Procedia Computer Science最新文献_第8页

A systematic review of active learning approaches in the selection of medical images 主动学习方法在医学图像选择中的系统回顾

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.02.186

Maria Santos , Goreti Marreiros

Background: Active Learning has been proven to be an effective way to maximize the model’s learning capacity, using fewer amounts of labeled data. In the field of medical imaging data, data and annotations can be scarce and very expensive to obtain, so techniques like Active Learning can be a useful solution. Methods: For this systematic review, the data sources were obtained through IEEE Explore, PubMed, and ACM Digital Library, between the period of 2018 and 2023. Only studies that belonged to the field of healthcare (using medical images as a dataset) and machine learning, written in English and that were not a book, or a survey were used. Covidence was used as a tool to synthesize the results. Results: From 336 records, 51 were included in this review. Interpretation: Most studies showed that Active Learning can have a positive impact on the construction of models, however, it is important to not consider only the informativeness/uncertainty of the sample, but also the distribution of the data, reducing the probability of selecting samples that are not representative enough of the dataset or outliers. Active Learning is usually an iterative process until a stop criterion is met, for example, the model’s performance. To evaluate an Active Learning solution, the proposed method is usually compared with random sampling, or other Active Learning queries.

背景：主动学习已经被证明是一种有效的方法来最大化模型的学习能力，使用较少的标记数据。在医学成像数据领域，数据和注释可能是稀缺的，并且获取起来非常昂贵，因此像主动学习这样的技术可能是一个有用的解决方案。方法：本系统综述的数据来源为2018年至2023年，通过IEEE Explore、PubMed和ACM数字图书馆获得。只使用了属于医疗保健领域（使用医学图像作为数据集）和机器学习的研究，这些研究是用英语编写的，不是一本书，也不是一项调查。以covid为工具综合结果。结果：从336例记录中，51例纳入本综述。解释：大多数研究表明，主动学习可以对模型的构建产生积极的影响，然而，重要的是不仅要考虑样本的信息性/不确定性，还要考虑数据的分布，减少选择样本的概率，这些样本不足以代表数据集或异常值。主动学习通常是一个迭代过程，直到满足停止标准，例如，模型的性能。为了评估主动学习的解决方案，通常将所提出的方法与随机抽样或其他主动学习查询进行比较。

{"title":"A systematic review of active learning approaches in the selection of medical images","authors":"Maria Santos , Goreti Marreiros","doi":"10.1016/j.procs.2025.02.186","DOIUrl":"10.1016/j.procs.2025.02.186","url":null,"abstract":"<div><div>Background: Active Learning has been proven to be an effective way to maximize the model’s learning capacity, using fewer amounts of labeled data. In the field of medical imaging data, data and annotations can be scarce and very expensive to obtain, so techniques like Active Learning can be a useful solution. Methods: For this systematic review, the data sources were obtained through IEEE Explore, PubMed, and ACM Digital Library, between the period of 2018 and 2023. Only studies that belonged to the field of healthcare (using medical images as a dataset) and machine learning, written in English and that were not a book, or a survey were used. Covidence was used as a tool to synthesize the results. Results: From 336 records, 51 were included in this review. Interpretation: Most studies showed that Active Learning can have a positive impact on the construction of models, however, it is important to not consider only the informativeness/uncertainty of the sample, but also the distribution of the data, reducing the probability of selecting samples that are not representative enough of the dataset or outliers. Active Learning is usually an iterative process until a stop criterion is met, for example, the model’s performance. To evaluate an Active Learning solution, the proposed method is usually compared with random sampling, or other Active Learning queries.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"256 ","pages":"Pages 843-851"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143592862","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

K-operator as a predictor for Alzheimer-Perusini’s disease k算子作为阿尔茨海默病的预测因子

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.02.173

Maria Mannone , Norbert Marwan , Peppino Fazio , Patrizia Ribino

Progressive memory loss occurring in age-related neurological diseases contributes to the disgregation of the individual, with serious personal and social consequences. We model the brain network damage provoked by a neurological disease through a physics-inspired mathematical operator, K. Acting on a diseased brain, K provides the disease time evolution. Focusing on Alzheimer-Perusini’s disease (AD), we approximate the K-operator considering selected patients of the ADNI 2 dataset. We also propose K as a predictor for the disease progress over time and give its preliminary evaluation in the AD progression from the cognitive normal (CN) stage to AD through intermediate mild cognitive impairment (MCI) stages.

与年龄相关的神经系统疾病中发生的进行性记忆丧失有助于个体的分散，具有严重的个人和社会后果。我们通过一个受物理学启发的数学算子K来模拟由神经系统疾病引起的大脑网络损伤。K作用于患病的大脑，提供了疾病的时间演化。针对阿尔茨海默-佩鲁西尼病（AD），我们考虑ADNI 2数据集的选定患者来近似k算子。我们还提出K作为疾病随时间进展的预测因子，并对AD从认知正常（CN）阶段到中度轻度认知障碍（MCI）阶段的进展进行了初步评估。

引用次数: 0

Neural Network Ensemble for Detecting Parasite Eggs in Microscopic Images 显微图像中寄生虫卵检测的神经网络集成

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.02.174

Matheus L.L. Bessa , Geraldo Braz Junior , João Dallyson Souza de Almeida

Intestinal parasite infections are a global health problem. In 2022, the WHO estimates that up to 1.2 billion people will be infected with Ascaris lumbricoides. Diagnosis is conducted by analyzing faecall samples under a microscope. However, this process is laborious and prone to error. Considering this, this study proposes a methodology to automate the detection of parasite eggs in microscope images. This methodology applies multiple object detectors in an ensemble and submits a model to reduce false negatives in the public dataset Chula-ParasiteEgg-11, with 11,000 images and 11 classes of parasites. Using this approach, it was possible to reduce the false negative rate and improve the f1 score up to 0.94. The results suggest that the proposed model leads to a reduction of false negatives and an improvement in recall.

肠道寄生虫感染是一个全球性的健康问题。世卫组织估计，到2022年，将有多达12亿人感染类蛔虫。通过在显微镜下分析粪便样本进行诊断。然而，这个过程很费力，而且容易出错。考虑到这一点，本研究提出了一种在显微镜图像中自动检测寄生虫卵的方法。该方法在集成中应用多个目标检测器，并提交一个模型来减少公共数据集Chula-ParasiteEgg-11中的假阴性，该数据集包含11,000张图像和11类寄生虫。使用这种方法，可以降低假阴性率，并将f1分数提高到0.94。结果表明，提出的模型导致假阴性的减少和召回的提高。

引用次数: 0

Optimization Application of Dynamic Programming Algorithm in Computer Security Management 动态规划算法在计算机安全管理中的优化应用

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.238

Ying Li

With the continuous development of information technology and the increasing complexity of network attacks, computer security management faces increasingly severe challenges, especially in how to optimize resource allocation and improve security protection efficiency. In order to cope with these problems, this paper introduces an optimization method based on dynamic programming algorithm to improve the effectiveness of computer security management. Firstly, a multi-level security management model is constructed. Then, the dynamic programming algorithm is applied to optimize the response sequence and resource allocation strategy of security incidents. Finally, the optimal strategies under different attack modes are designed, and the simulated annealing algorithm is used to further improve the quality of the solution. Experimental results show that this method significantly outperforms traditional strategies on multiple indicators. In terms of security incident response time, the optimization strategy shortens the response time by an average of 30%. In terms of resource utilization, the average resource utilization of the optimization strategy reaches 87.5%, which is significantly higher than the traditional strategy. In addition, in terms of security cost and risk control, the optimization strategy reduces costs by 4.7% and risks by 4.4%. The experiment verifies the effectiveness of the optimization strategy based on dynamic programming in improving response efficiency, optimizing resource allocation and reducing security risks, providing new ideas for computer security management.

随着信息技术的不断发展和网络攻击的日益复杂，计算机安全管理面临着越来越严峻的挑战，特别是如何优化资源配置，提高安全防护效率。为了解决这些问题，本文提出了一种基于动态规划算法的优化方法，以提高计算机安全管理的有效性。首先，构建了多级安全管理模型。然后，应用动态规划算法优化安全事件的响应顺序和资源分配策略。最后，设计了不同攻击模式下的最优策略，并利用模拟退火算法进一步提高了解的质量。实验结果表明，该方法在多个指标上明显优于传统策略。在安全事件响应时间方面，优化策略使响应时间平均缩短了30%。在资源利用率方面，优化策略的平均资源利用率达到87.5%，明显高于传统策略。此外，在安全成本和风险控制方面，优化策略使成本降低4.7%，风险降低4.4%。实验验证了基于动态规划的优化策略在提高响应效率、优化资源配置、降低安全风险等方面的有效性，为计算机安全管理提供了新的思路。

{"title":"Optimization Application of Dynamic Programming Algorithm in Computer Security Management","authors":"Ying Li","doi":"10.1016/j.procs.2025.04.238","DOIUrl":"10.1016/j.procs.2025.04.238","url":null,"abstract":"<div><div>With the continuous development of information technology and the increasing complexity of network attacks, computer security management faces increasingly severe challenges, especially in how to optimize resource allocation and improve security protection efficiency. In order to cope with these problems, this paper introduces an optimization method based on dynamic programming algorithm to improve the effectiveness of computer security management. Firstly, a multi-level security management model is constructed. Then, the dynamic programming algorithm is applied to optimize the response sequence and resource allocation strategy of security incidents. Finally, the optimal strategies under different attack modes are designed, and the simulated annealing algorithm is used to further improve the quality of the solution. Experimental results show that this method significantly outperforms traditional strategies on multiple indicators. In terms of security incident response time, the optimization strategy shortens the response time by an average of 30%. In terms of resource utilization, the average resource utilization of the optimization strategy reaches 87.5%, which is significantly higher than the traditional strategy. In addition, in terms of security cost and risk control, the optimization strategy reduces costs by 4.7% and risks by 4.4%. The experiment verifies the effectiveness of the optimization strategy based on dynamic programming in improving response efficiency, optimizing resource allocation and reducing security risks, providing new ideas for computer security management.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 494-503"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124573","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Construction of Intelligent Electronic Fence System Based on Computer Vision Algorithm 基于计算机视觉算法的智能电子围栏系统构建

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.239

Yaokuan Wen, Qingyu Zhi, Kan Zhang, Yong Li, Yichen Cui, Haiyang Du

With the continuous development of technology, electronic fences face more and more security issues and challenges. This paper used convolutional neural network (CNN) technology to establish an intrusion detection system to achieve high-precision recognition and real-time response to intrusion behavior. The system used image preprocessing technology to improve image quality and reduce environmental interference, and used multi-sensor information fusion to improve system robustness. In order to improve real-time response capabilities, the system uses multi-threaded design and model optimization to achieve rapid and accurate identification of safety hazards in complex environments. At the same time, the system also integrates functions such as behavior recognition and remote control to achieve automated intrusion defense and rapid response. The results show that the intelligent electronic fence system is superior to the traditional system in terms of response time, with an average response time of 109.1 milliseconds. The false alarm rate and missed alarm rate are significantly lower than those of the traditional system. The false alarm rate and missed alarm rate for flame detection are 0.7% and 0.1% respectively, and the detection range is superior to other systems under different conditions. The intelligent electronic fence system has significant advantages in improving security and protection capabilities, and provides a new technical solution for modern security protection.

随着科技的不断发展，电子围栏面临越来越多的安全问题和挑战。本文利用卷积神经网络（CNN）技术建立入侵检测系统，实现对入侵行为的高精度识别和实时响应。该系统采用图像预处理技术提高图像质量，减少环境干扰，采用多传感器信息融合技术提高系统鲁棒性。为了提高实时响应能力，系统采用多线程设计和模型优化，实现了复杂环境下安全隐患的快速准确识别。同时，系统还集成了行为识别、远程控制等功能，实现了自动化入侵防御和快速响应。结果表明，智能电子围栏系统在响应时间上优于传统系统，平均响应时间为109.1毫秒。虚警率和漏警率明显低于传统系统。火焰检测虚警率和漏警率分别为0.7%和0.1%，不同条件下的检测范围优于其他系统。智能电子围栏系统在提高安全防护能力方面具有显著优势，为现代安全防护提供了新的技术解决方案。

{"title":"Construction of Intelligent Electronic Fence System Based on Computer Vision Algorithm","authors":"Yaokuan Wen, Qingyu Zhi, Kan Zhang, Yong Li, Yichen Cui, Haiyang Du","doi":"10.1016/j.procs.2025.04.239","DOIUrl":"10.1016/j.procs.2025.04.239","url":null,"abstract":"<div><div>With the continuous development of technology, electronic fences face more and more security issues and challenges. This paper used convolutional neural network (CNN) technology to establish an intrusion detection system to achieve high-precision recognition and real-time response to intrusion behavior. The system used image preprocessing technology to improve image quality and reduce environmental interference, and used multi-sensor information fusion to improve system robustness. In order to improve real-time response capabilities, the system uses multi-threaded design and model optimization to achieve rapid and accurate identification of safety hazards in complex environments. At the same time, the system also integrates functions such as behavior recognition and remote control to achieve automated intrusion defense and rapid response. The results show that the intelligent electronic fence system is superior to the traditional system in terms of response time, with an average response time of 109.1 milliseconds. The false alarm rate and missed alarm rate are significantly lower than those of the traditional system. The false alarm rate and missed alarm rate for flame detection are 0.7% and 0.1% respectively, and the detection range is superior to other systems under different conditions. The intelligent electronic fence system has significant advantages in improving security and protection capabilities, and provides a new technical solution for modern security protection.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 504-511"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124574","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Research on Crack Segmentation and Detection of Red Brick Wall Structure based on Deep Learning 基于深度学习的红砖墙体结构裂缝分割与检测研究

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.240

Wenjuan Peng , Wei Zhao , Qiusheng Zhang , Zhuoya Bai , Ying Zeng , Mingyang Qi , Jinshun Nan

The present paper discusses a technique for crack segmentation and detection in red brick walls that is based on deep learning. This technology is designed to improve the efficiency and accuracy of assessing building safety. With the development of the construction industry, the detection of cracks in red brick walls has become particularly important. Traditional detection methods are labor-intensive and error-prone, while deep learning models provide an efficient and reliable solution. In this paper, we study a variety of deep learning models, including PSPNet, DeepLabV3+, ERFNet, ANN, CCNet, and SegFormer, and compare their performance in the wall crack detection and segmentation task through experiments that use a real scene dataset to validate the model’s accuracy and generalization ability in the presence of interfering factors. Experimental results show that the SegFormer model performs best in IoU, F1, ACC and Recall, reaching 65.99%, 77.37%, 99.87%, and 80.79%, respectively, and with the addition of the attention mechanism to the SegFormer model for optimization, the model’s IoU and F1 are improved by 1.16% and 1.13%, respectively. The performance was significantly improved. The results provide technical support for detecting and repairing cracks in red brick walls, which helps to detect and repair potential safety hazards in a timely manner.

本文讨论了一种基于深度学习的红砖墙体裂缝分割与检测技术。该技术旨在提高建筑安全评估的效率和准确性。随着建筑业的发展，红砖墙体裂缝的检测变得尤为重要。传统的检测方法是劳动密集型且容易出错的，而深度学习模型提供了高效可靠的解决方案。本文研究了多种深度学习模型，包括PSPNet、DeepLabV3+、ERFNet、ANN、CCNet和SegFormer，并通过使用真实场景数据集的实验，比较了它们在墙体裂缝检测和分割任务中的性能，验证了模型在干扰因素存在下的准确性和推广能力。实验结果表明，SegFormer模型在IoU、F1、ACC和Recall方面表现最好，分别达到65.99%、77.37%、99.87%和80.79%，并且在SegFormer模型中加入注意机制进行优化后，模型的IoU和F1分别提高了1.16%和1.13%。性能得到了显著提高。研究结果为红砖墙体裂缝检测与修复提供了技术支持，有助于及时发现和修复安全隐患。

{"title":"Research on Crack Segmentation and Detection of Red Brick Wall Structure based on Deep Learning","authors":"Wenjuan Peng , Wei Zhao , Qiusheng Zhang , Zhuoya Bai , Ying Zeng , Mingyang Qi , Jinshun Nan","doi":"10.1016/j.procs.2025.04.240","DOIUrl":"10.1016/j.procs.2025.04.240","url":null,"abstract":"<div><div>The present paper discusses a technique for crack segmentation and detection in red brick walls that is based on deep learning. This technology is designed to improve the efficiency and accuracy of assessing building safety. With the development of the construction industry, the detection of cracks in red brick walls has become particularly important. Traditional detection methods are labor-intensive and error-prone, while deep learning models provide an efficient and reliable solution. In this paper, we study a variety of deep learning models, including PSPNet, DeepLabV3+, ERFNet, ANN, CCNet, and SegFormer, and compare their performance in the wall crack detection and segmentation task through experiments that use a real scene dataset to validate the model’s accuracy and generalization ability in the presence of interfering factors. Experimental results show that the SegFormer model performs best in IoU, F1, ACC and Recall, reaching 65.99%, 77.37%, 99.87%, and 80.79%, respectively, and with the addition of the attention mechanism to the SegFormer model for optimization, the model’s IoU and F1 are improved by 1.16% and 1.13%, respectively. The performance was significantly improved. The results provide technical support for detecting and repairing cracks in red brick walls, which helps to detect and repair potential safety hazards in a timely manner.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 512-519"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124575","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Emotion Recognition and Intervention Technology for Autistic Children Based on the Fusion of Neural Networks and Biological Signals 基于神经网络与生物信号融合的自闭症儿童情绪识别与干预技术

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.243

Yifei Wang

Given the significant difficulties that children with autism face in emotion recognition and intervention, there is an urgent need to develop accurate and efficient technical means to improve their social interaction and emotional understanding abilities. This study discusses a biological signal emotion recognition and intervention technology that integrates Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM). First, this paper collects a variety of biological signal data of autistic children in different emotional states, including heart rate, galvanic skin response (GSR) and electroencephalogram (EEG), and preprocesses and extracts features of the data. Next, this paper builds and trains a deep learning model that integrates CNN and LSTM, classifies and analyzes the extracted features into emotional states, and achieves high-precision emotion recognition. Finally, this paper designs personalized intervention strategies based on the recognition results, and provides emotional guidance and intervention to children through a real-time feedback system. In the experimental conclusion, the accuracy of emotion recognition of the proposed fusion model in the training set and the verification set is 97.5% and 94.2% respectively, which is significantly better than the single mode signal processing method. In addition, the personalized intervention strategy based on this model achieved improvements of 45%, 3.8 points, and 4.2 points in reducing the amplitude of emotional fluctuations, enhancing emotional regulation ability, and improving social behavior, respectively, demonstrating the significant advantages and application potential of multimodal biosignal fusion in improving emotion recognition and intervention effects in children with autism.

鉴于自闭症儿童在情绪识别和干预方面存在显著困难，迫切需要开发准确、高效的技术手段来提高自闭症儿童的社会交往和情绪理解能力。本研究探讨了一种融合卷积神经网络（CNN）和长短期记忆网络（LSTM）的生物信号情绪识别与干预技术。首先，收集自闭症儿童在不同情绪状态下的各种生物信号数据，包括心率、皮肤电反应（GSR）和脑电图（EEG），并对数据进行预处理和特征提取。接下来，本文构建并训练了一个融合CNN和LSTM的深度学习模型，将提取的特征分类并分析为情绪状态，实现了高精度的情绪识别。最后，根据识别结果设计个性化干预策略，并通过实时反馈系统对儿童进行情感引导和干预。实验结论表明，本文提出的融合模型在训练集和验证集上的情绪识别准确率分别为97.5%和94.2%，明显优于单模信号处理方法。此外，基于该模型的个性化干预策略在降低情绪波动幅度、增强情绪调节能力和改善社会行为方面分别提高了45%、3.8分和4.2分，显示了多模态生物信号融合在提高自闭症儿童情绪识别和干预效果方面的显著优势和应用潜力。

{"title":"Emotion Recognition and Intervention Technology for Autistic Children Based on the Fusion of Neural Networks and Biological Signals","authors":"Yifei Wang","doi":"10.1016/j.procs.2025.04.243","DOIUrl":"10.1016/j.procs.2025.04.243","url":null,"abstract":"<div><div>Given the significant difficulties that children with autism face in emotion recognition and intervention, there is an urgent need to develop accurate and efficient technical means to improve their social interaction and emotional understanding abilities. This study discusses a biological signal emotion recognition and intervention technology that integrates Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM). First, this paper collects a variety of biological signal data of autistic children in different emotional states, including heart rate, galvanic skin response (GSR) and electroencephalogram (EEG), and preprocesses and extracts features of the data. Next, this paper builds and trains a deep learning model that integrates CNN and LSTM, classifies and analyzes the extracted features into emotional states, and achieves high-precision emotion recognition. Finally, this paper designs personalized intervention strategies based on the recognition results, and provides emotional guidance and intervention to children through a real-time feedback system. In the experimental conclusion, the accuracy of emotion recognition of the proposed fusion model in the training set and the verification set is 97.5% and 94.2% respectively, which is significantly better than the single mode signal processing method. In addition, the personalized intervention strategy based on this model achieved improvements of 45%, 3.8 points, and 4.2 points in reducing the amplitude of emotional fluctuations, enhancing emotional regulation ability, and improving social behavior, respectively, demonstrating the significant advantages and application potential of multimodal biosignal fusion in improving emotion recognition and intervention effects in children with autism.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 538-547"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124578","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Innovation of Multimodal Learning Paths Based on Learning Behavior and Sentiment Analysis in AI Digital Intelligence Platform AI数字智能平台中基于学习行为和情感分析的多模态学习路径创新

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.246

Lei Wang , Nan Peng , Lu Liu , Sheng Wei

This study develops an intelligent multimodal learning path recommendation system to address the problems of lack of personalized learning paths and insufficient attention to students’ emotional impact in traditional digital education models. By integrating learning behavior analysis and sentiment analysis, the Random Forest (RF) model is used to deeply mine students’ learning behavior data, such as analyzing learning time, resource access frequency, etc., to precisely understand students’ learning patterns. At the same time, with the help of sentiment analysis technology based on BERT (Bidirectional Encoder Representations from Transformers), students’ emotional states during the learning process are monitored in real-time, including positive, negative, neutral, and other emotions. The experimental results show that by dynamically adjusting the learning path and monitoring students’ emotional states, students’ average scores increase by 6.0%, and homework completion rate increases by 4.4%. This study not only improves the overall quality of education, but also provides educators with more scientific decision-making support tools.

针对传统数字教育模式中缺乏个性化学习路径和对学生情感影响关注不足的问题，本研究开发了一种智能多模态学习路径推荐系统。通过整合学习行为分析和情感分析，利用随机森林（Random Forest， RF）模型对学生的学习行为数据进行深度挖掘，如分析学习时间、资源访问频率等，精确了解学生的学习模式。同时，借助基于BERT （Bidirectional Encoder Representations from Transformers）的情绪分析技术，实时监控学生在学习过程中的情绪状态，包括积极情绪、消极情绪、中性情绪等。实验结果表明，通过动态调整学习路径和监控学生的情绪状态，学生的平均成绩提高了6.0%，作业完成率提高了4.4%。本研究不仅提高了教育的整体质量，也为教育工作者提供了更科学的决策支持工具。

{"title":"Innovation of Multimodal Learning Paths Based on Learning Behavior and Sentiment Analysis in AI Digital Intelligence Platform","authors":"Lei Wang , Nan Peng , Lu Liu , Sheng Wei","doi":"10.1016/j.procs.2025.04.246","DOIUrl":"10.1016/j.procs.2025.04.246","url":null,"abstract":"<div><div>This study develops an intelligent multimodal learning path recommendation system to address the problems of lack of personalized learning paths and insufficient attention to students’ emotional impact in traditional digital education models. By integrating learning behavior analysis and sentiment analysis, the Random Forest (RF) model is used to deeply mine students’ learning behavior data, such as analyzing learning time, resource access frequency, etc., to precisely understand students’ learning patterns. At the same time, with the help of sentiment analysis technology based on BERT (Bidirectional Encoder Representations from Transformers), students’ emotional states during the learning process are monitored in real-time, including positive, negative, neutral, and other emotions. The experimental results show that by dynamically adjusting the learning path and monitoring students’ emotional states, students’ average scores increase by 6.0%, and homework completion rate increases by 4.4%. This study not only improves the overall quality of education, but also provides educators with more scientific decision-making support tools.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 566-573"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124581","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Application of Voice Recognition Technology in Diary Applications 语音识别技术在日记应用中的应用

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.250

Mi Zhou

Traditional diary applications mainly rely on keyboard input, which makes it difficult for users to quickly record their thoughts and feelings when their emotions fluctuate violently. This paper uses voice recognition technology to innovate the recording method of diary applications and optimize the user experience. This paper uses multiple voice data sets for training to ensure the accuracy and generalization ability of the model; a voice recognition method is constructed based on a one-dimensional convolutional neural network (1D CNN), which can accurately extract features from continuous voices and achieve high-quality voice transcription. The AM and NLP technology are introduced to further process the recognized text and improve the accuracy of its grammar, logic and emotional expression. Experimental results show that the method based on 1D CNN has an accuracy rate, word missing rate and vocabulary coverage of 94.61%, 3.17% and 93.11% respectively. Regarding time efficiency, the average input time of 1D CNN is 6.46 seconds. Voice recognition technology has great potential in diary applications. It can significantly improve recording efficiency and user experience, making diary content more authentic, fluent and personalized.

传统的日记应用主要依靠键盘输入，当用户情绪剧烈波动时，很难快速记录自己的想法和感受。本文利用语音识别技术创新日记应用的记录方式，优化用户体验。本文使用多个语音数据集进行训练，保证了模型的准确性和泛化能力；构建了一种基于一维卷积神经网络（1D CNN）的语音识别方法，该方法能够准确提取连续语音的特征，实现高质量的语音转录。引入AM和NLP技术对识别文本进行进一步处理，提高其语法、逻辑和情感表达的准确性。实验结果表明，基于1D CNN的方法准确率为94.61%，缺词率为3.17%，词汇覆盖率为93.11%。在时间效率方面，1D CNN的平均输入时间为6.46秒。语音识别技术在日记应用中具有很大的潜力。显著提高记录效率和用户体验，使日记内容更加真实、流畅、个性化。

{"title":"Application of Voice Recognition Technology in Diary Applications","authors":"Mi Zhou","doi":"10.1016/j.procs.2025.04.250","DOIUrl":"10.1016/j.procs.2025.04.250","url":null,"abstract":"<div><div>Traditional diary applications mainly rely on keyboard input, which makes it difficult for users to quickly record their thoughts and feelings when their emotions fluctuate violently. This paper uses voice recognition technology to innovate the recording method of diary applications and optimize the user experience. This paper uses multiple voice data sets for training to ensure the accuracy and generalization ability of the model; a voice recognition method is constructed based on a one-dimensional convolutional neural network (1D CNN), which can accurately extract features from continuous voices and achieve high-quality voice transcription. The AM and NLP technology are introduced to further process the recognized text and improve the accuracy of its grammar, logic and emotional expression. Experimental results show that the method based on 1D CNN has an accuracy rate, word missing rate and vocabulary coverage of 94.61%, 3.17% and 93.11% respectively. Regarding time efficiency, the average input time of 1D CNN is 6.46 seconds. Voice recognition technology has great potential in diary applications. It can significantly improve recording efficiency and user experience, making diary content more authentic, fluent and personalized.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 598-604"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144125106","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Accurate Detection and Classification of Surface Defects in Electric Porcelain Insulators Based on Deep Learning Intelligent Algorithms 基于深度学习智能算法的电瓷绝缘子表面缺陷精确检测与分类

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.248

Liwei Tan, Yuhan Hu

The traditional surface defect detection method of porcelain insulators has obvious shortcomings in accuracy. This paper introduces an intelligent algorithm based on deep learning to achieve accurate detection and classification of surface defects of porcelain insulators. First, a high-resolution image acquisition system is used to comprehensively scan the surface of porcelain insulators to obtain high-quality original image data. Then, the original data is processed based on data enhancement technology to generate diversified training samples to improve the generalization ability of the model. Then, a deep learning model that integrates YOLOv5 (You Only Look Once Version 5) and ResNet50 (Residual Networks 50) is designed. The pre-training weights are optimized through transfer learning, which improves the recognition effect of the model on complex defect types. Finally, in order to further improve the detection accuracy, multi-scale detection and feature fusion technology are used to solve the problems in small-size defects and large-scale image data processing. The deep learning model proposed in this study has an accuracy of 91.50%, a recall of 89.00%, a precision of 90.20%, and an F1-score of 89.60% in the detection and classification of defects in porcelain insulators without using data enhancement. Finally, the triple data enhancement combination of rotation, cropping, and brightness adjustment further improves the performance of the model, with an accuracy of 94.30%, a recall of 92.50%, and a precision of 93.20%, respectively, and an F1-score of 92.80%. This method not only has high accuracy and robustness but also can achieve efficient and automated defect detection in actual industrial applications, providing a strong guarantee for the safe operation of the power system.

传统的瓷绝缘子表面缺陷检测方法在精度上存在明显的不足。介绍了一种基于深度学习的智能算法，实现了瓷绝缘子表面缺陷的准确检测和分类。首先，利用高分辨率图像采集系统对瓷绝缘子表面进行全面扫描，获得高质量的原始图像数据。然后，基于数据增强技术对原始数据进行处理，生成多样化的训练样本，提高模型的泛化能力。然后，设计了一个集成了YOLOv5 （You Only Look Once Version 5）和ResNet50 （Residual Networks 50）的深度学习模型。通过迁移学习优化预训练权值，提高了模型对复杂缺陷类型的识别效果。最后，为了进一步提高检测精度，采用多尺度检测和特征融合技术解决小尺寸缺陷和大规模图像数据处理的问题。本研究提出的深度学习模型在不使用数据增强的情况下，对瓷绝缘子缺陷的检测和分类准确率为91.50%，召回率为89.00%，精密度为90.20%，f1分数为89.60%。最后，通过旋转、裁剪和亮度调整的三重数据增强组合，进一步提高了模型的性能，准确率为94.30%，召回率为92.50%，精度为93.20%，f1得分为92.80%。该方法不仅精度高、鲁棒性好，而且在实际工业应用中可以实现高效、自动化的缺陷检测，为电力系统的安全运行提供有力保障。

{"title":"Accurate Detection and Classification of Surface Defects in Electric Porcelain Insulators Based on Deep Learning Intelligent Algorithms","authors":"Liwei Tan, Yuhan Hu","doi":"10.1016/j.procs.2025.04.248","DOIUrl":"10.1016/j.procs.2025.04.248","url":null,"abstract":"<div><div>The traditional surface defect detection method of porcelain insulators has obvious shortcomings in accuracy. This paper introduces an intelligent algorithm based on deep learning to achieve accurate detection and classification of surface defects of porcelain insulators. First, a high-resolution image acquisition system is used to comprehensively scan the surface of porcelain insulators to obtain high-quality original image data. Then, the original data is processed based on data enhancement technology to generate diversified training samples to improve the generalization ability of the model. Then, a deep learning model that integrates YOLOv5 (You Only Look Once Version 5) and ResNet50 (Residual Networks 50) is designed. The pre-training weights are optimized through transfer learning, which improves the recognition effect of the model on complex defect types. Finally, in order to further improve the detection accuracy, multi-scale detection and feature fusion technology are used to solve the problems in small-size defects and large-scale image data processing. The deep learning model proposed in this study has an accuracy of 91.50%, a recall of 89.00%, a precision of 90.20%, and an F1-score of 89.60% in the detection and classification of defects in porcelain insulators without using data enhancement. Finally, the triple data enhancement combination of rotation, cropping, and brightness adjustment further improves the performance of the model, with an accuracy of 94.30%, a recall of 92.50%, and a precision of 93.20%, respectively, and an F1-score of 92.80%. This method not only has high accuracy and robustness but also can achieve efficient and automated defect detection in actual industrial applications, providing a strong guarantee for the safe operation of the power system.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 582-588"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144125103","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0