者： Yiwei Lu Zhengtao Yu Yantuan Xian Yuxin Huang Yan Xiang Faculty of Information Engineering and Automation Kunming University of Science and Technology Kunming Yunnan China

ISBN: (纸本)9798400712647

Unsupervised social media event detection methods have been able to effectively construct message graphs by fusing attribute edges and semantic edges. However, the performance of such methods is limited by the quality of the composition and the subsequent community discovery algorithm. The composition has a great impact., and there is a lot of noise in the attribute edges. Although semantics can supplement it to a certain extent and obtain more accurate edge associations, the addition of semantic edges will have a certain impact on the structure of the graph. If there are too many semantic edge associations, noise will be introduced. If there are few edge associations, the graph structure is too sparse, which makes it difficult for clustering algorithms to discover event associations between messages. To address these challenges, we propose a novel unsupervised social media event detection method combining semantic edge pruning and community discovery. Structural entropy can measure the hierarchical structure of the graph and decode real knowledge without being disturbed by noisy data. Therefore, we first propose to construct a message graph through structural entropy. Based on the principle of minimizing structural entropy, we select a reasonable threshold to simplify the edge relationships with weak semantic associations and construct a message graph that is more conducive to event discovery. Secondly, we use community discovery algorithms to achieve efficient detection of social media events. Experimental results show that compared with the existing baseline model, this method improves NMI, AMI and ARI by 1%, 2% and 8% on average respectively, and reduces the detection time consumption by 92.73%.

关键词： Edge Pruning

来源：评论

学校读者我要写书评

暂无评论

Multi-Modal and Cross-Scale Feature Fusion Network for Vehicle Detection with Transformers

Multi-Modal and Cross-Scale Feature Fusion Network for Vehic...

引用

2023 International Conference on Machine Vision, Image Processing and Imaging technology, MVIPIT 2023

作者： Ouyang, Junlin Wang, Qingwang Liu, Jiali Qu, Xin Song, Jian Shen, Tao Kunming University of Science and Technology School of Information Engineering and Automation Kunming China Beijing Institute of Mechanical Equipment Beijing China

ISBN: (纸本)9798350306545

Aerial image vehicle detection plays a significant role in intelligent traffic planning and disaster relief. Researchers have conducted extensive studies in this field and made significant progress. However, most of these methods are based on visible images, which struggle to ensure the reliability of detection systems in various complex real-world scenarios, especially under low-light or nighttime conditions. Introducing multispectral images can alleviate this issue, such as infrared images. Nonetheless, due to the significant semantic gap between different modalities, it is challenging for models to extract complementary information while suppressing redundant information. To address this dilemma, we have innovatively extended the deformable DETR to multi-modal vehicle detection task, propose a multi-modal and cross-scale feature fusion vehicle detection network called MC-DETR. We incorporate aspect ratio and scale into the decoder query. Then, we treat the multi-scale feature maps extracted from visible and infrared modalities as a 3D feature space. Our method adaptively extracts multi-modal and cross-scale complementary information from the 3D feature space with deformable attention. Through comparisons with existing vehicle detection methods on the DroneVehicle dataset, our proposed method achieves a 3.8% higher mAP50 compared to the second-ranked method. © 2023 IEEE.

关键词： Aspect ratio

来源：评论

学校读者我要写书评

暂无评论

EEG Emotion Recognition Based on TQWT-features and Multiscale Convolution Superimposed Transformer 25

EEG Emotion Recognition Based on TQWT-features and Multiscal...

引用

Proceedings of the 2025 2nd International Conference on Generative Artificial Intelligence and information Security

作者： Shuaicheng Li Faculty of Information Engineering and Automation Kunming University of Science and Technology Kunming Yunnan China

ISBN: (纸本)9798400713453

Emotional recognition can help machines to capture the emotional state of human beings, which is an indispensable technology for human-computer interaction and the realization of artificial intelligence. Electroencephalogram (EEG) has been applied to emotion recognition due to its good temporal and spatial resolution. How to capture the correlation of long-distance time-series EEG signal data from both temporal and spatial dimensions and directly affect the effect of emotion recognition is a difficulty in the BCI research. Transformer model can extract long-term correlations of time-series signals. In this paper, a new electroencephalogram emotion recognition method based on spatiotemporal fusion Vision Transformer is proposed. The feature of EEG signals is reconstructed by the tunable Q-factor wavelet transform (TQWT) method. The spatiotemporal correlation between the acquisition channel and the temporal signal is established by position coding for the convolved features. The hybrid model of multiscale convolutional and Vision Transformer (MC-ViT) is utilized for feature extraction and emotion classification. The EEG emotion recognition method based on spatio-temporal fusion vision transformer proposed in this paper has achieved good EEG emotion classification effect on the SEED dataset, which outperforms state-of-the-art deep learning methods.

关键词： Deep learning (DL)

来源：评论

学校读者我要写书评

暂无评论

Research and Practice on the Multi-dimensional Course Goal Achievement Evaluation of Software engineering

引用

计算机教育 2023年第12期 265-271页

作者： Ying Jiang Hongbin Wang Jiaman Ding the Faculty of Information Engineering and Automation Kunming University of Science and TechnologyKunming650500China

According to the standards of engineering education accreditation,the achievement paths and evaluation criteria of course goals are presented,aimed at the objectives of software engineering courses and the characteristics of hybrid teaching in kunming university of science and *** a multi-dimensional evaluation system for course goal achievement of software engineering is *** practice’s results show that the multi-dimensional course goal achievement evaluation is helpful to the continuous improvement of course teaching,which can effectively support the evaluation of graduation outcomes.

关键词： Software engineering Course goal achievement evaluation Multi-dimensional

来源：评论

学校读者我要写书评

暂无评论

Lightweight Surface Defect Detection Algorithm Based on Improved YOLOv5

Lightweight Surface Defect Detection Algorithm Based on Impr...

引用

Mechatronics technology and Intelligent Manufacturing (ICMTIM), International Conference on

作者： Kaijun Yang Tao Chen Faculty of Information Engineering And AutoMation Kunming University of Science and Technology Kunming China

ISBN: (数字)9798350363265

ISBN: (纸本)9798350363272

In response to the current issues of low detection accuracy and slow detection speed of detection algorithms in detecting surface defects on metal surfaces, a lightweight surface defect detection algorithm based on improved YOLOv5 is proposed. Firstly, a lightweight spatial attention module is introduced into the SPP module to extract effective information in the spatial dimension. Secondly, a new feature fusion network Amsf is designed to replace the original feature fusion network, which integrates feature information of different scales. Furthermore, the FAM attention mechanism is added to this network to enhance the detection capability for different types of targets. Finally, the SIoU loss function is used to replace the GIoU loss function to reduce boundary loss errors and improve the accuracy of target localization. Experimental results show that the improved model achieves an average detection accuracy of 81.97% on the NEU-DET dataset, which is 7.74% higher than YOLOv5. The model size and detection speed are comparable to YOLOv5. Effective improvement in detection accuracy is achieved while ensuring detection speed.

关键词： YOLO Location awareness Accuracy Attention mechanisms Mechatronics Metals Feature extraction

来源：评论

学校读者我要写书评

暂无评论

An Adaptive Deep Learning Method Integrating Transfer Learning and Concept Drift Detection for Wind Power Forecasting 43

An Adaptive Deep Learning Method Integrating Transfer Learni...

引用

43rd Chinese Control Conference, CCC 2024

作者： Wang, Jingdi Cao, Yundong Jin, Huaikang Fan, Shouyuan Wang, Bin Jin, Huaiping Kunming University of Science and Technology Faculty of Information Engineering and Automation Kunming650500 China Kunming University of Science and Technology Yunnan Key Laboratory of Artificial Intelligence Kunming650500 China Huaneng Renewables Co. Ltd. Yunnan Branch Kunming650000 China Yunnan Electric Power Design Institute Kunming650051 China Energy Engineering Group China

ISBN: (纸本)9789887581581

With the continuous advancement of new energy technologies, wind power is becoming an increasingly crucial source of clean energy. Due to the stochastic and fluctuating nature of wind power data, there is a significant challenge known as concept drift, where the data distribution changes over time. This poses a substantial obstacle to accurate prediction, leading to performance degradation of predictors. To address this issue, an adaptive deep-learning wind power prediction method combining transfer learning and concept drift detection is proposed. This method incorporates the concept of online transfer learning, transferring offline-trained model parameters to adapt quickly to new data distributions during the online phase. Simultaneously, a combination of an online moving window and Kullback-Leibler (KL) divergence is employed to capture and analyze changes in the online predictor's output layer network parameters. This facilitates a more precise detection of concept drift in wind power prediction, determining whether model updates are necessary. By dynamically updating model parameters and effectively handling concept drift, the approach can maintain high accuracy and stability in wind power prediction. The effectiveness and superiority of the proposed method are validated through the analysis of real-world wind power field data. © 2024 Technical Committee on Control Theory, Chinese Association of automation.

关键词： Wind forecasting

来源：评论

学校读者我要写书评

暂无评论

A Columnar Storage Cone Search Method for Massive Astronomical Data

A Columnar Storage Cone Search Method for Massive Astronomic...

引用

information science, Parallel and Distributed Systems (ISPDS), International Conference on

作者： Yu-hai Liu Faculty of Information Engineering and Automation Kunming University of Science and Technology Kunming China

ISBN: (数字)9798350352719

ISBN: (纸本)9798350352726

In recent years, with the continuous advancement of astronomical instrument manufacturing technology, astronomical observation data has shown an exponential growth trend, and traditional database technologies are no longer able to meet the storage and retrieval performance requirements. To address the challenges of storing massive astronomical data and efficient cone search, a solution utilizing columnar storage is proposed, leveraging the separate storage of each attribute column in columnar storage for independent retrieval to enhance data read efficiency. Building upon this, a specialized cone search area retrieval method tailored for columnar storage mode is introduced to reduce data read and computation in cone search, further enhancing retrieval efficiency.

关键词： information science Search methods Instruments Buildings Relational databases Market research Manufacturing

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：