最新刊期
卷 23 ， 期 11 ， 2024
- “在自然语言处理领域，专家对语义文本相似度计算方法进行了深入分析，为衡量文本间语义相似度提供新思路。”摘要：Semantic text similarity calculation is a key task in the field of natural language processing， which aims to measure the degree of semantic similarity between two texts. Based on the summary of the traditional and current mainstream semantic text similarity calculation methods， these methods are divided into traditional methods and deep learning-based methods. The traditional methods are divided into literal matching， statistics and rule-based methods. The methods based on deep learning are further divided into the methods based on word embedding， sentence vector and pre-trained model. On the basis of further subdivision of each category， the typical methods of each subcategory are introduced in detail， and the basic ideas， advantages and limitations of each method are deeply analyzed and summarized. Finally， the possible development direction of semantic text similarity calculation is prospected.关键词：text similarity;semantic similarity;natural language processing;deep learning;pretrained model390|393|0更新时间：2024-11-19
- “语音风格迁移技术最新研究进展，分析了特征提取、语料对齐、迁移模型和声码器等关键因素，为该领域发展提供新方向。”摘要：Voice style transfer technology is to convert the voice style or timbre of the source speaker into the voice style or timbre of the target speaker on the premise of keeping the voice content unchanged. In order to quickly understand the latest development of key technologies of speech style transfer， combined with domestic and foreign research in the field of speech style transfer in recent years， the research status is analyzed from four important factors， including feature extraction， corpus alignment， transfer model and coder， mainly including the comparison of multiple feature extraction methods， the selection of parallel corpus and non-parallel corpus. wavenet vocoder and several related improved vocoders， finally， the latest research progress of deep neural networks in style transfer model is introduced， the current research status in this field is summarized， the key problems and technical challenges are identified， and the future research direction and potential applications are prospected.关键词：voice style transfer;deep learning;transfer learning;voice conversion;vocoder109|141|0更新时间：2024-11-19
- “在云控制系统领域，同态加密技术为数据保密性提供解决方案，推动了云加密控制系统的未来发展。”摘要：With the rapid development of cloud computing and control theory， cloud control systems have emerged to achieve efficient control computing. However， the transmission and processing of sensitive data between communication networks and third-party clouds are vulnerable to network attacks such as eavesdropping and data tampering. Due to homomorphic encryption， this security vulnerability can be addressed by encrypting data at various levels of transmission and computation， providing data confidentiality for the entire control loop. Firstly， a comprehensive analysis was conducted on the research of using homomorphic encryption for security protection of cloud control systems. Then， the research progress of linear control， distributed control， and model predictive control based on homomorphic encryption was introduced； Then analyze the characteristics and limitations of different methods for solving the problem of limited running time of dynamic controllers； Finally， summarize the quantization techniques used for encrypting data and look forward to the future development direction of cloud encryption control systems， in order to provide reference and inspiration for researchers of cloud control systems based on homomorphic encryption.关键词：homomorphic encryption;cloud control systems;encrypted controller;security43|30|0更新时间：2024-11-19
- “在数据可视化领域，研究者提出了基于知识图谱嵌入模型的新方法，通过sTransH模型学习实体与关系嵌入，有效推理可视化设计选择。”摘要：Visualization recommendation or automatic visualization generation can help users with no background in data visualization to quickly create effective data visualizations. Aiming at the problems of heavy manual effort and poor interpretability in existing rule-based methods and machine learning-based methods， this paper proposes a knowledge graph embedding model based method， which uses cache self-adversarial negative sampling to learn embeddings of entities and relations， and constructs a knowledge graph to reason about visualization designs. Firstly， a rule-based method is used to extract features from the data set to construct a knowledge graph. Secondly， the improved TransH（sTransH） model is used to learn the embedding of entities and relations to solve the one-to-many/many-to-one situation. Finally， we reason about visualization design choices based on the embedded projections and translation vectors of entities. Experimental results show that sTransH outperforms the existing TransE， TransH， RotatE， and TransE-adv baseline methods in terms of accuracy， mean rank （MR）， and rank ratio （Hit@2）， which verifies the effectiveness of the proposed sTransH model in visualization recommendation.关键词：data visualization;visualization recommendation;knowledge graph embedding;translation model150|140|0更新时间：2024-11-19
- “在强对流天气识别领域，研究者提出了基于DOCnet的分类识别方法，有效提高了分类准确度。”摘要：Severe convective weather will pose a serious threat to people's lives and property， accurate identification of it is beneficial to better forecasting it. In order to solve the problems that traditional machine learning methods ignore the different shape expressions of different severe convective weather on radar and the traditional machine learning methods have a large amount of calculation， a classification and identification method of multi-category severe convective weather based on DOCnet is proposed. This network uses deepwise octave convolution to extract high-frequency features and low-frequency features of radar respectively. This not only removes the redundant space in low-frequency feature maps and reduces the amount of parameters， but also increases the receptive field of the convolution layer for extracting low-frequency feature maps and high-frequency feature maps. which enables the network to fully extract the image features of the radar and improve the model's classification accuracy for severe convective weather. Finally， the model is trained through the flooding method to improve the generalization ability of the model.In the test set of the storm event image （SEVIR） data， the DOCnet model has an average probability of detection of 90.54% ， an average critical success index of 81.2% and the average false alarm ration of 11.9% for the four types of severe convective weather： heavy precipitation， thunderstorms， hail， and tornadoes. Compared with the baseline model， DOCnet achieves a probability of detection improvement of 15.02 percentage points， and compared with the best performing MobileNet V2， DOCnet achieves a hit rate 5.87% higher than that of MobileNet V2. The experimental results show that DOCnet can effectively improve the classification effect of strong convective weather.关键词：classification;severe convective weather;octave convolution;depthwise convolution;radar map107|44|0更新时间：2024-11-19
- “在文档检索领域，研究者提出了基于知识图谱的零样本文档检索伪查询生成方法KGQG，有效提升了检索性能并减轻了手动标记训练数据的工作量。”摘要：To improve the performance of document retrieval models and reduce the workload of manually labeling training data， a zero sample document retrieval pseudo query generation method KGQG based on knowledge graph is proposed. This method utilizes knowledge graphs to enhance pseudo queries， combining external information with pseudo queries to generate richer and more informative pseudo queries. The experimental results showed that in the 12 public datasets of the BEIR benchmark test， the KGQG method improved the normalized discounted cumulative benefit （NDCG） index by 4.6， 11.88， and 7.96 percentage points respectively compared to the classical sparse retrieval model， dense retrieval model， and the latest zero sample dense retrieval model based on external knowledge extension. The KGQG method not only improves retrieval performance， but also reduces the need for manual labeling of training data， providing useful references for future research and application of document retrieval models.关键词：dense retrieval;information retrieval;zero-shot learning;query expansion;knowledge graph;natural language processing37|91|0更新时间：2024-11-19
- “在教育服务智能化领域，研究者提出了融合图嵌入和注意力机制的自适应学习路径推荐模型GE-MAKT，有效提升了个性化学习规划的准确性。”摘要：Adaptive learning path recommendation is a core application of intelligent technology in the intelligentization of educational services. Knowledge tracing predicts students' future knowledge levels by analyzing their historical learning records， providing personalized learning plans and recommendations. However， existing knowledge tracing methods face data sparsity issues when dealing with the complexity of students' online learning behaviors and often ignore temporal factors and students' forgetting mechanisms. This leads to models failing to accurately capture students' state changes， thus affecting recommendation effectiveness. This paper integrates graph embedding and attention mechanisms into the knowledge tracing model， designing a novel adaptive learning path recommendation model supported by deep learning （GE-MAKT model）. Experimental results show that compared to traditional methods， the GE-MAKT model significantly improves the AUC and ACC evaluation metrics， enhancing the ability to assess students' knowledge mastery levels， thereby providing more personalized learning path recommendations for students.关键词：online learning;deep learning;knowledge tracing;graph embedding;attention mechanism;path recommendation66|193|0更新时间：2024-11-19
- “在微表情识别领域，研究者提出了改进伪三维残差网络方法，有效提升了识别率和鲁棒性。”摘要：Aiming at the problems of difficulty in capturing micro expression features and low recognition rate， an improved pseudo 3D residual network micro expression recognition method is proposed. Firstly， after preprocessing the dataset samples， different bottleneck structures were designed for Pseudo-3D-SS， Pseudo-3D-PS， and Pseudo-3D-SSS in the design phase of the Pseudo 3D Residual Network （Pseudo 3DResNet） to address three issues： unreasonable interlayer order of basic 3D residual units， unstable output values， and information propagation blocking. These structures were applied to four residual blocks. Secondly， two independently designed convolutional layers and pooling layers are applied to filter weakly characterized micro expression sequences between different residual blocks， in order to further highlight valuable feature information and remove redundancy， achieving spatiotemporal feature extraction. Finally， in order to further reduce the impact of brightness changes on optical flow feature extraction and improve the speed of tracking feature points， the L1 norm based total variation optical flow method was improved. The TV-OFM method was used to extract optical flow features to obtain horizontal and vertical optical flow sequences for each micro expression. The experiment showed that the proposed method improved the unweighted F1 score （UF1） and unweighted average recall （UAR） by 3.61% and 2.92%， respectively， compared to recent methods； In the comparative experiment of optical flow method， the proposed method improved the speed of tracking feature points by 30.08% compared to the comparative method； Comparing the four improved residual block variants with other network variants， it was found that the Avg （UF1+UAR） of the proposed model increased by 2.5%， and the network structure has stronger generalization and feature extraction capabilities； Overlaying convolution and pooling operations in layers can extract features more evenly， further improve the recognition rate， and make the model more robust and progressiveness.关键词：micro-expression recognition;TV-OFM;Pseudo-3DResNet;residual block;LOOCV37|53|0更新时间：2024-11-19
- “在高校育人管理领域，专家构建了大学生群体综合特征画像标签体系，为个性化教育管理提供支持。”摘要：Precision education is the natural orientation of university education management under the current "people-oriented" education concept. User profiling technology， as a means of accurately analyzing user characteristics and depicting user behavior， can be well applied to the analysis of characteristics of college students， promoting personalized and information-based education. Aiming at the problem of low efficiency and poor authenticity of the current comprehensive feature analysis methods for college students， a label system and construction framework for the comprehensive feature portrait of college students based on user portrait technology are proposed， aiming to accurately analyze the comprehensive characteristics of college students. By crawling relevant information platforms， distributing questionnaire surveys， and contacting relevant departments of universities， publicly available data was obtained for experimental analysis. K-means algorithm clustering analysis was used to identify five groups of college students， namely comprehensive， academic overachiever， popular， interest oriented， and depressed. The experimental results show that the portrait label system and construction framework can effectively reflect the comprehensive characteristics of college students， provide support for personalized and precise education management in universities， and also help promote college students' self-awareness and management.关键词：user portrait;college students;group characteristics;targeted education147|495|0更新时间：2024-11-19
- “在气象预报领域，专家提出了基于U-net和3D注意力机制的风场订正模型，有效提高了数值天气预报的准确率。”摘要：At present， numerical weather forecasting products have been widely used in meteorological forecasting operations. A wind field correction model based on U-net and 3D attention mechanism that integrates multiple spatiotemporal features is proposed to further improve the accuracy of numerical forecasting models due to their inherent shortcomings. This model is used to correct the deviation of the near ground 10 meter wind field forecast by the GRAPES-3KM model developed by the China Meteorological Administration. RMSE and MAE are used as evaluation indicators to compare with the original numerical forecast products， traditional correction methods， and U-net and CU net models. The experimental results show that the RMSE of the 10 m meridional wind corrected by the proposed model has decreased by 4.19% to 42.67% compared to the original forecast data， LASSO regression， U-net， and CU net models， and the MAE has decreased by 6.06% to 45.29%； The RMSE index of the 10 meter meridional wind decreased by 8.55%~41.35%， and the MAE decreased by 6.54%~40.82%； The RMSE of 10 m full wind speed decreased by 6.14%~29.41%， and the MAE decreased by 1.5%~21.08%. The proposed model has better correction effect compared to the control model， and there is no situation where the correction results are too smooth.关键词：numerical weather prediction;deep learning;bias correction;multifactor fusion;3D attention134|114|0更新时间：2024-11-19
- “在暗涵排口信息提取领域，专家提出了基于点云孔洞边界检测的新方法，有效提高了边界检测准确率，降低了误判率，为暗涵排查信息化提供了参考。”摘要：In view of the large amount of point cloud data and serious noise interference of the 3D laser scanning model of the culvert， the manual extraction of outlet information is inefficient and easy to misjudge. In this paper， a method of extracting the information of the culvert outlet based on the hole boundary detection of the point cloud is proposed. Firstly， the spherical neighborhood is established according to the average distance of the point cloud， and projected into the tangent plane. Then， the boundary points are detected by the combined criterion of the maximum angle and force， and the noise points， and non-hole boundaries are removed by clustering and plane fitting. Finally， the coordinates， elevation and size information of the culvert outlet are extracted according to the convex hull of the boundary points. The results show that the spherical neighborhood is more robust than the kd-tree. When the calculation time of the combined criterion is only increased by 5 %， the accuracy of boundary detection is 3% higher than that of the single criterion， and the misjudgment rate is reduced by nearly half. This method provides a certain reference for the informatization of culvert investigation.关键词：culvert;outlet;point cloud;hole boundary detection;combined criteria12|37|0更新时间：2024-11-19
- “在医疗物联网领域，专家利用NLP技术提取论文语义特征，构建技术发展网络图，分析技术趋势。”摘要：With the rapid development of science and technology， identifying the development trajectory and research hotspots in technology domains is of great significance. However， the explosive growth of technical data has made it costly and time-consuming to monitor the technology trend through manually. Therefore， this paper proposes a technology development network maps generation method based on semantic and citation analysis to enable technology trend analysis. The proposed method involves extracting latent semantic features from scientific papers using natural language processing （NLP） techniques， and combines these semantic features to achieve sub-topic division of a technical domain through cluster analysis. Then， a technology development network map composed of main papers is constructed by combining paper topic classification and paper citation analysis， and graph algorithms are used to analyze the generated network map to obtain meaningful technology trend results. By using the papers in the field of healthcare IoT， this paper realizes the sub-topic division in this field， and generates the technology development network map in this field to realize the technical trend analysis. Especially，the application of graph algorithm to the generated technology development network map is helpful to find the hot spots and trends in the field of healthcare Internet of Things.关键词：technology development network map;trend analysis;topic modeling;natural language processing（NLP）;word embedding;healthcare IoT25|80|0更新时间：2024-11-19
- “在城市空气污染预测领域，专家提出了基于深度学习的多城市联合预测模型Res-Att-SimVP，有效缓解周边城市污染物扩散影响，为空气污染物浓度预测提供解决方案。”摘要：The traditional urban air pollution prediction model only models the historical data of Chengdu over time， ignoring the influence of the spatial dimension of pollutant diffusion in surrounding cities. To this end， a deep learning based multi city joint prediction model for air pollution， Res Att SimVP， is proposed， which includes spatial feature extraction module， temporal feature extraction module， feature screening module， and prediction module. Among them， the spatial feature extraction module uses residual networks to extract spatial feature information between cities； The time feature extraction module uses Inception network for temporal modeling； The feature filtering module uses channel attention mechanism and spatial attention mechanism to selectively focus on feature information in different regions and time periods； The prediction module uses a fully connected network to achieve air pollutant concentration prediction. Integrating all modules can effectively alleviate the impact of pollutant diffusion in surrounding cities. The hourly meteorological data， air pollutant data， and distance information between Chengdu and surrounding cities were used to test the predictive performance of this method and the control method for air pollutant concentrations at different time scales. The results showed that compared with the Res ConvLSTM model， the Res Att SimVP model reduced the average RMSE and MAE of PM2.5， PM10， and NO2 by 12% and 15%， 11% and 15%， and 12% and 22%， respectively， at different prediction durations， demonstrating high accuracy.关键词：air pollution prediction;deep learning;spatio-temporal data;multi-city joint;time series33|73|0更新时间：2024-11-19
- “车载自组织网络数据规模激增，DFPC算法提升车辆卸载成功率，降低负载方差。”摘要：The data volume carried by vehicle ad-hoc networks（VANETs） exhibits an explosive growth trend， targeting the issue of a significant decline in offloading success rate due to uneven loads on multi-access edge computing （MEC） servers， this study proposes a software-defined networking （SDN）-based vehicular network multi-MEC dynamic load balancing algorithm DFPC.This algorithm combines the two methods of first-come-first-serve and priority service in queuing theory， where the SDN controller periodically collects the current batch of tasks after a certain waiting delay. It uses an improved k-means clustering algorithm to rapidly categorize multidimensional tasks， prioritizing tasks with higher relative urgency for queuing. And using the MEC context information collected by the SDN controller at regular intervals， it implements dynamic feedback adjustment for task distribution among multiple MECs， resolving the issue of dynamic load imbalance among multiple MECs. This approach optimally utilizes MECs computational resources， ultimately enhancing the overall success rate of offloading for vehicles.To validate the effectiveness of the DFPC algorithm in real dynamic scenarios， a multi-MEC online load balancing framework， named MOLF， is designed. Performance testing for load balancing in online offloading scenarios is conducted through a low-cost hardware deployment model. Experimental results indicate that， compared to the baseline solution， the DFPC algorithm increased the average offloading success rate by 28% and reduced the average load variance by 73%.关键词：vehicular ad-hoc network（VANET）;mobile edge computing（MEC）;load balancing;online offloading;software-defined networking（SDN）34|93|0更新时间：2024-11-19
- “在离群点检测领域，K-iForest算法通过核函数重新采样提升性能，实验显示其AUC值优于其他算法0.1%~100.2%。”摘要：The isolation forest algorithm based on random subsampling does not take into account the relative density between sample points from different regions in the subsampling. Therefore， a kernel based isolation forest algorithm K-iForest is proposed to improve the performance of the isolation forest algorithm by resampling based on the probability density function. The effectiveness and efficiency of the K-iForest algorithm are validated on the Annthyroid， ForestCover， Mulcross， Shuttle in the Outlier Detection Database （ODDS）， and Http （KDD Cup 1999）， Smtp （KDD Cup 1999）， and KDD CUP 99 datasets， and compared it with the iForest algorithm， EIF algorithm， RRCF algorithm， GIF algorithm， and HIF algorithm. The experimental results show that the AUC value of the K-iForest algorithm is 0.1% to 100.2% higher than other algorithms.关键词：kernel function;outlier detection;isolated forest algorithm;probability density;relative density12|48|0更新时间：2024-11-19
- “在图像拼接领域，专家利用K-means算法优化KNN匹配，提升匹配率近1倍，缩短匹配时间约0.2秒，并改进最佳缝合线算法，有效去除重叠区域“鬼影”错位现象。”摘要：Image matching and fusion are the core steps of image mosaic. In this paper， in the process of image matching， there are many wrong matches in the traditional KNN algorithm. It is observed that the slope of the correct matching point on the image is almost parallel， and the K-means algorithm is used to cluster and screen out the correct matching. Through experiments， the algorithm can remove a large number of false matches， and the matching rate and matching time are improved. The optimal suture algorithm is improved in image fusion， and compared with the gradual fusion algorithm， it is proved that the improved suture algorithm can effectively remove the ghost and dislocation phenomenon in the overlapping area of the image.关键词：KNN matching;K-means clustering;RANSAC algorithm;optimal suture line;fade in and fade out algorithm19|97|0更新时间：2024-11-19
- “在防灾减灾领域，专家构建了多灾害多源异构数据库与管理系统，实现了极端灾害数据的高效采集和标准化，为减小灾害风险提供技术支持。”摘要：The frequent occurrence of extreme weather around the world in recent years has caused lots of extreme disasters such as floods， mudslides and collapses. Research on the design and implementation of multi-disaster and multi-source heterogeneous database management system is of great significance for disaster prevention and reduction， emergency rescue and reconstruction after disaster. This paper first introduces the framework of multi-disaster and multi-source heterogeneous database and management system. Then， the management system design based on MySQL relational database is introduced in detail. The data collection program is developed by using Python language to realize the efficient collection and standardization of data information of earthquake， tsunami， flood， wave and fire. Then， two key technologies， serialization and deserialization and transaction processing， are applied to realize efficient storage and retrieval of database. Finally， the function test of database management system is carried out， and the results show that the database and management system established in this paper run well. The construction of the database and the implementation of the system can provide important technique support for reducing extreme disaster risk and increasing disaster resilience everywhere.关键词：extreme disaster;disaster database;MySQL database;data acquisition;multi-source heterogeneous31|166|0更新时间：2024-11-19
- “在个人信息安全领域，专家设计了结合口令与击键特征的双因素认证方案，提高了身份认证的安全性和便捷性。”摘要：With the rapid development of Internet technology， personal information security has been paid more and more attention. The traditional identity authentication method based on username and password carries the risk of password leakage. Biometric recognition technology， mainly based on facial recognition and fingerprint recognition， is greatly affected in its universality due to expensive equipment and complex algorithms. The use of biometric behavioral features to identify user identity has become a research hotspot. To this end， a dual factor authentication scheme combining password and keystroke features is designed. By collecting input passwords and personal keystroke behavior characteristics during password input， combined with the new dimension information priority idea， statistical algorithms and Manhattan algorithm are used to analyze and model the collected user keystroke behavior feature data， and match it with the current user's keystroke behavior feature for similarity to authenticate the user's identity. The system designed based on this scheme not only has the characteristics of user-friendly interface， easy operation， and high security， but also can automatically display the keystroke characteristic curve of each user， which can serve as the first security gate of important information systems.关键词：information security;identify authentication;keystroke features;statistics algorithm;Manhattan algorithm46|107|0更新时间：2024-11-19
- “在城市规划和智慧旅游领域，精准识别建筑物至关重要。针对建筑物标注数据稀缺和视觉外观多样性问题，专家提出了基于余弦注意力机制的少样本建筑识别方法，有效提升了模型性能。”摘要：As the core landmark in urban space and a crucial medium for human activities， the accurate identification of buildings is of great significance in urban planning，smart tourism and other fields. However， collecting and labeling a sufficient amount of data is a costly and time-consuming endeavor. Aiming at the problem that some buildings' labeling data are scarce and the visual appearance diversity leads to insufficient feature representation， this paper proposes a method of building identification with few samples based on cosine attention mechanism. This method fully captures the features of the target object by using an adaptive prototype representation methods，and replace the scaled dot-product attention mechanism in Transformer with a cosine attention mechanism to optimize model performance. Firstly， the paper collects sample data from open resources， then constructs a small sample classification data set containing a variety of historical buildings in Qingdao， and then uses this data set to verify the effectiveness of the proposed method.Experimental results show that the proposed method achieves 58.08% and 77.15% accuracy in 1-shot and 5-shot learning scenarios， respectively， which shows the ability and effect of this method in building identification with few samples.关键词：few-shot learning;building recognition;Transformer;cosine attention;Xception network27|57|0更新时间：2024-11-19
- “在云计算中心智能运维领域，专家基于ARIMA模型和Isolation Forest算法，提出了大数据分析方法，有效预测容量趋势和监测异常，提高了系统稳定性和健壮性。”摘要：Aiming at the problem of unpredictable capacity trends and inability to monitor anomalies in cloud computing centers， a method for intelligent operation and maintenance of cloud computing centers based on big data analysis is proposed using ARIMA time series algorithm and Isolation Forest anomaly detection algorithm. The experiment shows that the accuracy of the server anomaly detection method based on Isolation Forest is about 88%， and the average absolute error of disk capacity prediction based on ARIMA model is 0.158 6， proving that this system has better stability， robustness， and service capabilities compared to traditional operation and maintenance platforms.关键词：cloud computing center;intelligent operation and maintenance;isolation forest;ARIMA model;trend prediction;anomaly detection21|112|0更新时间：2024-11-19
- “在医学影像领域，研究者提出了MU-Net算法，有效提升了脑胶质瘤分割的准确度，为脑肿瘤治疗提供新方案。”摘要：Magnetic resonance imaging segmentation is crucial for the treatment of brain tumor patients， but issues such as variable tumor morphology and blurred boundaries result in poor edge segmentation performance. To solve the above problems， a brain glioma automated segmentation algorithm MU Net based on multiple U-Net networks is proposed. Firstly， using U-Net as the backbone network， a residual dilated convolution module is designed as a short connection in the encoding stage to enhance the connection of long-distance information of encoded features， thereby improving the feature extraction effect； Secondly， an improved efficient channel attention mechanism is introduced at the network skip connections， while using average pooling and maximum pooling to fully utilize spatial and channel information to improve segmentation accuracy； Finally， multiple dual output U-Nets of different depths are designed at the skip connections processed by the improved efficient channel attention mechanism as links between encoding and decoding to enhance the network's adaptability to brain tumors of different scales. A large number of experiments were conducted on the BraTS2020 dataset， and the results showed that the MU Net algorithm had Dice coefficients of 86.75%， 77.76%， and 76.21% for intact tumors， tumor cores， and enhanced tumors， respectively. Compared with the benchmark model， it improved by 2.6%， 2.55%， and 2.41%， respectively， and had better segmentation performance.关键词：brain tumor;magnetic resonance imaging segmentation;multi U-Net network;residual module;efficient channel attention mechanism;dual output U-Net75|93|0更新时间：2024-11-19
- “在水果采摘领域，研究者提出了一种新型目标检测方法，有效提高了复杂环境下小尺寸水果的检测识别准确率，为自动采摘提供理论方法和实践经验。”摘要：In the context of small fruits， partial occlusion， changing light， cluttered background， etc.， it is difficult for the picking robot to improve the accuracy of detection and recognition， leading to issues such as missed and erroneous picking during the picking process. To solve this problem， a novel object detection method is proposed by integrating the Hough circle detection algorithm and the improved template matching algorithm. It aims to effectively detect and recognize small-sized and multi-class fruits in complicated situations. Firstly， denoising pre-processing， such as adaptive threshold binarization， is performed on the target fruit data. Secondly， this is followed by the application of the Hough circle detection algorithm to detect the position of the fruit. And then the template matching algorithm is enhanced by calculating the matching similarity from four levels of pixel， brightness， contrast and structure to identify the fruit variety. Finally， It is validated experimentally using the fruits of strawberries， jujubes and blueberries. And the results indicate that the proposed algorithm takes about 3.2ms and exhibits the mean average accuracy of 95% when detecting fruits， which is superior to the three algorithms of SSD， YOLOv8 and YOLOv9. Additionally， the recognition accuracy of the algorithm in complex environments is 92%. The proposed algorithm offers an effective theoretical method and practical experience for automatic picking of small-sized fruits.关键词：picking robots;fruit detection;Hough circle detection;template matching;matching similarity27|76|0更新时间：2024-11-19
-
Detection Algorithm of Electric Power Operation Wearable Device Based on GSConv Lightweight YOLOv5s AI导读
“在电力作业安全领域，专家提出了融合GSConv轻量化YOLOv5s的检测算法，有效监测作业人员是否正确佩戴安全设备。”摘要：In order to determine whether power workers are wearing safety equipment such as helmets， insulated boots， insulated gloves， and insulated clothing， and ensure the safety of power construction sites， a lightweight YOLOv5s fusion GSConv algorithm for power operation safety equipment detection is proposed. Firstly， a new feature extraction end is constructed using C3Ghost convolution and deep separation convolution （DWConv）， and GSConv is used instead of ordinary convolution in the feature fusion end to reduce model complexity， reduce parameter and computational complexity， and improve detection accuracy while improving algorithm calculation speed； Secondly， use K-mean+clustering algorithm to obtain the candidate box setting values at the output of YOLOv5s algorithm； Finally， the PSA self attention mechanism is used to improve the feature extraction end of the YOLOv5s algorithm， enhancing the channel resolution and spatial resolution of safety equipment images in power operation scenarios， while preserving key node information of occluded small targets. The experimental results show that the average accuracy mean （IoU=0.5） of the algorithm reaches 0.962， which is 1.50% higher than the original network detection performance. At the same time， the model parameters are reduced from 7.02 M to 2.77 M， and the computational complexity is reduced from 15.8 GFLOPs to 5.7 GFLOPs. The proposed algorithm can meet the real-time detection requirements of power operations and effectively monitor whether the operators are wearing safety equipment correctly in the presence of obstructions and omissions in the power operation scene.关键词：lightweight YOLOv5s algorithm;electricity operations;safety wearable devices;PSA amechanism;GSConv12|78|0更新时间：2024-11-19 - “在实时手势检测领域，研究者提出了基于YOLOv5的手势识别算法，通过优化主干网络，实现了轻量化并有效解决了实时检测问题。”摘要：To meet the demand for real-time hand gesture detection， this paper presents a YOLOv5-based gesture recognition algorithm. By replacing CSPNet-53 in YOLOv5s with the lightweight MobileNetV3， the optimized backbone integrates depthwise separable convolutions and the SE attention mechanism， forming the M_YOLO_N model. Compared to the original， M_YOLO_N reduces parameters by 33% and decreases computational complexity by 54%. On a custom dataset， mAP@0.5 increased by 2.4%. This model achieves both lightweight design and real-time detection.For multi-scale detection， the SPPF module is retained， and the normalized Wasserstein distance （NWD） is introduced，proposes a new bounding box loss function NewIoU. Without increasing parameters， detection confidence improved by 20%.关键词：YOLOv5;gesture recognition;depth separable convolution;attention mechanism45|86|0更新时间：2024-11-19
- “街景图像语义分割研究取得新进展，提出了结合生成对抗网络与混合注意力机制的分割模型，有效提高了分割精度和模型稳定性。”摘要：Semantic segmentation of street view images is one of the main research tasks in the field of autonomous driving. At present， semantic segmentation of street view images mainly suffers from inaccurate segmentation of small target objects and overfitting of models. Therefore， a street view image semantic segmentation model combining generative adversarial networks and hybrid attention mechanisms is proposed. Specifically， a multi-scale hybrid attention module is proposed to enhance contextual semantic information， improve feature representation ability， and adaptability to multi-scale targets. At the same time， in order to reduce overfitting， a BN layer is introduced and combined with a DCGAN network to construct a generative adversarial network segmentation model. The training is constrained by both discriminative loss and segmentation loss to enhance model stability and improve segmentation accuracy. The experimental results showed that compared with DeepLabV3+， the proposed model improved segmentation accuracy by 2.4 percentage points on the Cityscapes dataset， with a mIoU value of 73.4%.关键词：street view semantic segmentation;generative adversarial network;hybrid attention mechanism;hybrid loss function41|45|0更新时间：2024-11-19
-
Lightweight and Improvement of Aerial Photography Small Target Detection Algorithm Based on YOLOv8s AI导读
“在目标检测领域，针对航拍小目标图像检测难题，提出了基于YOLOv8s改进的轻量化小目标检测算法。实验结果表明，所提算法在Visdrone数据集上的mAP50和mAP50：95分别达到41.4%和24.9%，相较原始YOLOv8s算法精度提高了1.6%和1.3%，参数量和运算量降低了36.8%和42.8%。”摘要：Small target detection has always been a hot topic in the field of object detection， and aerial images of small targets have the characteristics of large scale changes， diverse perspectives and poses， which can easily lead to missed or false detections. To address these issues， a lightweight small object detection algorithm based on YOLOv8s improvement is proposed. Firstly， the CESE-C2f module is used in the Backbone section to enhance detection performance and reduce parameter count by utilizing a lightweight attention mechanism； Secondly， design a new network structure based on the BiFPN concept， fully utilizing existing network information to enhance the fusion of deep and shallow feature information； Finally， the new lightweight detection head LW-Detect is used to significantly reduce the computational and parameter requirements to meet real-time requirements. The experimental results showed that the mAP50 and mAP50：95 of the proposed algorithm on the Visdrone dataset reached 41.4% and 24.9%， respectively. Compared with the original YOLOv8s algorithm， the accuracy was improved by 1.6% and 1.3%， and the parameter and computational complexity were reduced by 36.8% and 42.8%， respectively.关键词：YOLOv8s;small target;lightweight;BiFPN;target detection36|72|0更新时间：2024-11-19 - “在工业生产领域，针对钢表面缺陷检测问题，专家改进YOLOv8模型，提出YOLOv8-LSD算法，有效提高检测精度和速度。”摘要：Due to imperfect manufacturing processes and external factors， steel surfaces often have defects that can seriously affect their life and usability.Therefore， surface defect detection is a necessary process in industrial production. Traditional surface defect detection algorithms have the drawbacks of low accuracy and slow speed.Therefore， this article improves on the YOLOv8 model. Replace the original loss function CIoU with the SIoU function， and introduce the ShuffleAttention （SA） attention mechanism in the Backbone section to improve the extraction ability of shallow and deep feature information in the image. Finally， add a small object detection layer in the network based on the characteristics of the dataset to enhance the feature extraction ability.Experiments show that the improved YOLOv8-LSD algorithm， which improves the mAP value by 3.9% over the original algorithm， reduces the false detection and leakage rate of defects.关键词：steel surface defect;YOLOv8;SIOU;ShuffleAttention（SA）;detection layer32|139|0更新时间：2024-11-19
- “在人工智能教学领域，专家探讨了基于OBE理念与案例驱动教学法的改革策略，旨在提升学生兴趣和思政素养，增强实践应用能力。”摘要：Traditional artificial intelligence teaching methods face many challenges， including numerous algorithms， obscure and difficult to understand content， lack of student interest， and insufficient ideological and political education. Therefore， taking the Introduction to Artificial Intelligence course as the research object， this paper explores teaching reform strategies based on the OBE concept and case-based teaching method. The reform measures include reshaping course knowledge， ability， and literacy goals， optimizing and reconstructing teaching content， implementing a blended online and offline teaching model， and proposing a multidimensional assessment mechanism， aiming to build a closed-loop and modular teaching unit to meet the needs of the industry and students. Practice has shown that the proposed measures can enhance students' interest and confidence in artificial intelligence courses， cultivate their ideological and political literacy， strengthen their practical application and problem-solving abilities， and better meet the needs of cultivating applied talents in the field of computer artificial intelligence.关键词：OBE;case-driven;teaching reform;artificial intelligence;ideological and political education32|48|0更新时间：2024-11-19
- “在网络安全人才培养领域，高校与企业合作建立开放式实验室，探索安全人才培养新模式，有效提升毕业生就业率和薪资水平，具有推广价值。”摘要：The cultivation and reserve of cybersecurity talents is the fundamental guarantee of national cyberspace sovereignty and an urgent issue that universities need to address. In response to the imbalance between the output of network security talent cultivation and the demand for job positions in the recruitment market， a new model of open laboratory construction and security talent cultivation in ordinary undergraduate colleges is proposed. This model collaborates with network security enterprises to establish an open laboratory based on campus network security system and guided by security attack and defense practical combat. This article elaborates on the exploration and practical process of the training model from six aspects： the formulation of training plans， student selection， construction of practical environments， cultivation of dual qualified teachers， practical evaluation， and school enterprise cooperation. By comparing the employment reports of school graduates over the years， it was found that the employment rate of laboratory graduates has increased by 15%， and the average salary is 1.9 times that of graduates in similar majors. This model actively promotes the cultivation of network security talents and the construction of practical teaching staff in ordinary undergraduate colleges， and has great promotion value.关键词：network security talents;cultivation mode;open laboratory;double-qualified teacher;practical teaching27|53|0更新时间：2024-11-19
- Technical support is provided by Beijing Founder electronics co., LTD 鄂ICP备18021494号-4
鄂公网安备42010602004949号 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0